diff --git a/experiments/prolific_mova_1932/accepted_annotations/69eb6e157e23c6c0b2b7fb0d_baa12307967541c0b3f848b18ff0db47.json b/experiments/prolific_mova_1932/accepted_annotations/69eb6e157e23c6c0b2b7fb0d_baa12307967541c0b3f848b18ff0db47.json new file mode 100644 index 0000000000000000000000000000000000000000..e862993f4b5c87649a3628409ea148db080696f5 --- /dev/null +++ b/experiments/prolific_mova_1932/accepted_annotations/69eb6e157e23c6c0b2b7fb0d_baa12307967541c0b3f848b18ff0db47.json @@ -0,0 +1,3503 @@ +{ + "schema_version": 1, + "annotator_id": "69eb6e157e23c6c0b2b7fb0d", + "assignment_id": "baa12307967541c0b3f848b18ff0db47", + "group_id": "five_axis_26_82", + "quality_control": { + "correct_fields": 20, + "total_fields": 25, + "accuracy": 0.8, + "threshold": 0.7 + }, + "annotation_count": 21, + "annotations": [ + { + "id": "ef741e2433e041a2b54bb22045fc6964", + "video_id": "tree-then-thumbs_down_left-then-snap_left__order__MOVA", + "user": "69eb6e157e23c6c0b2b7fb0d", + "annotations": { + "responses": { + "events": { + "speech": { + "a": { + "label": "PASS", + "rationale": "" + }, + "v": { + "label": "PASS", + "rationale": "" + }, + "avb": { + "label": "PASS", + "rationale": "" + }, + "tcCount": { + "value": "1", + "rationale": "" + } + }, + "silent_action": { + "v": { + "label": "FAIL", + "rationale": "" + }, + "tcCount": { + "value": "0", + "rationale": "" + } + }, + "sounding_action": { + "a": { + "label": "FAIL", + "rationale": "" + }, + "v": { + "label": "FAIL", + "rationale": "" + }, + "avb": { + "label": "FAIL", + "rationale": "" + }, + "tcCount": { + "value": "0", + "rationale": "" + } + } + }, + "video": { + "tcRel": { + "label": "FAIL", + "rationale": "The requested bundle is not completed, it only passed audio and visual correctness when it says tree, the other is not passed. Also he spoke the word that is not requested." + } + } + }, + "axis_definitions": [ + { + "id": "a", + "key": "A", + "title": "Audio Event Correctness", + "granularity": "per-event", + "summary": "Whether the required speech or sound event is present and correct in the audio.", + "detail": "PASS when every required speech/sound component is present, clear, and correct. FAIL when a required audio event is missing, the content or sound type is wrong, or the audio is too unclear to identify. Judge audibility leniently (a soft or quiet sound still counts); judge identity leniently within an acoustic family (clap/slap/tap/thud/pat) and strictly across families." + }, + { + "id": "v", + "key": "V", + "title": "Visual Event Correctness", + "granularity": "per-event", + "summary": "Whether the required visual event is present and correct. For speech events this evaluates visible speaking behavior (mouth/lip motion), not the spoken content.", + "detail": "PASS when every required visual component is present, visible, and correct in actor, action, side, body part, object, and contact target. FAIL when it's missing; the wrong actor/side/body part/object/action is used; occlusion, cropping, or blur prevents verification; or (for speech) the mouth movement is absent or inconsistent with the required utterance." + }, + { + "id": "avb", + "key": "AVB", + "title": "Audio-Visual Binding", + "granularity": "per-event", + "summary": "Whether the audio and visual components within the same event are correctly synchronized and causally aligned.", + "detail": "Applies only to events with both audio and visual components (speech, sounding action — not silent action). PASS when the sound is synchronized with the corresponding visible action, impact, or mouth movement. FAILS if A or V fails, or the two are clearly misaligned in time. Temporal relations between different events are handled under TC_REL, not AVB." + }, + { + "id": "tcRel", + "key": "TC_REL", + "title": "Temporal Relation", + "granularity": "per-video", + "summary": "Whether the specified relationship between events holds - synchronization, order, and absolute timing.", + "detail": "Compare the defined start times of the required events. If any required event fails A, V, or AVB, TC_REL FAILS. Synchronization: PASS when events begin at the same moment (simple overlap is not enough). Order: PASS when events clearly begin in the required order (overlap allowed). Absolute timing: PASS when the event starts close to the required timestamp; +/-0.3s is a recommended guideline, not a strict cutoff - use judgment rather than a hard pass/fail line." + }, + { + "id": "tcCount", + "key": "TC_COUNT", + "title": "Temporal Count", + "granularity": "per-event", + "summary": "Whether each required event occurs the specified number of times, with no missing or extra occurrences.", + "detail": "Enter how many times the event actually occurs in the video. If no count is given, expect one. If the event fails A, V, or its applicable AVB, record its count as 0. A continuously held pose counts as one occurrence; releasing and re-forming it (or a clear distinct repetition without full release) counts as another." + } + ], + "events": [ + { + "id": "speech", + "modality": "speech", + "modalityLabel": "Speech", + "actor": "person_A", + "label": "Speech: say \"tree\"", + "definition": "The person clearly says the single word \"tree\". The speech is audible and easy to distinguish. No extra words are spoken unless explicitly required.", + "axes": { + "a": true, + "v": true, + "avb": true + } + }, + { + "id": "silent_action", + "modality": "silent_action", + "modalityLabel": "Silent action", + "actor": "person_A", + "label": "Silent action: thumbs down (left hand)", + "definition": "The person's left hand holds a thumbs-down pose, with the fingers curled into a fist and the thumb extended downward. The target is the pose, not the raising motion.", + "axes": { + "a": false, + "v": true, + "avb": false + } + }, + { + "id": "sounding_action", + "modality": "sounding_action", + "modalityLabel": "Sounding action", + "actor": "person_A", + "label": "Sounding action: snapping fingers (with left hand)", + "definition": "The person snaps the fingers of the left hand: the thumb and middle finger visibly snap, and a clear snap sound occurs exactly at the moment of the snap.", + "axes": { + "a": true, + "v": true, + "avb": true + } + } + ], + "temporal_constraint": { + "relationType": "order", + "timingType": "relative", + "value": { + "sequence": [ + "speech", + "silent_action", + "sounding_action" + ] + }, + "expectedCounts": { + "speech": 1, + "silent_action": 1, + "sounding_action": 1 + }, + "description": "Order: Speech: say \"tree\" -> then -> Silent action: thumbs down (left hand) -> then -> Sounding action: snapping fingers (with left hand)." + }, + "model": "MOVA", + "item_id": "tree-then-thumbs_down_left-then-snap_left__order", + "canonical_id": "snap_left_thumbs_down_left_tree_rel_order", + "variant": "v1", + "scene_hint": "plain indoor room", + "camera_hint": "static medium shot", + "prompt": "In a plain indoor room, a seated person faces the camera in a static medium shot, with the full face, mouth, upper body, both hands, and thighs clearly visible. The person starts seated on a chair with both hands initially resting on the knees or thighs. First, the person clearly says the single word \"tree\". After that speech ends, the person holds a thumbs-down pose with the left hand, fingers curled into a fist and the thumb extended downward. The target is the pose, not the raising motion. After that pose is visible, that hand returns to rest, then the person snaps the fingers of the left hand once so the thumb and middle finger visibly snap, with a clear snap sound exactly at the moment of the snap. No extra words or extra target sounds occur. After the final event, no new target action occurs." + }, + "created_at": "2026-11-15T00:27:31.340514Z", + "annotation_role": "full" + }, + { + "id": "60bb433449014818b42191d8538d2007", + "video_id": "down-with-ok_sign-with-slap_left__sync_count_x2__MOVA", + "user": "69eb6e157e23c6c0b2b7fb0d", + "annotations": { + "responses": { + "events": { + "speech": { + "a": { + "label": "FAIL", + "rationale": "" + }, + "v": { + "label": "PASS", + "rationale": "" + }, + "avb": { + "label": "PASS", + "rationale": "" + }, + "tcCount": { + "value": "1", + "rationale": "" + } + }, + "silent_action": { + "v": { + "label": "PASS", + "rationale": "" + }, + "tcCount": { + "value": "2", + "rationale": "" + } + }, + "sounding_action": { + "a": { + "label": "FAIL", + "rationale": "" + }, + "v": { + "label": "FAIL", + "rationale": "" + }, + "avb": { + "label": "FAIL", + "rationale": "" + }, + "tcCount": { + "value": "0", + "rationale": "" + } + } + }, + "video": { + "tcRel": { + "label": "FAIL", + "rationale": "the bundle not done twice, the person not spoke only one word (word), the ok sign was done, the slapping is done by right hand not left hand. All the action was not performed synchronously." + } + } + }, + "axis_definitions": [ + { + "id": "a", + "key": "A", + "title": "Audio Event Correctness", + "granularity": "per-event", + "summary": "Whether the required speech or sound event is present and correct in the audio.", + "detail": "PASS when every required speech/sound component is present, clear, and correct. FAIL when a required audio event is missing, the content or sound type is wrong, or the audio is too unclear to identify. Judge audibility leniently (a soft or quiet sound still counts); judge identity leniently within an acoustic family (clap/slap/tap/thud/pat) and strictly across families." + }, + { + "id": "v", + "key": "V", + "title": "Visual Event Correctness", + "granularity": "per-event", + "summary": "Whether the required visual event is present and correct. For speech events this evaluates visible speaking behavior (mouth/lip motion), not the spoken content.", + "detail": "PASS when every required visual component is present, visible, and correct in actor, action, side, body part, object, and contact target. FAIL when it's missing; the wrong actor/side/body part/object/action is used; occlusion, cropping, or blur prevents verification; or (for speech) the mouth movement is absent or inconsistent with the required utterance." + }, + { + "id": "avb", + "key": "AVB", + "title": "Audio-Visual Binding", + "granularity": "per-event", + "summary": "Whether the audio and visual components within the same event are correctly synchronized and causally aligned.", + "detail": "Applies only to events with both audio and visual components (speech, sounding action — not silent action). PASS when the sound is synchronized with the corresponding visible action, impact, or mouth movement. FAILS if A or V fails, or the two are clearly misaligned in time. Temporal relations between different events are handled under TC_REL, not AVB." + }, + { + "id": "tcRel", + "key": "TC_REL", + "title": "Temporal Relation", + "granularity": "per-video", + "summary": "Whether the specified relationship between events holds - synchronization, order, and absolute timing.", + "detail": "Compare the defined start times of the required events. If any required event fails A, V, or AVB, TC_REL FAILS. Synchronization: PASS when events begin at the same moment (simple overlap is not enough). Order: PASS when events clearly begin in the required order (overlap allowed). Absolute timing: PASS when the event starts close to the required timestamp; +/-0.3s is a recommended guideline, not a strict cutoff - use judgment rather than a hard pass/fail line." + }, + { + "id": "tcCount", + "key": "TC_COUNT", + "title": "Temporal Count", + "granularity": "per-event", + "summary": "Whether each required event occurs the specified number of times, with no missing or extra occurrences.", + "detail": "Enter how many times the event actually occurs in the video. If no count is given, expect one. If the event fails A, V, or its applicable AVB, record its count as 0. A continuously held pose counts as one occurrence; releasing and re-forming it (or a clear distinct repetition without full release) counts as another." + } + ], + "events": [ + { + "id": "speech", + "modality": "speech", + "modalityLabel": "Speech", + "actor": "person_A", + "label": "Speech: say \"down\"", + "definition": "The person clearly says the single word \"down\". The speech is audible and easy to distinguish. No extra words are spoken unless explicitly required.", + "axes": { + "a": true, + "v": true, + "avb": true + } + }, + { + "id": "silent_action", + "modality": "silent_action", + "modalityLabel": "Silent action", + "actor": "person_A", + "label": "Silent action: ok sign (right hand)", + "definition": "The person's right thumb and index fingertip touch to form a ring while the other three fingers remain extended and the palm faces the camera. The target is the held OK pose, not the raising motion.", + "axes": { + "a": false, + "v": true, + "avb": false + } + }, + { + "id": "sounding_action", + "modality": "sounding_action", + "modalityLabel": "Sounding action", + "actor": "person_A", + "label": "Sounding action: slapping the thigh (with left hand on left thigh)", + "definition": "The person's left hand strikes the top of the person's left thigh. Visible contact is made with the thigh, and a sharp slap sound occurs exactly at the moment of impact.", + "axes": { + "a": true, + "v": true, + "avb": true + } + } + ], + "temporal_constraint": { + "relationType": "sync_count", + "timingType": "relative", + "value": { + "events": [ + "speech", + "silent_action", + "sounding_action" + ], + "count": 2 + }, + "expectedCounts": { + "speech": 2, + "silent_action": 2, + "sounding_action": 2 + }, + "description": "Synchronization x2: Speech: say \"down\" + Silent action: ok sign (right hand) + Sounding action: slapping the thigh (with left hand on left thigh) must occur together, repeated exactly 2 times." + }, + "model": "MOVA", + "item_id": "down-with-ok_sign-with-slap_left__sync_count_x2", + "canonical_id": "down_ok_sign_slap_left_rel_sync_count", + "variant": "v1", + "scene_hint": "plain indoor room", + "camera_hint": "static medium shot", + "prompt": "In a plain indoor room, a seated person faces the camera in a static medium shot, with the full face, mouth, upper body, both hands, and thighs clearly visible. The person starts seated on a chair with both hands initially resting on the knees or thighs. The person performs the synchronized bundle exactly 2 times: each time, the person clearly says the single word \"down\", holds an OK sign with the right hand, thumb and index fingertip forming a ring and the other three fingers extended, palm facing the camera, and strikes the top of the left thigh with the left hand once, making visible contact and a sharp slap sound exactly at the moment of impact. The synchronized repetitions are distinct, with a brief reset between them if needed. The speech is audible and easy to distinguish, with no extra words. The target is the pose, not the raising motion. No extra words or extra target sounds occur. After the final event, no new target action occurs." + }, + "created_at": "2026-11-15T00:33:47.912141Z", + "annotation_role": "full" + }, + { + "id": "d382625f09684d3ca35ddc0a2d5f464b", + "video_id": "yes-with-slap_left-with-fist_pose__sync_count_x3__MOVA", + "user": "69eb6e157e23c6c0b2b7fb0d", + "annotations": { + "responses": { + "events": { + "speech": { + "a": { + "label": "FAIL", + "rationale": "" + }, + "v": { + "label": "PASS", + "rationale": "" + }, + "avb": { + "label": "PASS", + "rationale": "" + }, + "tcCount": { + "value": "1", + "rationale": "" + } + }, + "sounding_action": { + "a": { + "label": "FAIL", + "rationale": "" + }, + "v": { + "label": "FAIL", + "rationale": "" + }, + "avb": { + "label": "FAIL", + "rationale": "" + }, + "tcCount": { + "value": "0", + "rationale": "" + } + }, + "silent_action": { + "v": { + "label": "PASS", + "rationale": "" + }, + "tcCount": { + "value": "5", + "rationale": "" + } + } + }, + "video": { + "tcRel": { + "label": "FAIL", + "rationale": "the person spoke additional words, he didn't slap the thigh with left hand instead with right hand while performed fist pose" + } + } + }, + "axis_definitions": [ + { + "id": "a", + "key": "A", + "title": "Audio Event Correctness", + "granularity": "per-event", + "summary": "Whether the required speech or sound event is present and correct in the audio.", + "detail": "PASS when every required speech/sound component is present, clear, and correct. FAIL when a required audio event is missing, the content or sound type is wrong, or the audio is too unclear to identify. Judge audibility leniently (a soft or quiet sound still counts); judge identity leniently within an acoustic family (clap/slap/tap/thud/pat) and strictly across families." + }, + { + "id": "v", + "key": "V", + "title": "Visual Event Correctness", + "granularity": "per-event", + "summary": "Whether the required visual event is present and correct. For speech events this evaluates visible speaking behavior (mouth/lip motion), not the spoken content.", + "detail": "PASS when every required visual component is present, visible, and correct in actor, action, side, body part, object, and contact target. FAIL when it's missing; the wrong actor/side/body part/object/action is used; occlusion, cropping, or blur prevents verification; or (for speech) the mouth movement is absent or inconsistent with the required utterance." + }, + { + "id": "avb", + "key": "AVB", + "title": "Audio-Visual Binding", + "granularity": "per-event", + "summary": "Whether the audio and visual components within the same event are correctly synchronized and causally aligned.", + "detail": "Applies only to events with both audio and visual components (speech, sounding action — not silent action). PASS when the sound is synchronized with the corresponding visible action, impact, or mouth movement. FAILS if A or V fails, or the two are clearly misaligned in time. Temporal relations between different events are handled under TC_REL, not AVB." + }, + { + "id": "tcRel", + "key": "TC_REL", + "title": "Temporal Relation", + "granularity": "per-video", + "summary": "Whether the specified relationship between events holds - synchronization, order, and absolute timing.", + "detail": "Compare the defined start times of the required events. If any required event fails A, V, or AVB, TC_REL FAILS. Synchronization: PASS when events begin at the same moment (simple overlap is not enough). Order: PASS when events clearly begin in the required order (overlap allowed). Absolute timing: PASS when the event starts close to the required timestamp; +/-0.3s is a recommended guideline, not a strict cutoff - use judgment rather than a hard pass/fail line." + }, + { + "id": "tcCount", + "key": "TC_COUNT", + "title": "Temporal Count", + "granularity": "per-event", + "summary": "Whether each required event occurs the specified number of times, with no missing or extra occurrences.", + "detail": "Enter how many times the event actually occurs in the video. If no count is given, expect one. If the event fails A, V, or its applicable AVB, record its count as 0. A continuously held pose counts as one occurrence; releasing and re-forming it (or a clear distinct repetition without full release) counts as another." + } + ], + "events": [ + { + "id": "speech", + "modality": "speech", + "modalityLabel": "Speech", + "actor": "person_A", + "label": "Speech: say \"yes\"", + "definition": "The person clearly says the single word \"yes\". The speech is audible and easy to distinguish. No extra words are spoken unless explicitly required.", + "axes": { + "a": true, + "v": true, + "avb": true + } + }, + { + "id": "sounding_action", + "modality": "sounding_action", + "modalityLabel": "Sounding action", + "actor": "person_A", + "label": "Sounding action: slapping the thigh (with left hand on left thigh)", + "definition": "The person's left hand strikes the top of the person's left thigh. Visible contact is made with the thigh, and a sharp slap sound occurs exactly at the moment of impact.", + "axes": { + "a": true, + "v": true, + "avb": true + } + }, + { + "id": "silent_action", + "modality": "silent_action", + "modalityLabel": "Silent action", + "actor": "person_A", + "label": "Silent action: fist pose (right hand)", + "definition": "The person's right hand holds a closed fist, with all fingers curled and the thumb resting outside the curled fingers rather than extended upward. The target is the pose, not the raising motion.", + "axes": { + "a": false, + "v": true, + "avb": false + } + } + ], + "temporal_constraint": { + "relationType": "sync_count", + "timingType": "relative", + "value": { + "events": [ + "speech", + "sounding_action", + "silent_action" + ], + "count": 3 + }, + "expectedCounts": { + "speech": 3, + "sounding_action": 3, + "silent_action": 3 + }, + "description": "Synchronization x3: Speech: say \"yes\" + Sounding action: slapping the thigh (with left hand on left thigh) + Silent action: fist pose (right hand) must occur together, repeated exactly 3 times." + }, + "model": "MOVA", + "item_id": "yes-with-slap_left-with-fist_pose__sync_count_x3", + "canonical_id": "fist_pose_slap_left_yes_rel_sync_count", + "variant": "v1", + "scene_hint": "plain indoor room", + "camera_hint": "static medium shot", + "prompt": "In a plain indoor room, a seated person faces the camera in a static medium shot, with the full face, mouth, upper body, both hands, and thighs clearly visible. The person starts seated on a chair with both hands initially resting on the knees or thighs. The person performs the synchronized bundle exactly 3 times: each time, the person clearly says the single word \"yes\", strikes the top of the left thigh with the left hand once, making visible contact and a sharp slap sound exactly at the moment of impact, and holds a closed fist with the right hand, all fingers curled and the thumb resting outside the curled fingers rather than extended upward. The synchronized repetitions are distinct, with a brief reset between them if needed. The speech is audible and easy to distinguish, with no extra words. The target is the pose, not the raising motion. No extra words or extra target sounds occur. After the final event, no new target action occurs." + }, + "created_at": "2026-11-15T00:38:02.032434Z", + "annotation_role": "full" + }, + { + "id": "c3841a2949df4c49af5d092f1c0fb03d", + "video_id": "hello-then-point_pose-then-snap__order__MOVA", + "user": "69eb6e157e23c6c0b2b7fb0d", + "annotations": { + "responses": { + "events": { + "speech": { + "a": { + "label": "FAIL", + "rationale": "the person spoke another additional word" + }, + "v": { + "label": "PASS", + "rationale": "" + }, + "avb": { + "label": "PASS", + "rationale": "" + }, + "tcCount": { + "value": "1", + "rationale": "" + } + }, + "silent_action": { + "v": { + "label": "PASS", + "rationale": "" + }, + "tcCount": { + "value": "1", + "rationale": "" + } + }, + "sounding_action": { + "a": { + "label": "FAIL", + "rationale": "No snapping fingers action done" + }, + "v": { + "label": "FAIL", + "rationale": "No snapping fingers action done" + }, + "avb": { + "label": "FAIL", + "rationale": "No snapping fingers action done" + }, + "tcCount": { + "value": "0", + "rationale": "No snapping fingers action done" + } + } + }, + "video": { + "tcRel": { + "label": "FAIL", + "rationale": "Mostly fail because there is extra word spoken, and no snapping fingers done. Only point pose that passed" + } + } + }, + "axis_definitions": [ + { + "id": "a", + "key": "A", + "title": "Audio Event Correctness", + "granularity": "per-event", + "summary": "Whether the required speech or sound event is present and correct in the audio.", + "detail": "PASS when every required speech/sound component is present, clear, and correct. FAIL when a required audio event is missing, the content or sound type is wrong, or the audio is too unclear to identify. Judge audibility leniently (a soft or quiet sound still counts); judge identity leniently within an acoustic family (clap/slap/tap/thud/pat) and strictly across families." + }, + { + "id": "v", + "key": "V", + "title": "Visual Event Correctness", + "granularity": "per-event", + "summary": "Whether the required visual event is present and correct. For speech events this evaluates visible speaking behavior (mouth/lip motion), not the spoken content.", + "detail": "PASS when every required visual component is present, visible, and correct in actor, action, side, body part, object, and contact target. FAIL when it's missing; the wrong actor/side/body part/object/action is used; occlusion, cropping, or blur prevents verification; or (for speech) the mouth movement is absent or inconsistent with the required utterance." + }, + { + "id": "avb", + "key": "AVB", + "title": "Audio-Visual Binding", + "granularity": "per-event", + "summary": "Whether the audio and visual components within the same event are correctly synchronized and causally aligned.", + "detail": "Applies only to events with both audio and visual components (speech, sounding action — not silent action). PASS when the sound is synchronized with the corresponding visible action, impact, or mouth movement. FAILS if A or V fails, or the two are clearly misaligned in time. Temporal relations between different events are handled under TC_REL, not AVB." + }, + { + "id": "tcRel", + "key": "TC_REL", + "title": "Temporal Relation", + "granularity": "per-video", + "summary": "Whether the specified relationship between events holds - synchronization, order, and absolute timing.", + "detail": "Compare the defined start times of the required events. If any required event fails A, V, or AVB, TC_REL FAILS. Synchronization: PASS when events begin at the same moment (simple overlap is not enough). Order: PASS when events clearly begin in the required order (overlap allowed). Absolute timing: PASS when the event starts close to the required timestamp; +/-0.3s is a recommended guideline, not a strict cutoff - use judgment rather than a hard pass/fail line." + }, + { + "id": "tcCount", + "key": "TC_COUNT", + "title": "Temporal Count", + "granularity": "per-event", + "summary": "Whether each required event occurs the specified number of times, with no missing or extra occurrences.", + "detail": "Enter how many times the event actually occurs in the video. If no count is given, expect one. If the event fails A, V, or its applicable AVB, record its count as 0. A continuously held pose counts as one occurrence; releasing and re-forming it (or a clear distinct repetition without full release) counts as another." + } + ], + "events": [ + { + "id": "speech", + "modality": "speech", + "modalityLabel": "Speech", + "actor": "person_A", + "label": "Speech: say \"hello\"", + "definition": "The person clearly says the single word \"hello\". The speech is audible and easy to distinguish. No extra words are spoken unless explicitly required.", + "axes": { + "a": true, + "v": true, + "avb": true + } + }, + { + "id": "silent_action", + "modality": "silent_action", + "modalityLabel": "Silent action", + "actor": "person_A", + "label": "Silent action: point pose (right hand)", + "definition": "The person's right hand holds a pointing pose toward a location on the person's right, with the index finger extended and the other fingers curled. The hand stays in frame. The target is the pose, not the raising motion.", + "axes": { + "a": false, + "v": true, + "avb": false + } + }, + { + "id": "sounding_action", + "modality": "sounding_action", + "modalityLabel": "Sounding action", + "actor": "person_A", + "label": "Sounding action: snapping fingers (with right hand)", + "definition": "The person snaps the fingers of the right hand: the thumb and middle finger visibly snap, and a clear snap sound occurs exactly at the moment of the snap.", + "axes": { + "a": true, + "v": true, + "avb": true + } + } + ], + "temporal_constraint": { + "relationType": "order", + "timingType": "relative", + "value": { + "sequence": [ + "speech", + "silent_action", + "sounding_action" + ] + }, + "expectedCounts": { + "speech": 1, + "silent_action": 1, + "sounding_action": 1 + }, + "description": "Order: Speech: say \"hello\" -> then -> Silent action: point pose (right hand) -> then -> Sounding action: snapping fingers (with right hand)." + }, + "model": "MOVA", + "item_id": "hello-then-point_pose-then-snap__order", + "canonical_id": "hello_point_pose_snap_rel_order", + "variant": "v1", + "scene_hint": "plain indoor room", + "camera_hint": "static medium shot", + "prompt": "In a plain indoor room, a seated person faces the camera in a static medium shot, with the full face, mouth, upper body, both hands, and thighs clearly visible. The person starts seated on a chair with both hands initially resting on the knees or thighs. First, the person clearly says the single word \"hello\". After that speech ends, the person holds a pointing pose with the right hand toward a location on the person's right, index finger extended and the other fingers curled. The target is the pose, not the raising motion. After that pose is visible, that hand returns to rest, then the person snaps the fingers of the right hand once so the thumb and middle finger visibly snap, with a clear snap sound exactly at the moment of the snap. No extra words or extra target sounds occur. After the final event, no new target action occurs." + }, + "created_at": "2026-11-15T00:42:57.490226Z", + "annotation_role": "full" + }, + { + "id": "62390653bf9746699a0fcbeac67e0aac", + "video_id": "help-then-salute_pose_left-then-point_pose__order__MOVA", + "user": "69eb6e157e23c6c0b2b7fb0d", + "annotations": { + "responses": { + "events": { + "speech": { + "a": { + "label": "FAIL", + "rationale": "He spoke additional word" + }, + "v": { + "label": "PASS", + "rationale": "" + }, + "avb": { + "label": "PASS", + "rationale": "" + }, + "tcCount": { + "value": "1", + "rationale": "" + } + }, + "silent_action": { + "v": { + "label": "FAIL", + "rationale": "He uses left hand" + }, + "tcCount": { + "value": "0", + "rationale": "He uses left hand" + } + }, + "silent_action_2": { + "v": { + "label": "FAIL", + "rationale": "He did not point to his right" + }, + "tcCount": { + "value": "1", + "rationale": "" + } + } + }, + "video": { + "tcRel": { + "label": "FAIL", + "rationale": "The person spoke another additional word, the salute poses not using left hand, and point pose didn't point to the requested direction" + } + } + }, + "axis_definitions": [ + { + "id": "a", + "key": "A", + "title": "Audio Event Correctness", + "granularity": "per-event", + "summary": "Whether the required speech or sound event is present and correct in the audio.", + "detail": "PASS when every required speech/sound component is present, clear, and correct. FAIL when a required audio event is missing, the content or sound type is wrong, or the audio is too unclear to identify. Judge audibility leniently (a soft or quiet sound still counts); judge identity leniently within an acoustic family (clap/slap/tap/thud/pat) and strictly across families." + }, + { + "id": "v", + "key": "V", + "title": "Visual Event Correctness", + "granularity": "per-event", + "summary": "Whether the required visual event is present and correct. For speech events this evaluates visible speaking behavior (mouth/lip motion), not the spoken content.", + "detail": "PASS when every required visual component is present, visible, and correct in actor, action, side, body part, object, and contact target. FAIL when it's missing; the wrong actor/side/body part/object/action is used; occlusion, cropping, or blur prevents verification; or (for speech) the mouth movement is absent or inconsistent with the required utterance." + }, + { + "id": "avb", + "key": "AVB", + "title": "Audio-Visual Binding", + "granularity": "per-event", + "summary": "Whether the audio and visual components within the same event are correctly synchronized and causally aligned.", + "detail": "Applies only to events with both audio and visual components (speech, sounding action — not silent action). PASS when the sound is synchronized with the corresponding visible action, impact, or mouth movement. FAILS if A or V fails, or the two are clearly misaligned in time. Temporal relations between different events are handled under TC_REL, not AVB." + }, + { + "id": "tcRel", + "key": "TC_REL", + "title": "Temporal Relation", + "granularity": "per-video", + "summary": "Whether the specified relationship between events holds - synchronization, order, and absolute timing.", + "detail": "Compare the defined start times of the required events. If any required event fails A, V, or AVB, TC_REL FAILS. Synchronization: PASS when events begin at the same moment (simple overlap is not enough). Order: PASS when events clearly begin in the required order (overlap allowed). Absolute timing: PASS when the event starts close to the required timestamp; +/-0.3s is a recommended guideline, not a strict cutoff - use judgment rather than a hard pass/fail line." + }, + { + "id": "tcCount", + "key": "TC_COUNT", + "title": "Temporal Count", + "granularity": "per-event", + "summary": "Whether each required event occurs the specified number of times, with no missing or extra occurrences.", + "detail": "Enter how many times the event actually occurs in the video. If no count is given, expect one. If the event fails A, V, or its applicable AVB, record its count as 0. A continuously held pose counts as one occurrence; releasing and re-forming it (or a clear distinct repetition without full release) counts as another." + } + ], + "events": [ + { + "id": "speech", + "modality": "speech", + "modalityLabel": "Speech", + "actor": "person_A", + "label": "Speech: say \"help\"", + "definition": "The person clearly says the single word \"help\". The speech is audible and easy to distinguish. No extra words are spoken unless explicitly required.", + "axes": { + "a": true, + "v": true, + "avb": true + } + }, + { + "id": "silent_action", + "modality": "silent_action", + "modalityLabel": "Silent action", + "actor": "person_A", + "label": "Silent action: salute pose (left hand)", + "definition": "The person's left hand holds a salute pose beside the left temple, with the fingers straight and together and the eyes and mouth visible. The target is the pose, not the raising motion.", + "axes": { + "a": false, + "v": true, + "avb": false + } + }, + { + "id": "silent_action_2", + "modality": "silent_action", + "modalityLabel": "Silent action", + "actor": "person_A", + "label": "Silent action: point pose (right hand)", + "definition": "The person's right hand holds a pointing pose toward a location on the person's right, with the index finger extended and the other fingers curled. The hand stays in frame. The target is the pose, not the raising motion.", + "axes": { + "a": false, + "v": true, + "avb": false + } + } + ], + "temporal_constraint": { + "relationType": "order", + "timingType": "relative", + "value": { + "sequence": [ + "speech", + "silent_action", + "silent_action_2" + ] + }, + "expectedCounts": { + "speech": 1, + "silent_action": 1, + "silent_action_2": 1 + }, + "description": "Order: Speech: say \"help\" -> then -> Silent action: salute pose (left hand) -> then -> Silent action: point pose (right hand)." + }, + "model": "MOVA", + "item_id": "help-then-salute_pose_left-then-point_pose__order", + "canonical_id": "help_point_pose_salute_pose_left_rel_order", + "variant": "v1", + "scene_hint": "plain indoor room", + "camera_hint": "static medium shot", + "prompt": "In a plain indoor room, a seated person faces the camera in a static medium shot, with the full face, mouth, upper body, both hands, and thighs clearly visible. The person starts seated on a chair with both hands initially resting on the knees or thighs. First, the person clearly says the single word \"help\". After that speech ends, the person holds a salute pose with the left hand beside the left temple, fingers straight and together, with the eyes and mouth visible. The target is the pose, not the raising motion. After that pose is visible, that hand returns to rest, then the person holds a pointing pose with the right hand toward a location on the person's right, index finger extended and the other fingers curled. The target is the pose, not the raising motion. No extra words or extra target sounds occur. After the final event, no new target action occurs." + }, + "created_at": "2026-11-15T00:46:11.626028Z", + "annotation_role": "full" + }, + { + "id": "d8d9292b23c441ce8c9e909b3a6da421", + "video_id": "up-then-salute_pose-then-clap__order__MOVA", + "user": "69eb6e157e23c6c0b2b7fb0d", + "annotations": { + "responses": { + "events": { + "speech": { + "a": { + "label": "PASS", + "rationale": "" + }, + "v": { + "label": "PASS", + "rationale": "" + }, + "avb": { + "label": "PASS", + "rationale": "" + }, + "tcCount": { + "value": "1", + "rationale": "" + } + }, + "silent_action": { + "v": { + "label": "PASS", + "rationale": "" + }, + "tcCount": { + "value": "1", + "rationale": "" + } + }, + "sounding_action": { + "a": { + "label": "PASS", + "rationale": "" + }, + "v": { + "label": "PASS", + "rationale": "" + }, + "avb": { + "label": "FAIL", + "rationale": "" + }, + "tcCount": { + "value": "2", + "rationale": "" + } + } + }, + "video": { + "tcRel": { + "label": "FAIL", + "rationale": "Mostly pass except the clapping hands which didn't bind the audio-visual and pass the required count" + } + } + }, + "axis_definitions": [ + { + "id": "a", + "key": "A", + "title": "Audio Event Correctness", + "granularity": "per-event", + "summary": "Whether the required speech or sound event is present and correct in the audio.", + "detail": "PASS when every required speech/sound component is present, clear, and correct. FAIL when a required audio event is missing, the content or sound type is wrong, or the audio is too unclear to identify. Judge audibility leniently (a soft or quiet sound still counts); judge identity leniently within an acoustic family (clap/slap/tap/thud/pat) and strictly across families." + }, + { + "id": "v", + "key": "V", + "title": "Visual Event Correctness", + "granularity": "per-event", + "summary": "Whether the required visual event is present and correct. For speech events this evaluates visible speaking behavior (mouth/lip motion), not the spoken content.", + "detail": "PASS when every required visual component is present, visible, and correct in actor, action, side, body part, object, and contact target. FAIL when it's missing; the wrong actor/side/body part/object/action is used; occlusion, cropping, or blur prevents verification; or (for speech) the mouth movement is absent or inconsistent with the required utterance." + }, + { + "id": "avb", + "key": "AVB", + "title": "Audio-Visual Binding", + "granularity": "per-event", + "summary": "Whether the audio and visual components within the same event are correctly synchronized and causally aligned.", + "detail": "Applies only to events with both audio and visual components (speech, sounding action — not silent action). PASS when the sound is synchronized with the corresponding visible action, impact, or mouth movement. FAILS if A or V fails, or the two are clearly misaligned in time. Temporal relations between different events are handled under TC_REL, not AVB." + }, + { + "id": "tcRel", + "key": "TC_REL", + "title": "Temporal Relation", + "granularity": "per-video", + "summary": "Whether the specified relationship between events holds - synchronization, order, and absolute timing.", + "detail": "Compare the defined start times of the required events. If any required event fails A, V, or AVB, TC_REL FAILS. Synchronization: PASS when events begin at the same moment (simple overlap is not enough). Order: PASS when events clearly begin in the required order (overlap allowed). Absolute timing: PASS when the event starts close to the required timestamp; +/-0.3s is a recommended guideline, not a strict cutoff - use judgment rather than a hard pass/fail line." + }, + { + "id": "tcCount", + "key": "TC_COUNT", + "title": "Temporal Count", + "granularity": "per-event", + "summary": "Whether each required event occurs the specified number of times, with no missing or extra occurrences.", + "detail": "Enter how many times the event actually occurs in the video. If no count is given, expect one. If the event fails A, V, or its applicable AVB, record its count as 0. A continuously held pose counts as one occurrence; releasing and re-forming it (or a clear distinct repetition without full release) counts as another." + } + ], + "events": [ + { + "id": "speech", + "modality": "speech", + "modalityLabel": "Speech", + "actor": "person_A", + "label": "Speech: say \"up\"", + "definition": "The person clearly says the single word \"up\". The speech is audible and easy to distinguish. No extra words are spoken unless explicitly required.", + "axes": { + "a": true, + "v": true, + "avb": true + } + }, + { + "id": "silent_action", + "modality": "silent_action", + "modalityLabel": "Silent action", + "actor": "person_A", + "label": "Silent action: salute pose (right hand)", + "definition": "The person's right hand holds a salute pose beside the right temple, with the fingers straight and together and the eyes and mouth visible. The target is the pose, not the raising motion.", + "axes": { + "a": false, + "v": true, + "avb": false + } + }, + { + "id": "sounding_action", + "modality": "sounding_action", + "modalityLabel": "Sounding action", + "actor": "person_A", + "label": "Sounding action: clapping hands", + "definition": "The person brings both hands together in front of the torso to clap, with visible hand-to-hand contact and a single sharp clap sound at the moment of impact.", + "axes": { + "a": true, + "v": true, + "avb": true + } + } + ], + "temporal_constraint": { + "relationType": "order", + "timingType": "relative", + "value": { + "sequence": [ + "speech", + "silent_action", + "sounding_action" + ] + }, + "expectedCounts": { + "speech": 1, + "silent_action": 1, + "sounding_action": 1 + }, + "description": "Order: Speech: say \"up\" -> then -> Silent action: salute pose (right hand) -> then -> Sounding action: clapping hands." + }, + "model": "MOVA", + "item_id": "up-then-salute_pose-then-clap__order", + "canonical_id": "clap_salute_pose_up_rel_order", + "variant": "v1", + "scene_hint": "plain indoor room", + "camera_hint": "static medium shot", + "prompt": "In a plain indoor room, a seated person faces the camera in a static medium shot, with the full face, mouth, upper body, both hands, and thighs clearly visible. The person starts seated on a chair with both hands initially resting on the knees or thighs. First, the person clearly says the single word \"up\". After that speech ends, the person holds a salute pose with the right hand beside the right temple, fingers straight and together, with the eyes and mouth visible. The target is the pose, not the raising motion. After that pose is visible, that hand returns to rest, then the person claps both hands together in front of the torso once, with visible contact and a sharp clap sound exactly at the moment of impact. No extra words or extra target sounds occur. After the final event, no new target action occurs." + }, + "created_at": "2026-11-15T00:51:11.112815Z", + "annotation_role": "full" + }, + { + "id": "efcfe0f950424e39b3248f02467d4628", + "video_id": "hello_x2-then-point_pose_left_x2-then-clap_x2__order_count__MOVA", + "user": "69eb6e157e23c6c0b2b7fb0d", + "annotations": { + "responses": { + "events": { + "speech": { + "a": { + "label": "FAIL", + "rationale": "There's extra words" + }, + "v": { + "label": "PASS", + "rationale": "" + }, + "avb": { + "label": "PASS", + "rationale": "" + }, + "tcCount": { + "value": "1", + "rationale": "" + } + }, + "silent_action": { + "v": { + "label": "FAIL", + "rationale": "He uses right hand" + }, + "tcCount": { + "value": "0", + "rationale": "" + } + }, + "sounding_action": { + "a": { + "label": "PASS", + "rationale": "" + }, + "v": { + "label": "FAIL", + "rationale": "" + }, + "avb": { + "label": "FAIL", + "rationale": "" + }, + "tcCount": { + "value": "4", + "rationale": "" + } + } + }, + "video": { + "tcRel": { + "label": "FAIL", + "rationale": "Most of the requested action are failed to be followed" + } + } + }, + "axis_definitions": [ + { + "id": "a", + "key": "A", + "title": "Audio Event Correctness", + "granularity": "per-event", + "summary": "Whether the required speech or sound event is present and correct in the audio.", + "detail": "PASS when every required speech/sound component is present, clear, and correct. FAIL when a required audio event is missing, the content or sound type is wrong, or the audio is too unclear to identify. Judge audibility leniently (a soft or quiet sound still counts); judge identity leniently within an acoustic family (clap/slap/tap/thud/pat) and strictly across families." + }, + { + "id": "v", + "key": "V", + "title": "Visual Event Correctness", + "granularity": "per-event", + "summary": "Whether the required visual event is present and correct. For speech events this evaluates visible speaking behavior (mouth/lip motion), not the spoken content.", + "detail": "PASS when every required visual component is present, visible, and correct in actor, action, side, body part, object, and contact target. FAIL when it's missing; the wrong actor/side/body part/object/action is used; occlusion, cropping, or blur prevents verification; or (for speech) the mouth movement is absent or inconsistent with the required utterance." + }, + { + "id": "avb", + "key": "AVB", + "title": "Audio-Visual Binding", + "granularity": "per-event", + "summary": "Whether the audio and visual components within the same event are correctly synchronized and causally aligned.", + "detail": "Applies only to events with both audio and visual components (speech, sounding action — not silent action). PASS when the sound is synchronized with the corresponding visible action, impact, or mouth movement. FAILS if A or V fails, or the two are clearly misaligned in time. Temporal relations between different events are handled under TC_REL, not AVB." + }, + { + "id": "tcRel", + "key": "TC_REL", + "title": "Temporal Relation", + "granularity": "per-video", + "summary": "Whether the specified relationship between events holds - synchronization, order, and absolute timing.", + "detail": "Compare the defined start times of the required events. If any required event fails A, V, or AVB, TC_REL FAILS. Synchronization: PASS when events begin at the same moment (simple overlap is not enough). Order: PASS when events clearly begin in the required order (overlap allowed). Absolute timing: PASS when the event starts close to the required timestamp; +/-0.3s is a recommended guideline, not a strict cutoff - use judgment rather than a hard pass/fail line." + }, + { + "id": "tcCount", + "key": "TC_COUNT", + "title": "Temporal Count", + "granularity": "per-event", + "summary": "Whether each required event occurs the specified number of times, with no missing or extra occurrences.", + "detail": "Enter how many times the event actually occurs in the video. If no count is given, expect one. If the event fails A, V, or its applicable AVB, record its count as 0. A continuously held pose counts as one occurrence; releasing and re-forming it (or a clear distinct repetition without full release) counts as another." + } + ], + "events": [ + { + "id": "speech", + "modality": "speech", + "modalityLabel": "Speech", + "actor": "person_A", + "label": "Speech: say \"hello\"", + "definition": "The person clearly says the single word \"hello\". The speech is audible and easy to distinguish. No extra words are spoken unless explicitly required.", + "axes": { + "a": true, + "v": true, + "avb": true + } + }, + { + "id": "silent_action", + "modality": "silent_action", + "modalityLabel": "Silent action", + "actor": "person_A", + "label": "Silent action: point pose (left hand)", + "definition": "The person's left hand holds a pointing pose toward a location on the person's left, with the index finger extended and the other fingers curled. The hand stays in frame. The target is the pose, not the raising motion.", + "axes": { + "a": false, + "v": true, + "avb": false + } + }, + { + "id": "sounding_action", + "modality": "sounding_action", + "modalityLabel": "Sounding action", + "actor": "person_A", + "label": "Sounding action: clapping hands", + "definition": "The person brings both hands together in front of the torso to clap, with visible hand-to-hand contact and a single sharp clap sound at the moment of impact.", + "axes": { + "a": true, + "v": true, + "avb": true + } + } + ], + "temporal_constraint": { + "relationType": "order_count", + "timingType": "relative", + "value": { + "ordered_events": [ + "speech", + "silent_action", + "sounding_action" + ], + "event_counts": { + "speech": 2, + "silent_action": 2, + "sounding_action": 2 + }, + "expanded_sequence": [ + "speech", + "speech", + "silent_action", + "silent_action", + "sounding_action", + "sounding_action" + ] + }, + "expectedCounts": { + "speech": 2, + "silent_action": 2, + "sounding_action": 2 + }, + "description": "Order: Speech: say \"hello\" x2 -> then -> Silent action: point pose (left hand) x2 -> then -> Sounding action: clapping hands x2." + }, + "model": "MOVA", + "item_id": "hello_x2-then-point_pose_left_x2-then-clap_x2__order_count", + "canonical_id": "clap_hello_point_pose_left_rel_order_count", + "variant": "v1", + "scene_hint": "plain indoor room", + "camera_hint": "static medium shot", + "prompt": "In a plain indoor room, a seated person faces the camera in a static medium shot, with the full face, mouth, upper body, both hands, and thighs clearly visible. The person starts seated on a chair with both hands initially resting on the knees or thighs. First, the person must say the single word \"hello\" exactly 2 times, with a brief pause between repetitions so each repetition is distinct. After that speech ends, the person must hold a pointing pose with the left hand toward a location on the person's left, index finger extended and the other fingers curled, exactly 2 times, returning the hand to rest between repetitions so each repetition is distinct. The target is the pose, not the raising motion. After that pose is visible, that hand returns to rest, then the person must clap both hands together in front of the torso, with visible contact and a sharp clap sound exactly at the moment of impact, exactly 2 times, with a brief pause between repetitions so each repetition is distinct. The order must be strict: all repetitions of an event are completed before the next event begins. No extra words or extra target sounds occur. After the final event, no new target action occurs." + }, + "created_at": "2026-11-15T00:54:13.785007Z", + "annotation_role": "full" + }, + { + "id": "ad5b7f6594ba47e597bc16d86c80f774", + "video_id": "up_x2-then-ok_sign_x1-then-thumbs_down_x2__order_count__MOVA", + "user": "69eb6e157e23c6c0b2b7fb0d", + "annotations": { + "responses": { + "events": { + "speech": { + "a": { + "label": "FAIL", + "rationale": "Another extra words are spoken" + }, + "v": { + "label": "PASS", + "rationale": "" + }, + "avb": { + "label": "PASS", + "rationale": "" + }, + "tcCount": { + "value": "1", + "rationale": "The person only say the right word once" + } + }, + "silent_action": { + "v": { + "label": "PASS", + "rationale": "" + }, + "tcCount": { + "value": "1", + "rationale": "" + } + }, + "silent_action_2": { + "v": { + "label": "FAIL", + "rationale": "" + }, + "tcCount": { + "value": "1", + "rationale": "" + } + } + }, + "video": { + "tcRel": { + "label": "FAIL", + "rationale": "The requested action are mostly not done" + } + } + }, + "axis_definitions": [ + { + "id": "a", + "key": "A", + "title": "Audio Event Correctness", + "granularity": "per-event", + "summary": "Whether the required speech or sound event is present and correct in the audio.", + "detail": "PASS when every required speech/sound component is present, clear, and correct. FAIL when a required audio event is missing, the content or sound type is wrong, or the audio is too unclear to identify. Judge audibility leniently (a soft or quiet sound still counts); judge identity leniently within an acoustic family (clap/slap/tap/thud/pat) and strictly across families." + }, + { + "id": "v", + "key": "V", + "title": "Visual Event Correctness", + "granularity": "per-event", + "summary": "Whether the required visual event is present and correct. For speech events this evaluates visible speaking behavior (mouth/lip motion), not the spoken content.", + "detail": "PASS when every required visual component is present, visible, and correct in actor, action, side, body part, object, and contact target. FAIL when it's missing; the wrong actor/side/body part/object/action is used; occlusion, cropping, or blur prevents verification; or (for speech) the mouth movement is absent or inconsistent with the required utterance." + }, + { + "id": "avb", + "key": "AVB", + "title": "Audio-Visual Binding", + "granularity": "per-event", + "summary": "Whether the audio and visual components within the same event are correctly synchronized and causally aligned.", + "detail": "Applies only to events with both audio and visual components (speech, sounding action — not silent action). PASS when the sound is synchronized with the corresponding visible action, impact, or mouth movement. FAILS if A or V fails, or the two are clearly misaligned in time. Temporal relations between different events are handled under TC_REL, not AVB." + }, + { + "id": "tcRel", + "key": "TC_REL", + "title": "Temporal Relation", + "granularity": "per-video", + "summary": "Whether the specified relationship between events holds - synchronization, order, and absolute timing.", + "detail": "Compare the defined start times of the required events. If any required event fails A, V, or AVB, TC_REL FAILS. Synchronization: PASS when events begin at the same moment (simple overlap is not enough). Order: PASS when events clearly begin in the required order (overlap allowed). Absolute timing: PASS when the event starts close to the required timestamp; +/-0.3s is a recommended guideline, not a strict cutoff - use judgment rather than a hard pass/fail line." + }, + { + "id": "tcCount", + "key": "TC_COUNT", + "title": "Temporal Count", + "granularity": "per-event", + "summary": "Whether each required event occurs the specified number of times, with no missing or extra occurrences.", + "detail": "Enter how many times the event actually occurs in the video. If no count is given, expect one. If the event fails A, V, or its applicable AVB, record its count as 0. A continuously held pose counts as one occurrence; releasing and re-forming it (or a clear distinct repetition without full release) counts as another." + } + ], + "events": [ + { + "id": "speech", + "modality": "speech", + "modalityLabel": "Speech", + "actor": "person_A", + "label": "Speech: say \"up\"", + "definition": "The person clearly says the single word \"up\". The speech is audible and easy to distinguish. No extra words are spoken unless explicitly required.", + "axes": { + "a": true, + "v": true, + "avb": true + } + }, + { + "id": "silent_action", + "modality": "silent_action", + "modalityLabel": "Silent action", + "actor": "person_A", + "label": "Silent action: ok sign (right hand)", + "definition": "The person's right thumb and index fingertip touch to form a ring while the other three fingers remain extended and the palm faces the camera. The target is the held OK pose, not the raising motion.", + "axes": { + "a": false, + "v": true, + "avb": false + } + }, + { + "id": "silent_action_2", + "modality": "silent_action", + "modalityLabel": "Silent action", + "actor": "person_A", + "label": "Silent action: thumbs down (right hand)", + "definition": "The person's right hand holds a thumbs-down pose, with the fingers curled into a fist and the thumb extended downward. The target is the pose, not the raising motion.", + "axes": { + "a": false, + "v": true, + "avb": false + } + } + ], + "temporal_constraint": { + "relationType": "order_count", + "timingType": "relative", + "value": { + "ordered_events": [ + "speech", + "silent_action", + "silent_action_2" + ], + "event_counts": { + "speech": 2, + "silent_action": 1, + "silent_action_2": 2 + }, + "expanded_sequence": [ + "speech", + "speech", + "silent_action", + "silent_action_2", + "silent_action_2" + ] + }, + "expectedCounts": { + "speech": 2, + "silent_action": 1, + "silent_action_2": 2 + }, + "description": "Order: Speech: say \"up\" x2 -> then -> Silent action: ok sign (right hand) x1 -> then -> Silent action: thumbs down (right hand) x2." + }, + "model": "MOVA", + "item_id": "up_x2-then-ok_sign_x1-then-thumbs_down_x2__order_count", + "canonical_id": "ok_sign_thumbs_down_up_rel_order_count", + "variant": "v1", + "scene_hint": "plain indoor room", + "camera_hint": "static medium shot", + "prompt": "In a plain indoor room, a seated person faces the camera in a static medium shot, with the full face, mouth, upper body, both hands, and thighs clearly visible. The person starts seated on a chair with both hands initially resting on the knees or thighs. First, the person must say the single word \"up\" exactly 2 times, with a brief pause between repetitions so each repetition is distinct. After that speech ends, the person must hold an OK sign with the right hand, thumb and index fingertip forming a ring and the other three fingers extended, palm facing the camera once. The target is the pose, not the raising motion. After that pose is visible, that hand returns to rest, then the person must hold a thumbs-down pose with the right hand, fingers curled into a fist and the thumb extended downward, exactly 2 times, returning the hand to rest between repetitions so each repetition is distinct. The target is the pose, not the raising motion. The order must be strict: all repetitions of an event are completed before the next event begins. No extra words or extra target sounds occur. After the final event, no new target action occurs." + }, + "created_at": "2026-11-15T00:57:00.749322Z", + "annotation_role": "full" + }, + { + "id": "ce26da0c8cd5422cad7cd6f7302d0b75", + "video_id": "yes_x2-then-salute_pose_left_x2-then-snap_x1__order_count__MOVA", + "user": "69eb6e157e23c6c0b2b7fb0d", + "annotations": { + "responses": { + "events": { + "speech": { + "a": { + "label": "FAIL", + "rationale": "Additional words are spoken" + }, + "v": { + "label": "PASS", + "rationale": "" + }, + "avb": { + "label": "PASS", + "rationale": "" + }, + "tcCount": { + "value": "1", + "rationale": "Only done once, not twice as requested" + } + }, + "silent_action": { + "v": { + "label": "FAIL", + "rationale": "The person uses right hand instead" + }, + "tcCount": { + "value": "0", + "rationale": "" + } + }, + "sounding_action": { + "a": { + "label": "FAIL", + "rationale": "No snapping fingers are done" + }, + "v": { + "label": "FAIL", + "rationale": "No snapping fingers are done" + }, + "avb": { + "label": "FAIL", + "rationale": "No snapping fingers are done" + }, + "tcCount": { + "value": "0", + "rationale": "" + } + } + }, + "video": { + "tcRel": { + "label": "FAIL", + "rationale": "Most of the required actions are failed to be followed" + } + } + }, + "axis_definitions": [ + { + "id": "a", + "key": "A", + "title": "Audio Event Correctness", + "granularity": "per-event", + "summary": "Whether the required speech or sound event is present and correct in the audio.", + "detail": "PASS when every required speech/sound component is present, clear, and correct. FAIL when a required audio event is missing, the content or sound type is wrong, or the audio is too unclear to identify. Judge audibility leniently (a soft or quiet sound still counts); judge identity leniently within an acoustic family (clap/slap/tap/thud/pat) and strictly across families." + }, + { + "id": "v", + "key": "V", + "title": "Visual Event Correctness", + "granularity": "per-event", + "summary": "Whether the required visual event is present and correct. For speech events this evaluates visible speaking behavior (mouth/lip motion), not the spoken content.", + "detail": "PASS when every required visual component is present, visible, and correct in actor, action, side, body part, object, and contact target. FAIL when it's missing; the wrong actor/side/body part/object/action is used; occlusion, cropping, or blur prevents verification; or (for speech) the mouth movement is absent or inconsistent with the required utterance." + }, + { + "id": "avb", + "key": "AVB", + "title": "Audio-Visual Binding", + "granularity": "per-event", + "summary": "Whether the audio and visual components within the same event are correctly synchronized and causally aligned.", + "detail": "Applies only to events with both audio and visual components (speech, sounding action — not silent action). PASS when the sound is synchronized with the corresponding visible action, impact, or mouth movement. FAILS if A or V fails, or the two are clearly misaligned in time. Temporal relations between different events are handled under TC_REL, not AVB." + }, + { + "id": "tcRel", + "key": "TC_REL", + "title": "Temporal Relation", + "granularity": "per-video", + "summary": "Whether the specified relationship between events holds - synchronization, order, and absolute timing.", + "detail": "Compare the defined start times of the required events. If any required event fails A, V, or AVB, TC_REL FAILS. Synchronization: PASS when events begin at the same moment (simple overlap is not enough). Order: PASS when events clearly begin in the required order (overlap allowed). Absolute timing: PASS when the event starts close to the required timestamp; +/-0.3s is a recommended guideline, not a strict cutoff - use judgment rather than a hard pass/fail line." + }, + { + "id": "tcCount", + "key": "TC_COUNT", + "title": "Temporal Count", + "granularity": "per-event", + "summary": "Whether each required event occurs the specified number of times, with no missing or extra occurrences.", + "detail": "Enter how many times the event actually occurs in the video. If no count is given, expect one. If the event fails A, V, or its applicable AVB, record its count as 0. A continuously held pose counts as one occurrence; releasing and re-forming it (or a clear distinct repetition without full release) counts as another." + } + ], + "events": [ + { + "id": "speech", + "modality": "speech", + "modalityLabel": "Speech", + "actor": "person_A", + "label": "Speech: say \"yes\"", + "definition": "The person clearly says the single word \"yes\". The speech is audible and easy to distinguish. No extra words are spoken unless explicitly required.", + "axes": { + "a": true, + "v": true, + "avb": true + } + }, + { + "id": "silent_action", + "modality": "silent_action", + "modalityLabel": "Silent action", + "actor": "person_A", + "label": "Silent action: salute pose (left hand)", + "definition": "The person's left hand holds a salute pose beside the left temple, with the fingers straight and together and the eyes and mouth visible. The target is the pose, not the raising motion.", + "axes": { + "a": false, + "v": true, + "avb": false + } + }, + { + "id": "sounding_action", + "modality": "sounding_action", + "modalityLabel": "Sounding action", + "actor": "person_A", + "label": "Sounding action: snapping fingers (with right hand)", + "definition": "The person snaps the fingers of the right hand: the thumb and middle finger visibly snap, and a clear snap sound occurs exactly at the moment of the snap.", + "axes": { + "a": true, + "v": true, + "avb": true + } + } + ], + "temporal_constraint": { + "relationType": "order_count", + "timingType": "relative", + "value": { + "ordered_events": [ + "speech", + "silent_action", + "sounding_action" + ], + "event_counts": { + "speech": 2, + "silent_action": 2, + "sounding_action": 1 + }, + "expanded_sequence": [ + "speech", + "speech", + "silent_action", + "silent_action", + "sounding_action" + ] + }, + "expectedCounts": { + "speech": 2, + "silent_action": 2, + "sounding_action": 1 + }, + "description": "Order: Speech: say \"yes\" x2 -> then -> Silent action: salute pose (left hand) x2 -> then -> Sounding action: snapping fingers (with right hand) x1." + }, + "model": "MOVA", + "item_id": "yes_x2-then-salute_pose_left_x2-then-snap_x1__order_count", + "canonical_id": "salute_pose_left_snap_yes_rel_order_count", + "variant": "v1", + "scene_hint": "plain indoor room", + "camera_hint": "static medium shot", + "prompt": "In a plain indoor room, a seated person faces the camera in a static medium shot, with the full face, mouth, upper body, both hands, and thighs clearly visible. The person starts seated on a chair with both hands initially resting on the knees or thighs. First, the person must say the single word \"yes\" exactly 2 times, with a brief pause between repetitions so each repetition is distinct. After that speech ends, the person must hold a salute pose with the left hand beside the left temple, fingers straight and together, with the eyes and mouth visible, exactly 2 times, returning the hand to rest between repetitions so each repetition is distinct. The target is the pose, not the raising motion. After that pose is visible, that hand returns to rest, then the person must snap the fingers of the right hand so the thumb and middle finger visibly snap, with a clear snap sound exactly at the moment of the snap once. The order must be strict: all repetitions of an event are completed before the next event begins. No extra words or extra target sounds occur. After the final event, no new target action occurs." + }, + "created_at": "2026-11-15T00:59:04.382373Z", + "annotation_role": "full" + }, + { + "id": "9552c352281b409c93c750e7c9c4f9a7", + "video_id": "okay-with-point_pose_left-with-peace_sign__sync__MOVA", + "user": "69eb6e157e23c6c0b2b7fb0d", + "annotations": { + "responses": { + "events": { + "speech": { + "a": { + "label": "FAIL", + "rationale": "There's additional words" + }, + "v": { + "label": "PASS", + "rationale": "" + }, + "avb": { + "label": "PASS", + "rationale": "" + }, + "tcCount": { + "value": "1", + "rationale": "" + } + }, + "silent_action": { + "v": { + "label": "FAIL", + "rationale": "No action using left hand" + }, + "tcCount": { + "value": "0", + "rationale": "" + } + }, + "silent_action_2": { + "v": { + "label": "PASS", + "rationale": "" + }, + "tcCount": { + "value": "1", + "rationale": "" + } + } + }, + "video": { + "tcRel": { + "label": "FAIL", + "rationale": "All the action are not done in synchronization and mostly the action are incorrect" + } + } + }, + "axis_definitions": [ + { + "id": "a", + "key": "A", + "title": "Audio Event Correctness", + "granularity": "per-event", + "summary": "Whether the required speech or sound event is present and correct in the audio.", + "detail": "PASS when every required speech/sound component is present, clear, and correct. FAIL when a required audio event is missing, the content or sound type is wrong, or the audio is too unclear to identify. Judge audibility leniently (a soft or quiet sound still counts); judge identity leniently within an acoustic family (clap/slap/tap/thud/pat) and strictly across families." + }, + { + "id": "v", + "key": "V", + "title": "Visual Event Correctness", + "granularity": "per-event", + "summary": "Whether the required visual event is present and correct. For speech events this evaluates visible speaking behavior (mouth/lip motion), not the spoken content.", + "detail": "PASS when every required visual component is present, visible, and correct in actor, action, side, body part, object, and contact target. FAIL when it's missing; the wrong actor/side/body part/object/action is used; occlusion, cropping, or blur prevents verification; or (for speech) the mouth movement is absent or inconsistent with the required utterance." + }, + { + "id": "avb", + "key": "AVB", + "title": "Audio-Visual Binding", + "granularity": "per-event", + "summary": "Whether the audio and visual components within the same event are correctly synchronized and causally aligned.", + "detail": "Applies only to events with both audio and visual components (speech, sounding action — not silent action). PASS when the sound is synchronized with the corresponding visible action, impact, or mouth movement. FAILS if A or V fails, or the two are clearly misaligned in time. Temporal relations between different events are handled under TC_REL, not AVB." + }, + { + "id": "tcRel", + "key": "TC_REL", + "title": "Temporal Relation", + "granularity": "per-video", + "summary": "Whether the specified relationship between events holds - synchronization, order, and absolute timing.", + "detail": "Compare the defined start times of the required events. If any required event fails A, V, or AVB, TC_REL FAILS. Synchronization: PASS when events begin at the same moment (simple overlap is not enough). Order: PASS when events clearly begin in the required order (overlap allowed). Absolute timing: PASS when the event starts close to the required timestamp; +/-0.3s is a recommended guideline, not a strict cutoff - use judgment rather than a hard pass/fail line." + }, + { + "id": "tcCount", + "key": "TC_COUNT", + "title": "Temporal Count", + "granularity": "per-event", + "summary": "Whether each required event occurs the specified number of times, with no missing or extra occurrences.", + "detail": "Enter how many times the event actually occurs in the video. If no count is given, expect one. If the event fails A, V, or its applicable AVB, record its count as 0. A continuously held pose counts as one occurrence; releasing and re-forming it (or a clear distinct repetition without full release) counts as another." + } + ], + "events": [ + { + "id": "speech", + "modality": "speech", + "modalityLabel": "Speech", + "actor": "person_A", + "label": "Speech: say \"okay\"", + "definition": "The person clearly says the single word \"okay\". The speech is audible and easy to distinguish. No extra words are spoken unless explicitly required.", + "axes": { + "a": true, + "v": true, + "avb": true + } + }, + { + "id": "silent_action", + "modality": "silent_action", + "modalityLabel": "Silent action", + "actor": "person_A", + "label": "Silent action: point pose (left hand)", + "definition": "The person's left hand holds a pointing pose toward a location on the person's left, with the index finger extended and the other fingers curled. The hand stays in frame. The target is the pose, not the raising motion.", + "axes": { + "a": false, + "v": true, + "avb": false + } + }, + { + "id": "silent_action_2", + "modality": "silent_action", + "modalityLabel": "Silent action", + "actor": "person_A", + "label": "Silent action: peace sign (right hand)", + "definition": "The person's right hand forms and holds a clear peace sign: the index and middle fingers are extended and separated in a V shape while the remaining fingers are folded.", + "axes": { + "a": false, + "v": true, + "avb": false + } + } + ], + "temporal_constraint": { + "relationType": "synchronization", + "timingType": "relative", + "value": { + "events": [ + "speech", + "silent_action", + "silent_action_2" + ] + }, + "expectedCounts": { + "speech": 1, + "silent_action": 1, + "silent_action_2": 1 + }, + "description": "Synchronization: Speech: say \"okay\" + Silent action: point pose (left hand) + Silent action: peace sign (right hand) must all occur at the same moment." + }, + "model": "MOVA", + "item_id": "okay-with-point_pose_left-with-peace_sign__sync", + "canonical_id": "okay_peace_sign_point_pose_left_rel_synchronization", + "variant": "v1", + "scene_hint": "plain indoor room", + "camera_hint": "static medium shot", + "prompt": "In a plain indoor room, a seated person faces the camera in a static medium shot, with the full face, mouth, upper body, both hands, and thighs clearly visible. The person starts seated on a chair with both hands initially resting on the knees or thighs, then performs all three target events at the same time: clearly says the single word \"okay\", holds a pointing pose with the left hand toward a location on the person's left, index finger extended and the other fingers curled, and forms and holds a clear right-hand peace sign, with the index and middle fingers extended in a V and the remaining fingers folded. The speech is audible and easy to distinguish, with no extra words. The target is the pose, not the raising motion. The target is the pose, not the raising motion. After the final event, no new target action occurs." + }, + "created_at": "2026-11-15T01:00:34.535270Z", + "annotation_role": "full" + }, + { + "id": "7199afb097c24339a8324bebb57d576c", + "video_id": "crossed_arms_x2__count__MOVA", + "user": "69eb6e157e23c6c0b2b7fb0d", + "annotations": { + "responses": { + "events": { + "silent_action": { + "v": { + "label": "FAIL", + "rationale": "" + }, + "tcCount": { + "value": "1", + "rationale": "" + } + } + }, + "video": { + "tcRel": { + "label": "FAIL", + "rationale": "The person only did once, with additional unrequested words spoken" + } + } + }, + "axis_definitions": [ + { + "id": "a", + "key": "A", + "title": "Audio Event Correctness", + "granularity": "per-event", + "summary": "Whether the required speech or sound event is present and correct in the audio.", + "detail": "PASS when every required speech/sound component is present, clear, and correct. FAIL when a required audio event is missing, the content or sound type is wrong, or the audio is too unclear to identify. Judge audibility leniently (a soft or quiet sound still counts); judge identity leniently within an acoustic family (clap/slap/tap/thud/pat) and strictly across families." + }, + { + "id": "v", + "key": "V", + "title": "Visual Event Correctness", + "granularity": "per-event", + "summary": "Whether the required visual event is present and correct. For speech events this evaluates visible speaking behavior (mouth/lip motion), not the spoken content.", + "detail": "PASS when every required visual component is present, visible, and correct in actor, action, side, body part, object, and contact target. FAIL when it's missing; the wrong actor/side/body part/object/action is used; occlusion, cropping, or blur prevents verification; or (for speech) the mouth movement is absent or inconsistent with the required utterance." + }, + { + "id": "avb", + "key": "AVB", + "title": "Audio-Visual Binding", + "granularity": "per-event", + "summary": "Whether the audio and visual components within the same event are correctly synchronized and causally aligned.", + "detail": "Applies only to events with both audio and visual components (speech, sounding action — not silent action). PASS when the sound is synchronized with the corresponding visible action, impact, or mouth movement. FAILS if A or V fails, or the two are clearly misaligned in time. Temporal relations between different events are handled under TC_REL, not AVB." + }, + { + "id": "tcRel", + "key": "TC_REL", + "title": "Temporal Relation", + "granularity": "per-video", + "summary": "Whether the specified relationship between events holds - synchronization, order, and absolute timing.", + "detail": "Compare the defined start times of the required events. If any required event fails A, V, or AVB, TC_REL FAILS. Synchronization: PASS when events begin at the same moment (simple overlap is not enough). Order: PASS when events clearly begin in the required order (overlap allowed). Absolute timing: PASS when the event starts close to the required timestamp; +/-0.3s is a recommended guideline, not a strict cutoff - use judgment rather than a hard pass/fail line." + }, + { + "id": "tcCount", + "key": "TC_COUNT", + "title": "Temporal Count", + "granularity": "per-event", + "summary": "Whether each required event occurs the specified number of times, with no missing or extra occurrences.", + "detail": "Enter how many times the event actually occurs in the video. If no count is given, expect one. If the event fails A, V, or its applicable AVB, record its count as 0. A continuously held pose counts as one occurrence; releasing and re-forming it (or a clear distinct repetition without full release) counts as another." + } + ], + "events": [ + { + "id": "silent_action", + "modality": "silent_action", + "modalityLabel": "Silent action", + "actor": "person_A", + "label": "Silent action: crossed arms (both arms)", + "definition": "The person holds both arms crossed over the chest. The target is the pose, not the crossing motion.", + "axes": { + "a": false, + "v": true, + "avb": false + } + } + ], + "temporal_constraint": { + "relationType": "count", + "timingType": "relative", + "value": { + "event": "silent_action", + "count": 2 + }, + "expectedCounts": { + "silent_action": 2 + }, + "description": "Count: Silent action: crossed arms (both arms) must occur exactly 2 times." + }, + "model": "MOVA", + "item_id": "crossed_arms_x2__count", + "canonical_id": "crossed_arms_rel_count", + "variant": "v1", + "scene_hint": "plain indoor room", + "camera_hint": "static medium shot", + "prompt": "In a plain indoor room, a seated person faces the camera in a static medium shot, with the full face, mouth, upper body, both hands, and thighs clearly visible. The person starts seated on a chair with both hands initially resting on the knees or thighs. The person crosses both arms over the chest and holds the pose exactly 2 times, returning both arms to rest between repetitions so each repetition is distinct. The target is the pose, not the crossing motion. No extra words or extra target sounds occur. After the final event, no new target action occurs." + }, + "created_at": "2026-11-15T01:01:43.315359Z", + "annotation_role": "full" + }, + { + "id": "7a6ffa5a15a34b3ea98c2906e54c1c4d", + "video_id": "yes-then-knock__order__MOVA", + "user": "69eb6e157e23c6c0b2b7fb0d", + "annotations": { + "responses": { + "events": { + "speech": { + "a": { + "label": "FAIL", + "rationale": "The person spoke extra words" + }, + "v": { + "label": "PASS", + "rationale": "" + }, + "avb": { + "label": "PASS", + "rationale": "" + }, + "tcCount": { + "value": "1", + "rationale": "" + } + }, + "sounding_action": { + "a": { + "label": "FAIL", + "rationale": "He uses left hand and the table are not on his right side" + }, + "v": { + "label": "FAIL", + "rationale": "" + }, + "avb": { + "label": "FAIL", + "rationale": "" + }, + "tcCount": { + "value": "0", + "rationale": "" + } + } + }, + "video": { + "tcRel": { + "label": "FAIL", + "rationale": "The requested action are mostly done incorrectly." + } + } + }, + "axis_definitions": [ + { + "id": "a", + "key": "A", + "title": "Audio Event Correctness", + "granularity": "per-event", + "summary": "Whether the required speech or sound event is present and correct in the audio.", + "detail": "PASS when every required speech/sound component is present, clear, and correct. FAIL when a required audio event is missing, the content or sound type is wrong, or the audio is too unclear to identify. Judge audibility leniently (a soft or quiet sound still counts); judge identity leniently within an acoustic family (clap/slap/tap/thud/pat) and strictly across families." + }, + { + "id": "v", + "key": "V", + "title": "Visual Event Correctness", + "granularity": "per-event", + "summary": "Whether the required visual event is present and correct. For speech events this evaluates visible speaking behavior (mouth/lip motion), not the spoken content.", + "detail": "PASS when every required visual component is present, visible, and correct in actor, action, side, body part, object, and contact target. FAIL when it's missing; the wrong actor/side/body part/object/action is used; occlusion, cropping, or blur prevents verification; or (for speech) the mouth movement is absent or inconsistent with the required utterance." + }, + { + "id": "avb", + "key": "AVB", + "title": "Audio-Visual Binding", + "granularity": "per-event", + "summary": "Whether the audio and visual components within the same event are correctly synchronized and causally aligned.", + "detail": "Applies only to events with both audio and visual components (speech, sounding action — not silent action). PASS when the sound is synchronized with the corresponding visible action, impact, or mouth movement. FAILS if A or V fails, or the two are clearly misaligned in time. Temporal relations between different events are handled under TC_REL, not AVB." + }, + { + "id": "tcRel", + "key": "TC_REL", + "title": "Temporal Relation", + "granularity": "per-video", + "summary": "Whether the specified relationship between events holds - synchronization, order, and absolute timing.", + "detail": "Compare the defined start times of the required events. If any required event fails A, V, or AVB, TC_REL FAILS. Synchronization: PASS when events begin at the same moment (simple overlap is not enough). Order: PASS when events clearly begin in the required order (overlap allowed). Absolute timing: PASS when the event starts close to the required timestamp; +/-0.3s is a recommended guideline, not a strict cutoff - use judgment rather than a hard pass/fail line." + }, + { + "id": "tcCount", + "key": "TC_COUNT", + "title": "Temporal Count", + "granularity": "per-event", + "summary": "Whether each required event occurs the specified number of times, with no missing or extra occurrences.", + "detail": "Enter how many times the event actually occurs in the video. If no count is given, expect one. If the event fails A, V, or its applicable AVB, record its count as 0. A continuously held pose counts as one occurrence; releasing and re-forming it (or a clear distinct repetition without full release) counts as another." + } + ], + "events": [ + { + "id": "speech", + "modality": "speech", + "modalityLabel": "Speech", + "actor": "person_A", + "label": "Speech: say \"yes\"", + "definition": "The person clearly says the single word \"yes\". The speech is audible and easy to distinguish. No extra words are spoken unless explicitly required.", + "axes": { + "a": true, + "v": true, + "avb": true + } + }, + { + "id": "sounding_action", + "modality": "sounding_action", + "modalityLabel": "Sounding action", + "actor": "person_A", + "label": "Sounding action: knock (with right hand on right side table)", + "definition": "The person's right hand forms a loose fist and strikes the small side-table surface once with the knuckles, with visible contact and a sharp knock exactly at the moment of impact.", + "axes": { + "a": true, + "v": true, + "avb": true + } + } + ], + "temporal_constraint": { + "relationType": "order", + "timingType": "relative", + "value": { + "sequence": [ + "speech", + "sounding_action" + ] + }, + "expectedCounts": { + "speech": 1, + "sounding_action": 1 + }, + "description": "Order: Speech: say \"yes\" -> then -> Sounding action: knock (with right hand on right side table)." + }, + "model": "MOVA", + "item_id": "yes-then-knock__order", + "canonical_id": "knock_yes_rel_order", + "variant": "v1", + "scene_hint": "plain indoor room", + "camera_hint": "static medium shot", + "prompt": "In a plain indoor room, a seated person faces the camera in a static medium shot, with the full face, mouth, upper body, both hands, and thighs clearly visible. A small side table stands beside the person's right hip, outside the frontal silhouette of both thighs; its top and legs do not obscure either thigh, and its contact surface and the acting hand remain in the frame. The person starts seated on a chair with both hands initially resting on the knees or thighs. First, the person clearly says the single word \"yes\". After that speech ends, the person forms a loose fist with the right hand and strikes the small side-table surface once with the knuckles, with visible contact and a sharp knock exactly at the moment of impact. No extra words or extra target sounds occur. After the final event, no new target action occurs." + }, + "created_at": "2026-11-15T01:06:51.385148Z", + "annotation_role": "full" + }, + { + "id": "5a823e036156471bb1301c80c8a0df47", + "video_id": "bird_x2-then-open_palm_x2-then-slap_left_x1__order_count__MOVA", + "user": "69eb6e157e23c6c0b2b7fb0d", + "annotations": { + "responses": { + "events": { + "speech": { + "a": { + "label": "FAIL", + "rationale": "The person spoke extra words" + }, + "v": { + "label": "PASS", + "rationale": "" + }, + "avb": { + "label": "PASS", + "rationale": "" + }, + "tcCount": { + "value": "1", + "rationale": "" + } + }, + "silent_action": { + "v": { + "label": "FAIL", + "rationale": "Only done once" + }, + "tcCount": { + "value": "1", + "rationale": "" + } + }, + "sounding_action": { + "a": { + "label": "FAIL", + "rationale": "No action done" + }, + "v": { + "label": "FAIL", + "rationale": "" + }, + "avb": { + "label": "FAIL", + "rationale": "" + }, + "tcCount": { + "value": "0", + "rationale": "" + } + } + }, + "video": { + "tcRel": { + "label": "FAIL", + "rationale": "The required action is not met the instruction and failed to be followed" + } + } + }, + "axis_definitions": [ + { + "id": "a", + "key": "A", + "title": "Audio Event Correctness", + "granularity": "per-event", + "summary": "Whether the required speech or sound event is present and correct in the audio.", + "detail": "PASS when every required speech/sound component is present, clear, and correct. FAIL when a required audio event is missing, the content or sound type is wrong, or the audio is too unclear to identify. Judge audibility leniently (a soft or quiet sound still counts); judge identity leniently within an acoustic family (clap/slap/tap/thud/pat) and strictly across families." + }, + { + "id": "v", + "key": "V", + "title": "Visual Event Correctness", + "granularity": "per-event", + "summary": "Whether the required visual event is present and correct. For speech events this evaluates visible speaking behavior (mouth/lip motion), not the spoken content.", + "detail": "PASS when every required visual component is present, visible, and correct in actor, action, side, body part, object, and contact target. FAIL when it's missing; the wrong actor/side/body part/object/action is used; occlusion, cropping, or blur prevents verification; or (for speech) the mouth movement is absent or inconsistent with the required utterance." + }, + { + "id": "avb", + "key": "AVB", + "title": "Audio-Visual Binding", + "granularity": "per-event", + "summary": "Whether the audio and visual components within the same event are correctly synchronized and causally aligned.", + "detail": "Applies only to events with both audio and visual components (speech, sounding action — not silent action). PASS when the sound is synchronized with the corresponding visible action, impact, or mouth movement. FAILS if A or V fails, or the two are clearly misaligned in time. Temporal relations between different events are handled under TC_REL, not AVB." + }, + { + "id": "tcRel", + "key": "TC_REL", + "title": "Temporal Relation", + "granularity": "per-video", + "summary": "Whether the specified relationship between events holds - synchronization, order, and absolute timing.", + "detail": "Compare the defined start times of the required events. If any required event fails A, V, or AVB, TC_REL FAILS. Synchronization: PASS when events begin at the same moment (simple overlap is not enough). Order: PASS when events clearly begin in the required order (overlap allowed). Absolute timing: PASS when the event starts close to the required timestamp; +/-0.3s is a recommended guideline, not a strict cutoff - use judgment rather than a hard pass/fail line." + }, + { + "id": "tcCount", + "key": "TC_COUNT", + "title": "Temporal Count", + "granularity": "per-event", + "summary": "Whether each required event occurs the specified number of times, with no missing or extra occurrences.", + "detail": "Enter how many times the event actually occurs in the video. If no count is given, expect one. If the event fails A, V, or its applicable AVB, record its count as 0. A continuously held pose counts as one occurrence; releasing and re-forming it (or a clear distinct repetition without full release) counts as another." + } + ], + "events": [ + { + "id": "speech", + "modality": "speech", + "modalityLabel": "Speech", + "actor": "person_A", + "label": "Speech: say \"bird\"", + "definition": "The person clearly says the single word \"bird\". The speech is audible and easy to distinguish. No extra words are spoken unless explicitly required.", + "axes": { + "a": true, + "v": true, + "avb": true + } + }, + { + "id": "silent_action", + "modality": "silent_action", + "modalityLabel": "Silent action", + "actor": "person_A", + "label": "Silent action: raise open palm (right hand)", + "definition": "The person's right hand forms or holds a clear open-palm pose with the palm open and facing the camera, fingers together and extended. The target is the held pose, not a separate raising motion.", + "axes": { + "a": false, + "v": true, + "avb": false + } + }, + { + "id": "sounding_action", + "modality": "sounding_action", + "modalityLabel": "Sounding action", + "actor": "person_A", + "label": "Sounding action: slapping the thigh (with left hand on left thigh)", + "definition": "The person's left hand strikes the top of the person's left thigh. Visible contact is made with the thigh, and a sharp slap sound occurs exactly at the moment of impact.", + "axes": { + "a": true, + "v": true, + "avb": true + } + } + ], + "temporal_constraint": { + "relationType": "order_count", + "timingType": "relative", + "value": { + "ordered_events": [ + "speech", + "silent_action", + "sounding_action" + ], + "event_counts": { + "speech": 2, + "silent_action": 2, + "sounding_action": 1 + }, + "expanded_sequence": [ + "speech", + "speech", + "silent_action", + "silent_action", + "sounding_action" + ] + }, + "expectedCounts": { + "speech": 2, + "silent_action": 2, + "sounding_action": 1 + }, + "description": "Order: Speech: say \"bird\" x2 -> then -> Silent action: raise open palm (right hand) x2 -> then -> Sounding action: slapping the thigh (with left hand on left thigh) x1." + }, + "model": "MOVA", + "item_id": "bird_x2-then-open_palm_x2-then-slap_left_x1__order_count", + "canonical_id": "bird_open_palm_slap_left_rel_order_count", + "variant": "v1", + "scene_hint": "plain indoor room", + "camera_hint": "static medium shot", + "prompt": "In a plain indoor room, a seated person faces the camera in a static medium shot, with the full face, mouth, upper body, both hands, and thighs clearly visible. The person starts seated on a chair with both hands initially resting on the knees or thighs. First, the person must say the single word \"bird\" exactly 2 times, with a brief pause between repetitions so each repetition is distinct. After that speech ends, the person must form and hold a clear right-hand open-palm pose with the palm facing the camera and the fingers together and extended exactly 2 times, returning the hand to rest between repetitions so each repetition is distinct. The target is the pose, not the raising motion. After that pose is visible, that hand returns to rest, then the person must strike the top of the left thigh with the left hand, making visible contact and a sharp slap sound exactly at the moment of impact once. The order must be strict: all repetitions of an event are completed before the next event begins. No extra words or extra target sounds occur. After the final event, no new target action occurs." + }, + "created_at": "2026-11-15T01:08:56.572016Z", + "annotation_role": "full" + }, + { + "id": "d9068b2c95624fa0b2e5c0ccf4db9e4f", + "video_id": "house_x2-then-ok_sign_x2-then-nod_head_x2-then-knock_x2__order_count__MOVA", + "user": "69eb6e157e23c6c0b2b7fb0d", + "annotations": { + "responses": { + "events": { + "speech": { + "a": { + "label": "FAIL", + "rationale": "There are extra words spoken" + }, + "v": { + "label": "FAIL", + "rationale": "the lip motions are incorrect" + }, + "avb": { + "label": "FAIL", + "rationale": "The visual didn't match the audio" + }, + "tcCount": { + "value": "1", + "rationale": "Only done once not twice as requested" + } + }, + "silent_action": { + "v": { + "label": "FAIL", + "rationale": "No action done" + }, + "tcCount": { + "value": "0", + "rationale": "" + } + }, + "silent_action_2": { + "v": { + "label": "FAIL", + "rationale": "" + }, + "tcCount": { + "value": "0", + "rationale": "" + } + }, + "sounding_action": { + "a": { + "label": "PASS", + "rationale": "" + }, + "v": { + "label": "FAIL", + "rationale": "" + }, + "avb": { + "label": "FAIL", + "rationale": "" + }, + "tcCount": { + "value": "1", + "rationale": "" + } + } + }, + "video": { + "tcRel": { + "label": "FAIL", + "rationale": "Only the knock sound that passed, the other aspect are failed" + } + } + }, + "axis_definitions": [ + { + "id": "a", + "key": "A", + "title": "Audio Event Correctness", + "granularity": "per-event", + "summary": "Whether the required speech or sound event is present and correct in the audio.", + "detail": "PASS when every required speech/sound component is present, clear, and correct. FAIL when a required audio event is missing, the content or sound type is wrong, or the audio is too unclear to identify. Judge audibility leniently (a soft or quiet sound still counts); judge identity leniently within an acoustic family (clap/slap/tap/thud/pat) and strictly across families." + }, + { + "id": "v", + "key": "V", + "title": "Visual Event Correctness", + "granularity": "per-event", + "summary": "Whether the required visual event is present and correct. For speech events this evaluates visible speaking behavior (mouth/lip motion), not the spoken content.", + "detail": "PASS when every required visual component is present, visible, and correct in actor, action, side, body part, object, and contact target. FAIL when it's missing; the wrong actor/side/body part/object/action is used; occlusion, cropping, or blur prevents verification; or (for speech) the mouth movement is absent or inconsistent with the required utterance." + }, + { + "id": "avb", + "key": "AVB", + "title": "Audio-Visual Binding", + "granularity": "per-event", + "summary": "Whether the audio and visual components within the same event are correctly synchronized and causally aligned.", + "detail": "Applies only to events with both audio and visual components (speech, sounding action — not silent action). PASS when the sound is synchronized with the corresponding visible action, impact, or mouth movement. FAILS if A or V fails, or the two are clearly misaligned in time. Temporal relations between different events are handled under TC_REL, not AVB." + }, + { + "id": "tcRel", + "key": "TC_REL", + "title": "Temporal Relation", + "granularity": "per-video", + "summary": "Whether the specified relationship between events holds - synchronization, order, and absolute timing.", + "detail": "Compare the defined start times of the required events. If any required event fails A, V, or AVB, TC_REL FAILS. Synchronization: PASS when events begin at the same moment (simple overlap is not enough). Order: PASS when events clearly begin in the required order (overlap allowed). Absolute timing: PASS when the event starts close to the required timestamp; +/-0.3s is a recommended guideline, not a strict cutoff - use judgment rather than a hard pass/fail line." + }, + { + "id": "tcCount", + "key": "TC_COUNT", + "title": "Temporal Count", + "granularity": "per-event", + "summary": "Whether each required event occurs the specified number of times, with no missing or extra occurrences.", + "detail": "Enter how many times the event actually occurs in the video. If no count is given, expect one. If the event fails A, V, or its applicable AVB, record its count as 0. A continuously held pose counts as one occurrence; releasing and re-forming it (or a clear distinct repetition without full release) counts as another." + } + ], + "events": [ + { + "id": "speech", + "modality": "speech", + "modalityLabel": "Speech", + "actor": "person_A", + "label": "Speech: say \"house\"", + "definition": "The person clearly says the single word \"house\". The speech is audible and easy to distinguish. No extra words are spoken unless explicitly required.", + "axes": { + "a": true, + "v": true, + "avb": true + } + }, + { + "id": "silent_action", + "modality": "silent_action", + "modalityLabel": "Silent action", + "actor": "person_A", + "label": "Silent action: ok sign (right hand)", + "definition": "The person's right thumb and index fingertip touch to form a ring while the other three fingers remain extended and the palm faces the camera. The target is the held OK pose, not the raising motion.", + "axes": { + "a": false, + "v": true, + "avb": false + } + }, + { + "id": "silent_action_2", + "modality": "silent_action", + "modalityLabel": "Silent action", + "actor": "person_A", + "label": "Silent action: nod head (N/A head)", + "definition": "The person lowers the chin slightly once and returns to the neutral forward-facing position, without bowing the torso. One complete down-and-return cycle is one event. Keep the full face and mouth visible throughout.", + "axes": { + "a": false, + "v": true, + "avb": false + } + }, + { + "id": "sounding_action", + "modality": "sounding_action", + "modalityLabel": "Sounding action", + "actor": "person_A", + "label": "Sounding action: knock (with right hand on right side table)", + "definition": "The person's right hand forms a loose fist and strikes the small side-table surface once with the knuckles, with visible contact and a sharp knock exactly at the moment of impact.", + "axes": { + "a": true, + "v": true, + "avb": true + } + } + ], + "temporal_constraint": { + "relationType": "order_count", + "timingType": "relative", + "value": { + "ordered_events": [ + "speech", + "silent_action", + "silent_action_2", + "sounding_action" + ], + "event_counts": { + "speech": 2, + "silent_action": 2, + "silent_action_2": 2, + "sounding_action": 2 + }, + "expanded_sequence": [ + "speech", + "speech", + "silent_action", + "silent_action", + "silent_action_2", + "silent_action_2", + "sounding_action", + "sounding_action" + ] + }, + "expectedCounts": { + "speech": 2, + "silent_action": 2, + "silent_action_2": 2, + "sounding_action": 2 + }, + "description": "Order: Speech: say \"house\" x2 -> then -> Silent action: ok sign (right hand) x2 -> then -> Silent action: nod head (N/A head) x2 -> then -> Sounding action: knock (with right hand on right side table) x2." + }, + "model": "MOVA", + "item_id": "house_x2-then-ok_sign_x2-then-nod_head_x2-then-knock_x2__order_count", + "canonical_id": "house_knock_nod_head_ok_sign_rel_order_count", + "variant": "v1", + "scene_hint": "plain indoor room", + "camera_hint": "static medium shot", + "prompt": "In a plain indoor room, a seated person faces the camera in a static medium shot, with the full face, mouth, upper body, both hands, and thighs clearly visible. A small side table stands beside the person's right hip, outside the frontal silhouette of both thighs; its top and legs do not obscure either thigh, and its contact surface and the acting hand remain in the frame. The person starts seated on a chair with both hands initially resting on the knees or thighs. First, the person must say the single word \"house\" exactly 2 times, with a brief pause between repetitions so each repetition is distinct. After that speech ends, the person must hold an OK sign with the right hand, thumb and index fingertip forming a ring and the other three fingers extended, palm facing the camera, exactly 2 times, returning the hand to rest between repetitions so each repetition is distinct. The target is the pose, not the raising motion. After that pose is visible, that hand returns to rest, then the person must lower the chin slightly and returns to the neutral forward-facing position, without bowing the torso, exactly 2 times, with a brief pause between repetitions so each repetition is distinct. After that movement is complete, the person must form a loose fist with the right hand and strike the small side-table surface with the knuckles, with visible contact and a sharp knock exactly at the moment of impact, exactly 2 times, with a brief pause between repetitions so each repetition is distinct. The order must be strict: all repetitions of an event are completed before the next event begins. No extra words or extra target sounds occur. After the final event, no new target action occurs." + }, + "created_at": "2026-11-15T01:12:14.380214Z", + "annotation_role": "full" + }, + { + "id": "c14b898ff92f43b1ad81d15a5a25e0c7", + "video_id": "okay_x2-then-ok_sign_x1-then-snap_x2__order_count__MOVA", + "user": "69eb6e157e23c6c0b2b7fb0d", + "annotations": { + "responses": { + "events": { + "speech": { + "a": { + "label": "FAIL", + "rationale": "There are extra words spoken" + }, + "v": { + "label": "PASS", + "rationale": "" + }, + "avb": { + "label": "PASS", + "rationale": "" + }, + "tcCount": { + "value": "1", + "rationale": "" + } + }, + "silent_action": { + "v": { + "label": "PASS", + "rationale": "" + }, + "tcCount": { + "value": "1", + "rationale": "" + } + }, + "sounding_action": { + "a": { + "label": "FAIL", + "rationale": "No action done" + }, + "v": { + "label": "FAIL", + "rationale": "" + }, + "avb": { + "label": "FAIL", + "rationale": "" + }, + "tcCount": { + "value": "0", + "rationale": "" + } + } + }, + "video": { + "tcRel": { + "label": "FAIL", + "rationale": "Most of the requested action are not followed correctly" + } + } + }, + "axis_definitions": [ + { + "id": "a", + "key": "A", + "title": "Audio Event Correctness", + "granularity": "per-event", + "summary": "Whether the required speech or sound event is present and correct in the audio.", + "detail": "PASS when every required speech/sound component is present, clear, and correct. FAIL when a required audio event is missing, the content or sound type is wrong, or the audio is too unclear to identify. Judge audibility leniently (a soft or quiet sound still counts); judge identity leniently within an acoustic family (clap/slap/tap/thud/pat) and strictly across families." + }, + { + "id": "v", + "key": "V", + "title": "Visual Event Correctness", + "granularity": "per-event", + "summary": "Whether the required visual event is present and correct. For speech events this evaluates visible speaking behavior (mouth/lip motion), not the spoken content.", + "detail": "PASS when every required visual component is present, visible, and correct in actor, action, side, body part, object, and contact target. FAIL when it's missing; the wrong actor/side/body part/object/action is used; occlusion, cropping, or blur prevents verification; or (for speech) the mouth movement is absent or inconsistent with the required utterance." + }, + { + "id": "avb", + "key": "AVB", + "title": "Audio-Visual Binding", + "granularity": "per-event", + "summary": "Whether the audio and visual components within the same event are correctly synchronized and causally aligned.", + "detail": "Applies only to events with both audio and visual components (speech, sounding action — not silent action). PASS when the sound is synchronized with the corresponding visible action, impact, or mouth movement. FAILS if A or V fails, or the two are clearly misaligned in time. Temporal relations between different events are handled under TC_REL, not AVB." + }, + { + "id": "tcRel", + "key": "TC_REL", + "title": "Temporal Relation", + "granularity": "per-video", + "summary": "Whether the specified relationship between events holds - synchronization, order, and absolute timing.", + "detail": "Compare the defined start times of the required events. If any required event fails A, V, or AVB, TC_REL FAILS. Synchronization: PASS when events begin at the same moment (simple overlap is not enough). Order: PASS when events clearly begin in the required order (overlap allowed). Absolute timing: PASS when the event starts close to the required timestamp; +/-0.3s is a recommended guideline, not a strict cutoff - use judgment rather than a hard pass/fail line." + }, + { + "id": "tcCount", + "key": "TC_COUNT", + "title": "Temporal Count", + "granularity": "per-event", + "summary": "Whether each required event occurs the specified number of times, with no missing or extra occurrences.", + "detail": "Enter how many times the event actually occurs in the video. If no count is given, expect one. If the event fails A, V, or its applicable AVB, record its count as 0. A continuously held pose counts as one occurrence; releasing and re-forming it (or a clear distinct repetition without full release) counts as another." + } + ], + "events": [ + { + "id": "speech", + "modality": "speech", + "modalityLabel": "Speech", + "actor": "person_A", + "label": "Speech: say \"okay\"", + "definition": "The person clearly says the single word \"okay\". The speech is audible and easy to distinguish. No extra words are spoken unless explicitly required.", + "axes": { + "a": true, + "v": true, + "avb": true + } + }, + { + "id": "silent_action", + "modality": "silent_action", + "modalityLabel": "Silent action", + "actor": "person_A", + "label": "Silent action: ok sign (right hand)", + "definition": "The person's right thumb and index fingertip touch to form a ring while the other three fingers remain extended and the palm faces the camera. The target is the held OK pose, not the raising motion.", + "axes": { + "a": false, + "v": true, + "avb": false + } + }, + { + "id": "sounding_action", + "modality": "sounding_action", + "modalityLabel": "Sounding action", + "actor": "person_A", + "label": "Sounding action: snapping fingers (with right hand)", + "definition": "The person snaps the fingers of the right hand: the thumb and middle finger visibly snap, and a clear snap sound occurs exactly at the moment of the snap.", + "axes": { + "a": true, + "v": true, + "avb": true + } + } + ], + "temporal_constraint": { + "relationType": "order_count", + "timingType": "relative", + "value": { + "ordered_events": [ + "speech", + "silent_action", + "sounding_action" + ], + "event_counts": { + "speech": 2, + "silent_action": 1, + "sounding_action": 2 + }, + "expanded_sequence": [ + "speech", + "speech", + "silent_action", + "sounding_action", + "sounding_action" + ] + }, + "expectedCounts": { + "speech": 2, + "silent_action": 1, + "sounding_action": 2 + }, + "description": "Order: Speech: say \"okay\" x2 -> then -> Silent action: ok sign (right hand) x1 -> then -> Sounding action: snapping fingers (with right hand) x2." + }, + "model": "MOVA", + "item_id": "okay_x2-then-ok_sign_x1-then-snap_x2__order_count", + "canonical_id": "ok_sign_okay_snap_rel_order_count", + "variant": "v1", + "scene_hint": "plain indoor room", + "camera_hint": "static medium shot", + "prompt": "In a plain indoor room, a seated person faces the camera in a static medium shot, with the full face, mouth, upper body, both hands, and thighs clearly visible. The person starts seated on a chair with both hands initially resting on the knees or thighs. First, the person must say the single word \"okay\" exactly 2 times, with a brief pause between repetitions so each repetition is distinct. After that speech ends, the person must hold an OK sign with the right hand, thumb and index fingertip forming a ring and the other three fingers extended, palm facing the camera once. The target is the pose, not the raising motion. After that pose is visible, that hand returns to rest, then the person must snap the fingers of the right hand so the thumb and middle finger visibly snap, with a clear snap sound exactly at the moment of the snap, exactly 2 times, with a brief pause between repetitions so each repetition is distinct. The order must be strict: all repetitions of an event are completed before the next event begins. No extra words or extra target sounds occur. After the final event, no new target action occurs." + }, + "created_at": "2026-11-15T01:13:33.265329Z", + "annotation_role": "full" + }, + { + "id": "9cf7e7a00c4943f48fb46c579259ab43", + "video_id": "up-with-open_palm_left__sync__MOVA", + "user": "69eb6e157e23c6c0b2b7fb0d", + "annotations": { + "responses": { + "events": { + "speech": { + "a": { + "label": "FAIL", + "rationale": "Extra words are spoken" + }, + "v": { + "label": "PASS", + "rationale": "" + }, + "avb": { + "label": "PASS", + "rationale": "" + }, + "tcCount": { + "value": "1", + "rationale": "" + } + }, + "silent_action": { + "v": { + "label": "FAIL", + "rationale": "The person uses right hand instead" + }, + "tcCount": { + "value": "0", + "rationale": "" + } + } + }, + "video": { + "tcRel": { + "label": "FAIL", + "rationale": "The action is not done in synchronization" + } + } + }, + "axis_definitions": [ + { + "id": "a", + "key": "A", + "title": "Audio Event Correctness", + "granularity": "per-event", + "summary": "Whether the required speech or sound event is present and correct in the audio.", + "detail": "PASS when every required speech/sound component is present, clear, and correct. FAIL when a required audio event is missing, the content or sound type is wrong, or the audio is too unclear to identify. Judge audibility leniently (a soft or quiet sound still counts); judge identity leniently within an acoustic family (clap/slap/tap/thud/pat) and strictly across families." + }, + { + "id": "v", + "key": "V", + "title": "Visual Event Correctness", + "granularity": "per-event", + "summary": "Whether the required visual event is present and correct. For speech events this evaluates visible speaking behavior (mouth/lip motion), not the spoken content.", + "detail": "PASS when every required visual component is present, visible, and correct in actor, action, side, body part, object, and contact target. FAIL when it's missing; the wrong actor/side/body part/object/action is used; occlusion, cropping, or blur prevents verification; or (for speech) the mouth movement is absent or inconsistent with the required utterance." + }, + { + "id": "avb", + "key": "AVB", + "title": "Audio-Visual Binding", + "granularity": "per-event", + "summary": "Whether the audio and visual components within the same event are correctly synchronized and causally aligned.", + "detail": "Applies only to events with both audio and visual components (speech, sounding action — not silent action). PASS when the sound is synchronized with the corresponding visible action, impact, or mouth movement. FAILS if A or V fails, or the two are clearly misaligned in time. Temporal relations between different events are handled under TC_REL, not AVB." + }, + { + "id": "tcRel", + "key": "TC_REL", + "title": "Temporal Relation", + "granularity": "per-video", + "summary": "Whether the specified relationship between events holds - synchronization, order, and absolute timing.", + "detail": "Compare the defined start times of the required events. If any required event fails A, V, or AVB, TC_REL FAILS. Synchronization: PASS when events begin at the same moment (simple overlap is not enough). Order: PASS when events clearly begin in the required order (overlap allowed). Absolute timing: PASS when the event starts close to the required timestamp; +/-0.3s is a recommended guideline, not a strict cutoff - use judgment rather than a hard pass/fail line." + }, + { + "id": "tcCount", + "key": "TC_COUNT", + "title": "Temporal Count", + "granularity": "per-event", + "summary": "Whether each required event occurs the specified number of times, with no missing or extra occurrences.", + "detail": "Enter how many times the event actually occurs in the video. If no count is given, expect one. If the event fails A, V, or its applicable AVB, record its count as 0. A continuously held pose counts as one occurrence; releasing and re-forming it (or a clear distinct repetition without full release) counts as another." + } + ], + "events": [ + { + "id": "speech", + "modality": "speech", + "modalityLabel": "Speech", + "actor": "person_A", + "label": "Speech: say \"up\"", + "definition": "The person clearly says the single word \"up\". The speech is audible and easy to distinguish. No extra words are spoken unless explicitly required.", + "axes": { + "a": true, + "v": true, + "avb": true + } + }, + { + "id": "silent_action", + "modality": "silent_action", + "modalityLabel": "Silent action", + "actor": "person_A", + "label": "Silent action: raise open palm (left hand)", + "definition": "The person's left hand forms or holds a clear open-palm pose with the palm open and facing the camera, fingers together and extended. The target is the held pose, not a separate raising motion.", + "axes": { + "a": false, + "v": true, + "avb": false + } + } + ], + "temporal_constraint": { + "relationType": "synchronization", + "timingType": "relative", + "value": { + "events": [ + "speech", + "silent_action" + ] + }, + "expectedCounts": { + "speech": 1, + "silent_action": 1 + }, + "description": "Synchronization: Speech: say \"up\" + Silent action: raise open palm (left hand) must all occur at the same moment." + }, + "model": "MOVA", + "item_id": "up-with-open_palm_left__sync", + "canonical_id": "open_palm_left_up_rel_synchronization", + "variant": "v1", + "scene_hint": "plain indoor room", + "camera_hint": "static medium shot", + "prompt": "In a plain indoor room, a seated person faces the camera in a static medium shot, with the full face, mouth, upper body, both hands, and thighs clearly visible. The person starts seated on a chair with both hands initially resting on the knees or thighs, then at one synchronized moment the person clearly says the single word \"up\" while at the same time forms and holds a clear left-hand open-palm pose with the palm facing the camera and the fingers together and extended. The speech is audible and easy to distinguish, with no extra words. The target is the pose, not the raising motion. After the final event, no new target action occurs." + }, + "created_at": "2026-11-15T01:16:31.933358Z", + "annotation_role": "full" + }, + { + "id": "60fae7bd3eab4ea3b90a37d01db25cc9", + "video_id": "down-then-open_palm_left__order__MOVA", + "user": "69eb6e157e23c6c0b2b7fb0d", + "annotations": { + "responses": { + "events": { + "speech": { + "a": { + "label": "FAIL", + "rationale": "Extra words are spoken" + }, + "v": { + "label": "PASS", + "rationale": "" + }, + "avb": { + "label": "PASS", + "rationale": "" + }, + "tcCount": { + "value": "1", + "rationale": "" + } + }, + "silent_action": { + "v": { + "label": "FAIL", + "rationale": "The person uses right hand" + }, + "tcCount": { + "value": "0", + "rationale": "" + } + } + }, + "video": { + "tcRel": { + "label": "FAIL", + "rationale": "Most of the requested action are not followed correctly" + } + } + }, + "axis_definitions": [ + { + "id": "a", + "key": "A", + "title": "Audio Event Correctness", + "granularity": "per-event", + "summary": "Whether the required speech or sound event is present and correct in the audio.", + "detail": "PASS when every required speech/sound component is present, clear, and correct. FAIL when a required audio event is missing, the content or sound type is wrong, or the audio is too unclear to identify. Judge audibility leniently (a soft or quiet sound still counts); judge identity leniently within an acoustic family (clap/slap/tap/thud/pat) and strictly across families." + }, + { + "id": "v", + "key": "V", + "title": "Visual Event Correctness", + "granularity": "per-event", + "summary": "Whether the required visual event is present and correct. For speech events this evaluates visible speaking behavior (mouth/lip motion), not the spoken content.", + "detail": "PASS when every required visual component is present, visible, and correct in actor, action, side, body part, object, and contact target. FAIL when it's missing; the wrong actor/side/body part/object/action is used; occlusion, cropping, or blur prevents verification; or (for speech) the mouth movement is absent or inconsistent with the required utterance." + }, + { + "id": "avb", + "key": "AVB", + "title": "Audio-Visual Binding", + "granularity": "per-event", + "summary": "Whether the audio and visual components within the same event are correctly synchronized and causally aligned.", + "detail": "Applies only to events with both audio and visual components (speech, sounding action — not silent action). PASS when the sound is synchronized with the corresponding visible action, impact, or mouth movement. FAILS if A or V fails, or the two are clearly misaligned in time. Temporal relations between different events are handled under TC_REL, not AVB." + }, + { + "id": "tcRel", + "key": "TC_REL", + "title": "Temporal Relation", + "granularity": "per-video", + "summary": "Whether the specified relationship between events holds - synchronization, order, and absolute timing.", + "detail": "Compare the defined start times of the required events. If any required event fails A, V, or AVB, TC_REL FAILS. Synchronization: PASS when events begin at the same moment (simple overlap is not enough). Order: PASS when events clearly begin in the required order (overlap allowed). Absolute timing: PASS when the event starts close to the required timestamp; +/-0.3s is a recommended guideline, not a strict cutoff - use judgment rather than a hard pass/fail line." + }, + { + "id": "tcCount", + "key": "TC_COUNT", + "title": "Temporal Count", + "granularity": "per-event", + "summary": "Whether each required event occurs the specified number of times, with no missing or extra occurrences.", + "detail": "Enter how many times the event actually occurs in the video. If no count is given, expect one. If the event fails A, V, or its applicable AVB, record its count as 0. A continuously held pose counts as one occurrence; releasing and re-forming it (or a clear distinct repetition without full release) counts as another." + } + ], + "events": [ + { + "id": "speech", + "modality": "speech", + "modalityLabel": "Speech", + "actor": "person_A", + "label": "Speech: say \"down\"", + "definition": "The person clearly says the single word \"down\". The speech is audible and easy to distinguish. No extra words are spoken unless explicitly required.", + "axes": { + "a": true, + "v": true, + "avb": true + } + }, + { + "id": "silent_action", + "modality": "silent_action", + "modalityLabel": "Silent action", + "actor": "person_A", + "label": "Silent action: raise open palm (left hand)", + "definition": "The person's left hand forms or holds a clear open-palm pose with the palm open and facing the camera, fingers together and extended. The target is the held pose, not a separate raising motion.", + "axes": { + "a": false, + "v": true, + "avb": false + } + } + ], + "temporal_constraint": { + "relationType": "order", + "timingType": "relative", + "value": { + "sequence": [ + "speech", + "silent_action" + ] + }, + "expectedCounts": { + "speech": 1, + "silent_action": 1 + }, + "description": "Order: Speech: say \"down\" -> then -> Silent action: raise open palm (left hand)." + }, + "model": "MOVA", + "item_id": "down-then-open_palm_left__order", + "canonical_id": "down_open_palm_left_rel_order", + "variant": "v1", + "scene_hint": "plain indoor room", + "camera_hint": "static medium shot", + "prompt": "In a plain indoor room, a seated person faces the camera in a static medium shot, with the full face, mouth, upper body, both hands, and thighs clearly visible. The person starts seated on a chair with both hands initially resting on the knees or thighs. First, the person clearly says the single word \"down\". After that speech ends, the person forms and holds a clear left-hand open-palm pose with the palm facing the camera and the fingers together and extended. The target is the pose, not the raising motion. No extra words or extra target sounds occur. After the final event, no new target action occurs." + }, + "created_at": "2026-11-15T01:17:24.875205Z", + "annotation_role": "full" + }, + { + "id": "5307c8980ecb4ed6821abe3d317299a4", + "video_id": "tree-with-peace_sign-with-tap_pen_left__sync_count_x2__MOVA", + "user": "69eb6e157e23c6c0b2b7fb0d", + "annotations": { + "responses": { + "events": { + "speech": { + "a": { + "label": "FAIL", + "rationale": "Extra words are spoken" + }, + "v": { + "label": "FAIL", + "rationale": "The lip motions are incorrect" + }, + "avb": { + "label": "FAIL", + "rationale": "" + }, + "tcCount": { + "value": "1", + "rationale": "" + } + }, + "silent_action": { + "v": { + "label": "PASS", + "rationale": "" + }, + "tcCount": { + "value": "1", + "rationale": "" + } + }, + "sounding_action": { + "a": { + "label": "FAIL", + "rationale": "No action done" + }, + "v": { + "label": "FAIL", + "rationale": "No action done" + }, + "avb": { + "label": "FAIL", + "rationale": "No action done" + }, + "tcCount": { + "value": "0", + "rationale": "No action done" + } + } + }, + "video": { + "tcRel": { + "label": "FAIL", + "rationale": "The requested action are not followed correctly" + } + } + }, + "axis_definitions": [ + { + "id": "a", + "key": "A", + "title": "Audio Event Correctness", + "granularity": "per-event", + "summary": "Whether the required speech or sound event is present and correct in the audio.", + "detail": "PASS when every required speech/sound component is present, clear, and correct. FAIL when a required audio event is missing, the content or sound type is wrong, or the audio is too unclear to identify. Judge audibility leniently (a soft or quiet sound still counts); judge identity leniently within an acoustic family (clap/slap/tap/thud/pat) and strictly across families." + }, + { + "id": "v", + "key": "V", + "title": "Visual Event Correctness", + "granularity": "per-event", + "summary": "Whether the required visual event is present and correct. For speech events this evaluates visible speaking behavior (mouth/lip motion), not the spoken content.", + "detail": "PASS when every required visual component is present, visible, and correct in actor, action, side, body part, object, and contact target. FAIL when it's missing; the wrong actor/side/body part/object/action is used; occlusion, cropping, or blur prevents verification; or (for speech) the mouth movement is absent or inconsistent with the required utterance." + }, + { + "id": "avb", + "key": "AVB", + "title": "Audio-Visual Binding", + "granularity": "per-event", + "summary": "Whether the audio and visual components within the same event are correctly synchronized and causally aligned.", + "detail": "Applies only to events with both audio and visual components (speech, sounding action — not silent action). PASS when the sound is synchronized with the corresponding visible action, impact, or mouth movement. FAILS if A or V fails, or the two are clearly misaligned in time. Temporal relations between different events are handled under TC_REL, not AVB." + }, + { + "id": "tcRel", + "key": "TC_REL", + "title": "Temporal Relation", + "granularity": "per-video", + "summary": "Whether the specified relationship between events holds - synchronization, order, and absolute timing.", + "detail": "Compare the defined start times of the required events. If any required event fails A, V, or AVB, TC_REL FAILS. Synchronization: PASS when events begin at the same moment (simple overlap is not enough). Order: PASS when events clearly begin in the required order (overlap allowed). Absolute timing: PASS when the event starts close to the required timestamp; +/-0.3s is a recommended guideline, not a strict cutoff - use judgment rather than a hard pass/fail line." + }, + { + "id": "tcCount", + "key": "TC_COUNT", + "title": "Temporal Count", + "granularity": "per-event", + "summary": "Whether each required event occurs the specified number of times, with no missing or extra occurrences.", + "detail": "Enter how many times the event actually occurs in the video. If no count is given, expect one. If the event fails A, V, or its applicable AVB, record its count as 0. A continuously held pose counts as one occurrence; releasing and re-forming it (or a clear distinct repetition without full release) counts as another." + } + ], + "events": [ + { + "id": "speech", + "modality": "speech", + "modalityLabel": "Speech", + "actor": "person_A", + "label": "Speech: say \"tree\"", + "definition": "The person clearly says the single word \"tree\". The speech is audible and easy to distinguish. No extra words are spoken unless explicitly required.", + "axes": { + "a": true, + "v": true, + "avb": true + } + }, + { + "id": "silent_action", + "modality": "silent_action", + "modalityLabel": "Silent action", + "actor": "person_A", + "label": "Silent action: peace sign (right hand)", + "definition": "The person's right hand forms and holds a clear peace sign: the index and middle fingers are extended and separated in a V shape while the remaining fingers are folded.", + "axes": { + "a": false, + "v": true, + "avb": false + } + }, + { + "id": "sounding_action", + "modality": "sounding_action", + "modalityLabel": "Sounding action", + "actor": "person_A", + "label": "Sounding action: tap pen (with left hand on left side table)", + "definition": "The person's left hand holds a pen and taps its tip against the small side-table surface once, with visible contact and a distinct tapping sound exactly at the moment of impact.", + "axes": { + "a": true, + "v": true, + "avb": true + } + } + ], + "temporal_constraint": { + "relationType": "sync_count", + "timingType": "relative", + "value": { + "events": [ + "speech", + "silent_action", + "sounding_action" + ], + "count": 2 + }, + "expectedCounts": { + "speech": 2, + "silent_action": 2, + "sounding_action": 2 + }, + "description": "Synchronization x2: Speech: say \"tree\" + Silent action: peace sign (right hand) + Sounding action: tap pen (with left hand on left side table) must occur together, repeated exactly 2 times." + }, + "model": "MOVA", + "item_id": "tree-with-peace_sign-with-tap_pen_left__sync_count_x2", + "canonical_id": "peace_sign_tap_pen_left_tree_rel_sync_count", + "variant": "v1", + "scene_hint": "plain indoor room", + "camera_hint": "static medium shot", + "prompt": "In a plain indoor room, a seated person faces the camera in a static medium shot, with the full face, mouth, upper body, both hands, and thighs clearly visible. A small side table stands beside the person's left hip, outside the frontal silhouette of both thighs; its top and legs do not obscure either thigh, and its contact surface and the acting hand remain in the frame. A pen lies on the side table from the start. The person starts seated on a chair with both hands initially resting on the knees or thighs. After the initial pose, the person picks up the pen with the left hand; picking it up is preparation, not a target event. The person performs the synchronized bundle exactly 2 times: each time, the person clearly says the single word \"tree\", forms and holds a clear right-hand peace sign, with the index and middle fingers extended in a V and the remaining fingers folded, and taps the pen tip against the small side-table surface once with the left hand, with visible contact and a distinct tapping sound exactly at the moment of impact. The synchronized repetitions are distinct, with a brief reset between them if needed. The speech is audible and easy to distinguish, with no extra words. The target is the pose, not the raising motion. No extra words or extra target sounds occur. After the final event, no new target action occurs." + }, + "created_at": "2026-11-15T01:19:09.960645Z", + "annotation_role": "full" + }, + { + "id": "419bb311236d46e9993f7baf6bd3b6f2", + "video_id": "house-then-thumbs_up_left-then-fist_pose__order__MOVA", + "user": "69eb6e157e23c6c0b2b7fb0d", + "annotations": { + "responses": { + "events": { + "speech": { + "a": { + "label": "FAIL", + "rationale": "Extra words are spoken" + }, + "v": { + "label": "PASS", + "rationale": "" + }, + "avb": { + "label": "PASS", + "rationale": "" + }, + "tcCount": { + "value": "1", + "rationale": "" + } + }, + "silent_action": { + "v": { + "label": "FAIL", + "rationale": "The person uses right hand instead" + }, + "tcCount": { + "value": "0", + "rationale": "" + } + }, + "silent_action_2": { + "v": { + "label": "FAIL", + "rationale": "No action done" + }, + "tcCount": { + "value": "0", + "rationale": "" + } + } + }, + "video": { + "tcRel": { + "label": "FAIL", + "rationale": "Most of the required action are done incorrectly" + } + } + }, + "axis_definitions": [ + { + "id": "a", + "key": "A", + "title": "Audio Event Correctness", + "granularity": "per-event", + "summary": "Whether the required speech or sound event is present and correct in the audio.", + "detail": "PASS when every required speech/sound component is present, clear, and correct. FAIL when a required audio event is missing, the content or sound type is wrong, or the audio is too unclear to identify. Judge audibility leniently (a soft or quiet sound still counts); judge identity leniently within an acoustic family (clap/slap/tap/thud/pat) and strictly across families." + }, + { + "id": "v", + "key": "V", + "title": "Visual Event Correctness", + "granularity": "per-event", + "summary": "Whether the required visual event is present and correct. For speech events this evaluates visible speaking behavior (mouth/lip motion), not the spoken content.", + "detail": "PASS when every required visual component is present, visible, and correct in actor, action, side, body part, object, and contact target. FAIL when it's missing; the wrong actor/side/body part/object/action is used; occlusion, cropping, or blur prevents verification; or (for speech) the mouth movement is absent or inconsistent with the required utterance." + }, + { + "id": "avb", + "key": "AVB", + "title": "Audio-Visual Binding", + "granularity": "per-event", + "summary": "Whether the audio and visual components within the same event are correctly synchronized and causally aligned.", + "detail": "Applies only to events with both audio and visual components (speech, sounding action — not silent action). PASS when the sound is synchronized with the corresponding visible action, impact, or mouth movement. FAILS if A or V fails, or the two are clearly misaligned in time. Temporal relations between different events are handled under TC_REL, not AVB." + }, + { + "id": "tcRel", + "key": "TC_REL", + "title": "Temporal Relation", + "granularity": "per-video", + "summary": "Whether the specified relationship between events holds - synchronization, order, and absolute timing.", + "detail": "Compare the defined start times of the required events. If any required event fails A, V, or AVB, TC_REL FAILS. Synchronization: PASS when events begin at the same moment (simple overlap is not enough). Order: PASS when events clearly begin in the required order (overlap allowed). Absolute timing: PASS when the event starts close to the required timestamp; +/-0.3s is a recommended guideline, not a strict cutoff - use judgment rather than a hard pass/fail line." + }, + { + "id": "tcCount", + "key": "TC_COUNT", + "title": "Temporal Count", + "granularity": "per-event", + "summary": "Whether each required event occurs the specified number of times, with no missing or extra occurrences.", + "detail": "Enter how many times the event actually occurs in the video. If no count is given, expect one. If the event fails A, V, or its applicable AVB, record its count as 0. A continuously held pose counts as one occurrence; releasing and re-forming it (or a clear distinct repetition without full release) counts as another." + } + ], + "events": [ + { + "id": "speech", + "modality": "speech", + "modalityLabel": "Speech", + "actor": "person_A", + "label": "Speech: say \"house\"", + "definition": "The person clearly says the single word \"house\". The speech is audible and easy to distinguish. No extra words are spoken unless explicitly required.", + "axes": { + "a": true, + "v": true, + "avb": true + } + }, + { + "id": "silent_action", + "modality": "silent_action", + "modalityLabel": "Silent action", + "actor": "person_A", + "label": "Silent action: thumbs up (left hand)", + "definition": "The person's left hand forms or holds a thumbs-up pose: a fist closed with the thumb extended upward, clearly recognizable as a thumbs-up. The target is the pose, not a separate upward hand-raising motion.", + "axes": { + "a": false, + "v": true, + "avb": false + } + }, + { + "id": "silent_action_2", + "modality": "silent_action", + "modalityLabel": "Silent action", + "actor": "person_A", + "label": "Silent action: fist pose (right hand)", + "definition": "The person's right hand holds a closed fist, with all fingers curled and the thumb resting outside the curled fingers rather than extended upward. The target is the pose, not the raising motion.", + "axes": { + "a": false, + "v": true, + "avb": false + } + } + ], + "temporal_constraint": { + "relationType": "order", + "timingType": "relative", + "value": { + "sequence": [ + "speech", + "silent_action", + "silent_action_2" + ] + }, + "expectedCounts": { + "speech": 1, + "silent_action": 1, + "silent_action_2": 1 + }, + "description": "Order: Speech: say \"house\" -> then -> Silent action: thumbs up (left hand) -> then -> Silent action: fist pose (right hand)." + }, + "model": "MOVA", + "item_id": "house-then-thumbs_up_left-then-fist_pose__order", + "canonical_id": "fist_pose_house_thumbs_up_left_rel_order", + "variant": "v1", + "scene_hint": "plain indoor room", + "camera_hint": "static medium shot", + "prompt": "In a plain indoor room, a seated person faces the camera in a static medium shot, with the full face, mouth, upper body, both hands, and thighs clearly visible. The person starts seated on a chair with both hands initially resting on the knees or thighs. First, the person clearly says the single word \"house\". After that speech ends, the person forms and holds a clear left-hand thumbs-up pose with the fist closed and the thumb extended upward. The target is the pose, not the raising motion. After that pose is visible, that hand returns to rest, then the person holds a closed fist with the right hand, all fingers curled and the thumb resting outside the curled fingers rather than extended upward. The target is the pose, not the raising motion. No extra words or extra target sounds occur. After the final event, no new target action occurs." + }, + "created_at": "2026-11-15T01:20:57.869470Z", + "annotation_role": "full" + }, + { + "id": "c9077f14128c40b79f9b6527ed5be652", + "video_id": "ok_sign-with-slap_left__sync_count_x3__MOVA", + "user": "69eb6e157e23c6c0b2b7fb0d", + "annotations": { + "responses": { + "events": { + "silent_action": { + "v": { + "label": "PASS", + "rationale": "" + }, + "tcCount": { + "value": "3", + "rationale": "" + } + }, + "sounding_action": { + "a": { + "label": "FAIL", + "rationale": "No action done using left hand" + }, + "v": { + "label": "FAIL", + "rationale": "No action done using left hand" + }, + "avb": { + "label": "FAIL", + "rationale": "No action done using left hand" + }, + "tcCount": { + "value": "0", + "rationale": "" + } + } + }, + "video": { + "tcRel": { + "label": "FAIL", + "rationale": "The requested action are not done correctly" + } + } + }, + "axis_definitions": [ + { + "id": "a", + "key": "A", + "title": "Audio Event Correctness", + "granularity": "per-event", + "summary": "Whether the required speech or sound event is present and correct in the audio.", + "detail": "PASS when every required speech/sound component is present, clear, and correct. FAIL when a required audio event is missing, the content or sound type is wrong, or the audio is too unclear to identify. Judge audibility leniently (a soft or quiet sound still counts); judge identity leniently within an acoustic family (clap/slap/tap/thud/pat) and strictly across families." + }, + { + "id": "v", + "key": "V", + "title": "Visual Event Correctness", + "granularity": "per-event", + "summary": "Whether the required visual event is present and correct. For speech events this evaluates visible speaking behavior (mouth/lip motion), not the spoken content.", + "detail": "PASS when every required visual component is present, visible, and correct in actor, action, side, body part, object, and contact target. FAIL when it's missing; the wrong actor/side/body part/object/action is used; occlusion, cropping, or blur prevents verification; or (for speech) the mouth movement is absent or inconsistent with the required utterance." + }, + { + "id": "avb", + "key": "AVB", + "title": "Audio-Visual Binding", + "granularity": "per-event", + "summary": "Whether the audio and visual components within the same event are correctly synchronized and causally aligned.", + "detail": "Applies only to events with both audio and visual components (speech, sounding action — not silent action). PASS when the sound is synchronized with the corresponding visible action, impact, or mouth movement. FAILS if A or V fails, or the two are clearly misaligned in time. Temporal relations between different events are handled under TC_REL, not AVB." + }, + { + "id": "tcRel", + "key": "TC_REL", + "title": "Temporal Relation", + "granularity": "per-video", + "summary": "Whether the specified relationship between events holds - synchronization, order, and absolute timing.", + "detail": "Compare the defined start times of the required events. If any required event fails A, V, or AVB, TC_REL FAILS. Synchronization: PASS when events begin at the same moment (simple overlap is not enough). Order: PASS when events clearly begin in the required order (overlap allowed). Absolute timing: PASS when the event starts close to the required timestamp; +/-0.3s is a recommended guideline, not a strict cutoff - use judgment rather than a hard pass/fail line." + }, + { + "id": "tcCount", + "key": "TC_COUNT", + "title": "Temporal Count", + "granularity": "per-event", + "summary": "Whether each required event occurs the specified number of times, with no missing or extra occurrences.", + "detail": "Enter how many times the event actually occurs in the video. If no count is given, expect one. If the event fails A, V, or its applicable AVB, record its count as 0. A continuously held pose counts as one occurrence; releasing and re-forming it (or a clear distinct repetition without full release) counts as another." + } + ], + "events": [ + { + "id": "silent_action", + "modality": "silent_action", + "modalityLabel": "Silent action", + "actor": "person_A", + "label": "Silent action: ok sign (right hand)", + "definition": "The person's right thumb and index fingertip touch to form a ring while the other three fingers remain extended and the palm faces the camera. The target is the held OK pose, not the raising motion.", + "axes": { + "a": false, + "v": true, + "avb": false + } + }, + { + "id": "sounding_action", + "modality": "sounding_action", + "modalityLabel": "Sounding action", + "actor": "person_A", + "label": "Sounding action: slapping the thigh (with left hand on left thigh)", + "definition": "The person's left hand strikes the top of the person's left thigh. Visible contact is made with the thigh, and a sharp slap sound occurs exactly at the moment of impact.", + "axes": { + "a": true, + "v": true, + "avb": true + } + } + ], + "temporal_constraint": { + "relationType": "sync_count", + "timingType": "relative", + "value": { + "events": [ + "silent_action", + "sounding_action" + ], + "count": 3 + }, + "expectedCounts": { + "silent_action": 3, + "sounding_action": 3 + }, + "description": "Synchronization x3: Silent action: ok sign (right hand) + Sounding action: slapping the thigh (with left hand on left thigh) must occur together, repeated exactly 3 times." + }, + "model": "MOVA", + "item_id": "ok_sign-with-slap_left__sync_count_x3", + "canonical_id": "ok_sign_slap_left_rel_sync_count", + "variant": "v1", + "scene_hint": "plain indoor room", + "camera_hint": "static medium shot", + "prompt": "In a plain indoor room, a seated person faces the camera in a static medium shot, with the full face, mouth, upper body, both hands, and thighs clearly visible. The person starts seated on a chair with both hands initially resting on the knees or thighs. The person performs the synchronized bundle exactly 3 times: each time, the person holds an OK sign with the right hand, thumb and index fingertip forming a ring and the other three fingers extended, palm facing the camera and strikes the top of the left thigh with the left hand once, making visible contact and a sharp slap sound exactly at the moment of impact. The synchronized repetitions are distinct, with a brief reset between them if needed. The target is the pose, not the raising motion. No extra words or extra target sounds occur. After the final event, no new target action occurs." + }, + "created_at": "2026-11-15T01:22:57.441477Z", + "annotation_role": "full" + }, + { + "id": "74af462e93b74f698481994aa0df2536", + "video_id": "salute_pose-then-ok_sign_left-then-clap__order__MOVA", + "user": "69eb6e157e23c6c0b2b7fb0d", + "annotations": { + "responses": { + "events": { + "silent_action": { + "v": { + "label": "PASS", + "rationale": "" + }, + "tcCount": { + "value": "1", + "rationale": "" + } + }, + "silent_action_2": { + "v": { + "label": "FAIL", + "rationale": "The person uses right hand" + }, + "tcCount": { + "value": "0", + "rationale": "" + } + }, + "sounding_action": { + "a": { + "label": "PASS", + "rationale": "" + }, + "v": { + "label": "PASS", + "rationale": "" + }, + "avb": { + "label": "PASS", + "rationale": "" + }, + "tcCount": { + "value": "5", + "rationale": "" + } + } + }, + "video": { + "tcRel": { + "label": "FAIL", + "rationale": "There are unwanted words spoken, and most of the action are done incorrectly" + } + } + }, + "axis_definitions": [ + { + "id": "a", + "key": "A", + "title": "Audio Event Correctness", + "granularity": "per-event", + "summary": "Whether the required speech or sound event is present and correct in the audio.", + "detail": "PASS when every required speech/sound component is present, clear, and correct. FAIL when a required audio event is missing, the content or sound type is wrong, or the audio is too unclear to identify. Judge audibility leniently (a soft or quiet sound still counts); judge identity leniently within an acoustic family (clap/slap/tap/thud/pat) and strictly across families." + }, + { + "id": "v", + "key": "V", + "title": "Visual Event Correctness", + "granularity": "per-event", + "summary": "Whether the required visual event is present and correct. For speech events this evaluates visible speaking behavior (mouth/lip motion), not the spoken content.", + "detail": "PASS when every required visual component is present, visible, and correct in actor, action, side, body part, object, and contact target. FAIL when it's missing; the wrong actor/side/body part/object/action is used; occlusion, cropping, or blur prevents verification; or (for speech) the mouth movement is absent or inconsistent with the required utterance." + }, + { + "id": "avb", + "key": "AVB", + "title": "Audio-Visual Binding", + "granularity": "per-event", + "summary": "Whether the audio and visual components within the same event are correctly synchronized and causally aligned.", + "detail": "Applies only to events with both audio and visual components (speech, sounding action — not silent action). PASS when the sound is synchronized with the corresponding visible action, impact, or mouth movement. FAILS if A or V fails, or the two are clearly misaligned in time. Temporal relations between different events are handled under TC_REL, not AVB." + }, + { + "id": "tcRel", + "key": "TC_REL", + "title": "Temporal Relation", + "granularity": "per-video", + "summary": "Whether the specified relationship between events holds - synchronization, order, and absolute timing.", + "detail": "Compare the defined start times of the required events. If any required event fails A, V, or AVB, TC_REL FAILS. Synchronization: PASS when events begin at the same moment (simple overlap is not enough). Order: PASS when events clearly begin in the required order (overlap allowed). Absolute timing: PASS when the event starts close to the required timestamp; +/-0.3s is a recommended guideline, not a strict cutoff - use judgment rather than a hard pass/fail line." + }, + { + "id": "tcCount", + "key": "TC_COUNT", + "title": "Temporal Count", + "granularity": "per-event", + "summary": "Whether each required event occurs the specified number of times, with no missing or extra occurrences.", + "detail": "Enter how many times the event actually occurs in the video. If no count is given, expect one. If the event fails A, V, or its applicable AVB, record its count as 0. A continuously held pose counts as one occurrence; releasing and re-forming it (or a clear distinct repetition without full release) counts as another." + } + ], + "events": [ + { + "id": "silent_action", + "modality": "silent_action", + "modalityLabel": "Silent action", + "actor": "person_A", + "label": "Silent action: salute pose (right hand)", + "definition": "The person's right hand holds a salute pose beside the right temple, with the fingers straight and together and the eyes and mouth visible. The target is the pose, not the raising motion.", + "axes": { + "a": false, + "v": true, + "avb": false + } + }, + { + "id": "silent_action_2", + "modality": "silent_action", + "modalityLabel": "Silent action", + "actor": "person_A", + "label": "Silent action: ok sign (left hand)", + "definition": "The person's left thumb and index fingertip touch to form a ring while the other three fingers remain extended and the palm faces the camera. The target is the held OK pose, not the raising motion.", + "axes": { + "a": false, + "v": true, + "avb": false + } + }, + { + "id": "sounding_action", + "modality": "sounding_action", + "modalityLabel": "Sounding action", + "actor": "person_A", + "label": "Sounding action: clapping hands", + "definition": "The person brings both hands together in front of the torso to clap, with visible hand-to-hand contact and a single sharp clap sound at the moment of impact.", + "axes": { + "a": true, + "v": true, + "avb": true + } + } + ], + "temporal_constraint": { + "relationType": "order", + "timingType": "relative", + "value": { + "sequence": [ + "silent_action", + "silent_action_2", + "sounding_action" + ] + }, + "expectedCounts": { + "silent_action": 1, + "silent_action_2": 1, + "sounding_action": 1 + }, + "description": "Order: Silent action: salute pose (right hand) -> then -> Silent action: ok sign (left hand) -> then -> Sounding action: clapping hands." + }, + "model": "MOVA", + "item_id": "salute_pose-then-ok_sign_left-then-clap__order", + "canonical_id": "clap_ok_sign_left_salute_pose_rel_order", + "variant": "v1", + "scene_hint": "plain indoor room", + "camera_hint": "static medium shot", + "prompt": "In a plain indoor room, a seated person faces the camera in a static medium shot, with the full face, mouth, upper body, both hands, and thighs clearly visible. The person starts seated on a chair with both hands initially resting on the knees or thighs. First, the person holds a salute pose with the right hand beside the right temple, fingers straight and together, with the eyes and mouth visible. The target is the pose, not the raising motion. After that pose is visible, that hand returns to rest, then the person holds an OK sign with the left hand, thumb and index fingertip forming a ring and the other three fingers extended, palm facing the camera. The target is the pose, not the raising motion. After that pose is visible, that hand returns to rest, then the person claps both hands together in front of the torso once, with visible contact and a sharp clap sound exactly at the moment of impact. No extra words or extra target sounds occur. After the final event, no new target action occurs." + }, + "created_at": "2026-11-15T01:24:51.344598Z", + "annotation_role": "full" + } + ], + "accepted_at": "2026-11-15T01:24:51.424856Z" +} \ No newline at end of file diff --git a/experiments/prolific_mova_1932/allocation_state.json b/experiments/prolific_mova_1932/allocation_state.json index 08fe80592f7c0f9f3223a922cb30e35e73e89316..25c96b51d11bae28073b3fde884c032dc034d3a8 100644 --- a/experiments/prolific_mova_1932/allocation_state.json +++ b/experiments/prolific_mova_1932/allocation_state.json @@ -2934,7 +2934,7 @@ "members": { "6af360ae762e3bbd25b599d7": "in_progress", "6ad15f742c074f06659e05ea": "in_progress", - "69eb6e157e23c6c0b2b7fb0d": "in_progress" + "69eb6e157e23c6c0b2b7fb0d": "completed" } } ], diff --git a/experiments/prolific_mova_1932/annotations/20261115T012451344598Z_salute_pose-then-ok_sign_left-then-clap__order__MOVA_69eb6e157e23c6c0b2b7fb0d_74af462e93b74f698481994aa0df2536.json b/experiments/prolific_mova_1932/annotations/20261115T012451344598Z_salute_pose-then-ok_sign_left-then-clap__order__MOVA_69eb6e157e23c6c0b2b7fb0d_74af462e93b74f698481994aa0df2536.json new file mode 100644 index 0000000000000000000000000000000000000000..ac2d25f38c8acffdd5595c07ddc05e26014b6c42 --- /dev/null +++ b/experiments/prolific_mova_1932/annotations/20261115T012451344598Z_salute_pose-then-ok_sign_left-then-clap__order__MOVA_69eb6e157e23c6c0b2b7fb0d_74af462e93b74f698481994aa0df2536.json @@ -0,0 +1,164 @@ +{ + "id": "74af462e93b74f698481994aa0df2536", + "video_id": "salute_pose-then-ok_sign_left-then-clap__order__MOVA", + "user": "69eb6e157e23c6c0b2b7fb0d", + "annotations": { + "responses": { + "events": { + "silent_action": { + "v": { + "label": "PASS", + "rationale": "" + }, + "tcCount": { + "value": "1", + "rationale": "" + } + }, + "silent_action_2": { + "v": { + "label": "FAIL", + "rationale": "The person uses right hand" + }, + "tcCount": { + "value": "0", + "rationale": "" + } + }, + "sounding_action": { + "a": { + "label": "PASS", + "rationale": "" + }, + "v": { + "label": "PASS", + "rationale": "" + }, + "avb": { + "label": "PASS", + "rationale": "" + }, + "tcCount": { + "value": "5", + "rationale": "" + } + } + }, + "video": { + "tcRel": { + "label": "FAIL", + "rationale": "There are unwanted words spoken, and most of the action are done incorrectly" + } + } + }, + "axis_definitions": [ + { + "id": "a", + "key": "A", + "title": "Audio Event Correctness", + "granularity": "per-event", + "summary": "Whether the required speech or sound event is present and correct in the audio.", + "detail": "PASS when every required speech/sound component is present, clear, and correct. FAIL when a required audio event is missing, the content or sound type is wrong, or the audio is too unclear to identify. Judge audibility leniently (a soft or quiet sound still counts); judge identity leniently within an acoustic family (clap/slap/tap/thud/pat) and strictly across families." + }, + { + "id": "v", + "key": "V", + "title": "Visual Event Correctness", + "granularity": "per-event", + "summary": "Whether the required visual event is present and correct. For speech events this evaluates visible speaking behavior (mouth/lip motion), not the spoken content.", + "detail": "PASS when every required visual component is present, visible, and correct in actor, action, side, body part, object, and contact target. FAIL when it's missing; the wrong actor/side/body part/object/action is used; occlusion, cropping, or blur prevents verification; or (for speech) the mouth movement is absent or inconsistent with the required utterance." + }, + { + "id": "avb", + "key": "AVB", + "title": "Audio-Visual Binding", + "granularity": "per-event", + "summary": "Whether the audio and visual components within the same event are correctly synchronized and causally aligned.", + "detail": "Applies only to events with both audio and visual components (speech, sounding action \u2014 not silent action). PASS when the sound is synchronized with the corresponding visible action, impact, or mouth movement. FAILS if A or V fails, or the two are clearly misaligned in time. Temporal relations between different events are handled under TC_REL, not AVB." + }, + { + "id": "tcRel", + "key": "TC_REL", + "title": "Temporal Relation", + "granularity": "per-video", + "summary": "Whether the specified relationship between events holds - synchronization, order, and absolute timing.", + "detail": "Compare the defined start times of the required events. If any required event fails A, V, or AVB, TC_REL FAILS. Synchronization: PASS when events begin at the same moment (simple overlap is not enough). Order: PASS when events clearly begin in the required order (overlap allowed). Absolute timing: PASS when the event starts close to the required timestamp; +/-0.3s is a recommended guideline, not a strict cutoff - use judgment rather than a hard pass/fail line." + }, + { + "id": "tcCount", + "key": "TC_COUNT", + "title": "Temporal Count", + "granularity": "per-event", + "summary": "Whether each required event occurs the specified number of times, with no missing or extra occurrences.", + "detail": "Enter how many times the event actually occurs in the video. If no count is given, expect one. If the event fails A, V, or its applicable AVB, record its count as 0. A continuously held pose counts as one occurrence; releasing and re-forming it (or a clear distinct repetition without full release) counts as another." + } + ], + "events": [ + { + "id": "silent_action", + "modality": "silent_action", + "modalityLabel": "Silent action", + "actor": "person_A", + "label": "Silent action: salute pose (right hand)", + "definition": "The person's right hand holds a salute pose beside the right temple, with the fingers straight and together and the eyes and mouth visible. The target is the pose, not the raising motion.", + "axes": { + "a": false, + "v": true, + "avb": false + } + }, + { + "id": "silent_action_2", + "modality": "silent_action", + "modalityLabel": "Silent action", + "actor": "person_A", + "label": "Silent action: ok sign (left hand)", + "definition": "The person's left thumb and index fingertip touch to form a ring while the other three fingers remain extended and the palm faces the camera. The target is the held OK pose, not the raising motion.", + "axes": { + "a": false, + "v": true, + "avb": false + } + }, + { + "id": "sounding_action", + "modality": "sounding_action", + "modalityLabel": "Sounding action", + "actor": "person_A", + "label": "Sounding action: clapping hands", + "definition": "The person brings both hands together in front of the torso to clap, with visible hand-to-hand contact and a single sharp clap sound at the moment of impact.", + "axes": { + "a": true, + "v": true, + "avb": true + } + } + ], + "temporal_constraint": { + "relationType": "order", + "timingType": "relative", + "value": { + "sequence": [ + "silent_action", + "silent_action_2", + "sounding_action" + ] + }, + "expectedCounts": { + "silent_action": 1, + "silent_action_2": 1, + "sounding_action": 1 + }, + "description": "Order: Silent action: salute pose (right hand) -> then -> Silent action: ok sign (left hand) -> then -> Sounding action: clapping hands." + }, + "model": "MOVA", + "item_id": "salute_pose-then-ok_sign_left-then-clap__order", + "canonical_id": "clap_ok_sign_left_salute_pose_rel_order", + "variant": "v1", + "scene_hint": "plain indoor room", + "camera_hint": "static medium shot", + "prompt": "In a plain indoor room, a seated person faces the camera in a static medium shot, with the full face, mouth, upper body, both hands, and thighs clearly visible. The person starts seated on a chair with both hands initially resting on the knees or thighs. First, the person holds a salute pose with the right hand beside the right temple, fingers straight and together, with the eyes and mouth visible. The target is the pose, not the raising motion. After that pose is visible, that hand returns to rest, then the person holds an OK sign with the left hand, thumb and index fingertip forming a ring and the other three fingers extended, palm facing the camera. The target is the pose, not the raising motion. After that pose is visible, that hand returns to rest, then the person claps both hands together in front of the torso once, with visible contact and a sharp clap sound exactly at the moment of impact. No extra words or extra target sounds occur. After the final event, no new target action occurs." + }, + "created_at": "2026-11-15T01:24:51.344598Z", + "annotation_role": "full" +} \ No newline at end of file diff --git a/experiments/prolific_mova_1932/completion_log.jsonl b/experiments/prolific_mova_1932/completion_log.jsonl index 3290f044d371ab67dddc180e1aff8f1fc61126eb..9b573ef49182f7ab8f613fdb9033f5562d51d979 100644 --- a/experiments/prolific_mova_1932/completion_log.jsonl +++ b/experiments/prolific_mova_1932/completion_log.jsonl @@ -7869,3 +7869,4 @@ {"id":"c9077f14128c40b79f9b6527ed5be652","video_id":"ok_sign-with-slap_left__sync_count_x3__MOVA","user":"69eb6e157e23c6c0b2b7fb0d","created_at":"2026-11-15T01:22:57.441477Z","annotation_path":"experiments/prolific_mova_1932/annotations/20261115T012257441477Z_ok_sign-with-slap_left__sync_count_x3__MOVA_69eb6e157e23c6c0b2b7fb0d_c9077f14128c40b79f9b6527ed5be652.json"} {"id":"5b42b4cb53ba45aca80908e8cba94986","video_id":"ok_sign-with-slap_left__sync_count_x3__MOVA","user":"6af360ae762e3bbd25b599d7","created_at":"2026-11-15T01:24:05.605289Z","annotation_path":"experiments/prolific_mova_1932/annotations/20261115T012405605289Z_ok_sign-with-slap_left__sync_count_x3__MOVA_6af360ae762e3bbd25b599d7_5b42b4cb53ba45aca80908e8cba94986.json"} {"id":"c66ca525a3ce49d5be23e3a92c4209b4","video_id":"house-then-thumbs_up_left-then-fist_pose__order__MOVA","user":"6ad15f742c074f06659e05ea","created_at":"2026-11-15T01:24:14.306770Z","annotation_path":"experiments/prolific_mova_1932/annotations/20261115T012414306770Z_house-then-thumbs_up_left-then-fist_pose__order__MOVA_6ad15f742c074f06659e05ea_c66ca525a3ce49d5be23e3a92c4209b4.json"} +{"id":"74af462e93b74f698481994aa0df2536","video_id":"salute_pose-then-ok_sign_left-then-clap__order__MOVA","user":"69eb6e157e23c6c0b2b7fb0d","created_at":"2026-11-15T01:24:51.344598Z","annotation_path":"experiments/prolific_mova_1932/annotations/20261115T012451344598Z_salute_pose-then-ok_sign_left-then-clap__order__MOVA_69eb6e157e23c6c0b2b7fb0d_74af462e93b74f698481994aa0df2536.json"} diff --git a/experiments/prolific_mova_1932/quality_control_status.json b/experiments/prolific_mova_1932/quality_control_status.json index 441e38bcabc1a36ac0000d48ef64d36464d63bfa..2bc7326894fea8876897a3501a57b7bec9838e8e 100644 --- a/experiments/prolific_mova_1932/quality_control_status.json +++ b/experiments/prolific_mova_1932/quality_control_status.json @@ -1445,13 +1445,13 @@ "69eb6e157e23c6c0b2b7fb0d": { "status": "pass", "quality_passed": true, - "assignment_completed": false, - "analysis_eligible": false, + "assignment_completed": true, + "analysis_eligible": true, "accuracy": 0.8, "correct_fields": 20, "total_fields": 25, "threshold": 0.7, - "updated_at": "2026-11-15T01:14:52.144794Z" + "updated_at": "2026-11-15T01:24:51.424856Z" }, "6ad15f742c074f06659e05ea": { "status": "pass", diff --git a/experiments/prolific_mova_1932/users/69eb6e157e23c6c0b2b7fb0d.json b/experiments/prolific_mova_1932/users/69eb6e157e23c6c0b2b7fb0d.json index 1a4567dac02a985b0955aad6ed8cd2fe8620ce2a..0f235f4df4182233e26badfeb2d6a5a42317a0ca 100644 --- a/experiments/prolific_mova_1932/users/69eb6e157e23c6c0b2b7fb0d.json +++ b/experiments/prolific_mova_1932/users/69eb6e157e23c6c0b2b7fb0d.json @@ -96,6 +96,7 @@ "okay-with-point_pose_left-with-peace_sign__sync__MOVA", "okay_x2-then-ok_sign_x1-then-snap_x2__order_count__MOVA", "open_palm_x2-then-stop_x1__order_count__MOVA", + "salute_pose-then-ok_sign_left-then-clap__order__MOVA", "slap-then-help-then-thumbs_up__order__MOVA", "thumbs_up-then-slap__order__MOVA", "tree-then-thumbs_down_left-then-snap_left__order__MOVA", @@ -232,11 +233,16 @@ "id": "c9077f14128c40b79f9b6527ed5be652", "created_at": "2026-11-15T01:22:57.441477Z", "annotation_path": "experiments/prolific_mova_1932/annotations/20261115T012257441477Z_ok_sign-with-slap_left__sync_count_x3__MOVA_69eb6e157e23c6c0b2b7fb0d_c9077f14128c40b79f9b6527ed5be652.json" + }, + "salute_pose-then-ok_sign_left-then-clap__order__MOVA": { + "id": "74af462e93b74f698481994aa0df2536", + "created_at": "2026-11-15T01:24:51.344598Z", + "annotation_path": "experiments/prolific_mova_1932/annotations/20261115T012451344598Z_salute_pose-then-ok_sign_left-then-clap__order__MOVA_69eb6e157e23c6c0b2b7fb0d_74af462e93b74f698481994aa0df2536.json" } }, - "status": "in_progress", - "completion_code": null, - "completed_at": null, + "status": "completed", + "completion_code": "T2AV-TGM9NNVATRII", + "completed_at": "2026-11-15T01:24:51.424856Z", "quality_control": { "scores": { "clap-then-peace_sign-then-help__order__MOVA": { @@ -271,8 +277,8 @@ "accuracy": 0.8, "threshold": 0.7, "quality_passed": true, - "analysis_eligible": false, - "assignment_completed": false, + "analysis_eligible": true, + "assignment_completed": true, "completed_at": "2026-11-15T01:14:52.144794Z" } } \ No newline at end of file