Add Track 2 whole-body reconstruction dataset

.gitattributes

@@ -98,11 +98,3 @@ track_2/mesh/watering_can/watering_can.glb filter=lfs diff=lfs merge=lfs -text
 track_2/mesh/wet_floor_sign/wet_floor_sign.glb filter=lfs diff=lfs merge=lfs -text
 track_2/mesh/white_desk/white_desk.glb filter=lfs diff=lfs merge=lfs -text
 track_2/mesh/woven_basket/woven_basket.glb filter=lfs diff=lfs merge=lfs -text
-track_3/mesh/wooden_piece_1/wooden_piece_1_collision.glb filter=lfs diff=lfs merge=lfs -text
-track_3/mesh/wooden_piece_1/wooden_piece_1_visual.glb filter=lfs diff=lfs merge=lfs -text
-track_3/mesh/wooden_piece_2/wooden_piece_2_collision.glb filter=lfs diff=lfs merge=lfs -text
-track_3/mesh/wooden_piece_2/wooden_piece_2_visual.glb filter=lfs diff=lfs merge=lfs -text
-track_3/mesh/mini_sweeper/mini_sweeper_collision.glb filter=lfs diff=lfs merge=lfs -text
-track_3/mesh/mini_sweeper/mini_sweeper_visual.glb filter=lfs diff=lfs merge=lfs -text
-*.glb filter=lfs diff=lfs merge=lfs -text
-track_1/meta/episodes_stats.jsonl filter=lfs diff=lfs merge=lfs -text

README.md

@@ -1,120 +1,3 @@
 ---
-license: cc-by-4.0
-pretty_name: Video to Data (V2D) Challenge Dataset
-tags:
-  - robotics
-  - robot-learning
-  - manipulation
-  - human-object-interaction
-  - motion-capture
-  - video
-  - 4d-reconstruction
-  - reinforcement-learning
-  - egocentric-video
+license: apache-2.0
 ---
-
-# Video to Data (V2D) Challenge Dataset
-
-## Dataset Description
-
-The **Video to Data (V2D) Challenge Dataset** is an NVIDIA-developed benchmark for studying the complete path from human demonstration video to physics-grounded robot behavior. It supports three coupled challenge tracks over shared manipulation tasks: **4D human-object interaction reconstruction**, **robotic grounding**, and **end-to-end egocentric transfer**.
-
-- [Challenge website](https://nvidia-isaac.github.io/video_to_data/v2d_challenge/)
-- [Starter toolkit](https://github.com/nvidia-isaac/video_to_data)
-- [Dataset repository](https://huggingface.co/datasets/nvidia/video_to_data_challenge)
-- Contact: [v2d_challenge@nvidia.com](mailto:v2d_challenge@nvidia.com)
-
-## Challenge Tracks
-
-| Track | Input | Goal | Evaluation summary |
-| --- | --- | --- | --- |
-| **Track 1: Reconstruction** | Monocular third-person RGB video | Recover the human, object pose, and object geometry as a metric 4D human-object interaction scene in a consistent world frame | Reconstruction accuracy and physical plausibility relative to the multi-view reference |
-| **Track 2: Robotic Grounding** | Third-person 4D human-object interaction trajectories at different input-noise tiers | Retarget the demonstration and learn an executable policy for the robotic embodiment and simulator | Object-tracking performance at each input tier |
-| **Track 3: Egocentric** | Egocentric human demonstration video | Produce an executable robot policy through either an explicit reconstruct-and-retarget pipeline or an implicit end-to-end method | Final results produced with the official evaluation script |
-
-### Track 1: Reconstruction
-
-Track 1 evaluates monocular 4D human-object interaction reconstruction under challenging conditions including occlusion, bimanual coordination, and long-horizon manipulation.
-
-Participants reconstruct:
-
-- the human body and hands;
-- object pose trajectories;
-- object geometry; and
-- metric scale.
-
-Track 1 is evaluated along two equally weighted axes:
-
-1. **Accuracy**
-   - Chamfer distance to the multi-view human mesh
-   - Chamfer distance to the multi-view object mesh
-2. **Physical plausibility**
-   - Human-joint acceleration error
-   - Object acceleration error
-   - Contact penetration error
-
-### Track 2: Robotic Grounding
-
-Track 2 measures how upstream reconstruction quality affects human-to-robot transfer and downstream policy learning.
-
-The dataset provides three input tiers:
-
-1. **Tier 1 — Clean multi-view capture:** an upper-bound input for upstream reconstruction.
-2. **Tier 2 — Synthetic corruption:** trajectories with jitter, dropout, and contact errors sampled from Track 1 error distributions.
-3. **Tier 3 — Off-the-shelf reconstruction:** trajectories produced by current reconstruction methods.
-
-Each tier is scored separately. Metrics include **AUC**, **SP-SR**, **MP-SR**, and **MPPE**, as defined by the challenge evaluation protocol.
-
-### Track 3: Egocentric
-
-Track 3 evaluates the full pipeline from egocentric human video to robot behavior. The track is method-agnostic: participants may use an explicit reconstruction-and-retargeting pipeline, an end-to-end model, a pretrained vision-language-action model, a world-action model, or a hybrid approach.
-
-The NVIDIA-collected data includes human-object manipulation recordings and associated assets made available for development and evaluation. Depending on the released split, these assets may include:
-
-- egocentric videos;
-- motion-capture trajectories;
-- sequence metadata;
-- textured 3D object meshes; and
-- URDF object descriptions.
-
-Use `eval_e2e.py` from the starter toolkit to package the required reconstructions and recorded policy evaluations.
-
-## Download
-
-Install the Hugging Face Hub client:
-
-```bash
-python -m pip install -U "huggingface_hub"
-```
-
-Download the complete dataset repository while preserving its file structure:
-
-```bash
-hf download nvidia/video_to_data_challenge \
-  --repo-type dataset \
-  --local-dir ./video_to_data_challenge
-```
-
-If authentication is requested, first run:
-
-```bash
-hf auth login
-```
-
-The same operation can be performed from Python:
-
-```python
-from huggingface_hub import snapshot_download
-
-snapshot_download(
-    repo_id="nvidia/video_to_data_challenge",
-    repo_type="dataset",
-    local_dir="./video_to_data_challenge",
-)
-```
-
-Large assets are stored using Hugging Face's large-file infrastructure. Make sure sufficient disk space is available before downloading the complete repository.
-
-## Support
-
-For challenge or dataset questions, contact [v2d_challenge@nvidia.com](mailto:v2d_challenge@nvidia.com). For software issues, use the issue tracker in the [Video to Data repository](https://github.com/nvidia-isaac/video_to_data/issues).

track_1/.gitignore

@@ -1,2 +0,0 @@
-__pycache__/
-.DS_Store

track_1/README.md

@@ -1,79 +0,0 @@
-# Track 1
-
-Video-only single-view human-object interaction sequences for object tracking. Track 1
-contains 30 selected episodes from the FORM-HOI multiview recordings. Each episode provides
-one static third-person RGB video, its physical camera view, the target object identifier
-and prompt, and the original action description.
-
-| | |
-|---|---|
-| Episodes | 30 |
-| Frames | 16,563 (360–877 per episode, mean 552) |
-| Rate | 30 fps |
-| Tasks | 22 original action descriptions |
-| Cameras | 1 stream per episode, 30 clips, 4 distinct physical cameras |
-| Target objects | 10 |
-
-## Layout
-
-```
-track_1/
-├── data/chunk-000/episode_0000NN.parquet    LeRobot frame indexes only
-├── videos/chunk-000/observation.images.exo_camera/episode_0000NN.mp4
-└── meta/
-    ├── info.json                            LeRobot v2.1 schema and counts
-    ├── episodes.jsonl                       episode lengths and action descriptions
-    ├── episodes_metadata.jsonl              sequence, camera, and target-object metadata
-    ├── episodes_stats.jsonl                 video and index statistics
-    └── tasks.jsonl                          original action descriptions
-```
-
-## Per-episode metadata
-
-Each record in `meta/episodes_metadata.jsonl` contains only:
-
-- `episode_index`
-- `sequence_id`
-- `camera`
-- `object`
-- `object_prompt`
-- `video_key`
-
-The video path is determined from `meta/info.json` using `episode_index` and `video_key`.
-The `object` and `object_prompt` fields identify the object to track. The corresponding
-action description is resolved through the episode's `task_index` and
-`meta/tasks.jsonl`.
-
-## Video
-
-Videos are 1536×1152 H.264 (`yuv420p`) at 30 fps with no audio. They are byte-identical
-copies of the selected source videos and were not re-encoded.
-
-| Camera | Episodes |
-|---|---:|
-| `back_stereo_camera_left` | 5 |
-| `front_stereo_camera_left` | 9 |
-| `left_stereo_camera_left` | 11 |
-| `right_stereo_camera_left` | 5 |
-
-## LeRobot v2.1 indexes
-
-The per-episode Parquet files contain only the structural columns required to align video
-frames with LeRobot episode and task metadata:
-
-- `timestamp`
-- `frame_index`
-- `episode_index`
-- `index`
-- `task_index`
-- `next.done`
-
-No human pose, body reconstruction, object pose, visibility annotation, or object mesh is
-included in the current revision.
-
-Episode indexes follow the 30-sequence selection order. Each of the 10 target objects has three episodes.
-
-RGB statistics for retained videos are preserved. For new videos, channel statistics use
-uniformly sampled frames (sample count `min(N, max(100, min(10000, floor(N**0.75))))`),
-area-downsampled to 154×116 and normalized to [0, 1]. This downsampling is used only
-for statistics; the distributed videos retain their original bytes and resolution.

track_1/data/chunk-000/episode_000011.parquet

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:3b14398e694390c4d91da3e8afb5c552aff9d414ff853860db5fe4c8f722e072
-size 15797

track_1/data/chunk-000/episode_000013.parquet

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:7cfe25d7461fb478eaab646c3c09a77e9757786a612e7f03c9b53001e2ebef4e
-size 8516

track_1/data/chunk-000/episode_000018.parquet

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:14024fa8c62c39581a88250f83a9e63a59ca5137e0ab08528f5cf52ab527ea56
-size 10395

track_1/data/chunk-000/episode_000021.parquet

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:f407f21a5b83b6010bd683f7855245f729b7ae162bf0472bd3b488e29f4646ef
-size 10856

track_1/data/chunk-000/episode_000022.parquet

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:45a0b91861e3c1bad4e26573970412d4b72e31eeb9303e4a9218cf663bb332d5
-size 10371

track_1/data/chunk-000/episode_000023.parquet

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:fe1ad175219113990efe58e742b284befef88699f60876a35fa722c5bd173637
-size 10660

track_1/data/chunk-000/episode_000024.parquet

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:ce406755d4bfab45eebc08540b9f4c8a81037e8b13edcd2c473b93f8059e0725
-size 8426

track_1/data/chunk-000/episode_000026.parquet

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:7ca62e56805e5dd6312babcbec4edcf5bd53677ae0929d3cdc53bd23c11d7fbb
-size 8093

track_1/data/chunk-000/episode_000027.parquet

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:960f4e33745469bf806411d9a556d432eb8923b6530de25e3d4e63390306ebbc
-size 8720

track_1/meta/episodes.jsonl

@@ -1,30 +0,0 @@
-{"episode_index": 0, "length": 790, "tasks": ["Walk over to the hula hoop on the floor and step inside the hula hoop. Lift the hula hoop over your entire body and place back on the floor when done."]}
-{"episode_index": 1, "length": 668, "tasks": ["Walk over to the hula hoop on the floor and grasp it with your right hand keeping the hula hoop vertical. Pass your right foot  through the hula hoop and then your head and torso while shifting the hula hoop from your right hand to the left hand. Finally step through the hula hoop using your left leg. Place the hula hoop on the floor."]}
-{"episode_index": 2, "length": 866, "tasks": ["Walk over to the hula hoop on the floor and pick up the hula hoop using both hands. Jump through the hula hoop as if it was a jump rope 3x."]}
-{"episode_index": 3, "length": 592, "tasks": ["Walk over to the big red bowl on the round table and flip the big red bowl onto the round table. Step away when complete."]}
-{"episode_index": 4, "length": 747, "tasks": ["Walk over to the big red bowl on the round table and push the red bowl away you using both hands as you walk around the round table. Step away when complete."]}
-{"episode_index": 5, "length": 668, "tasks": ["Walk over to the big red bowl on the floor and pick up the red bowl using your right hand and then handover to your left hand and then place the big red bowl on the skinny wood chair. Step away when complete."]}
-{"episode_index": 6, "length": 816, "tasks": ["Walk over to the iron on the round table and pick it up with your right hand stand it up vertically. Pick up the iron with your left hand and set it down horizontally on the round table."]}
-{"episode_index": 7, "length": 777, "tasks": ["Walk over to the iron on the floor and pick up the iron using both hands and rotate the iron as you walk within the bounding area and place it on the ground, step away when complete."]}
-{"episode_index": 8, "length": 634, "tasks": ["Walk over to the iron on the ground and kneel near the iron and pick up the iron using your right hand and proceed to iron the ground infront of you. Place the iron on the ground when complete and stand up."]}
-{"episode_index": 9, "length": 415, "tasks": ["Walk to the white desk that is lying on its side. Place both hands on it and use them to carefully flip the desk onto the other side."]}
-{"episode_index": 10, "length": 577, "tasks": ["Walk to the white desk, use your left foot to push the desk."]}
-{"episode_index": 11, "length": 877, "tasks": ["Walk to the white desk, use both your hands to grasp on the edge of the table closest to you, and lower the desk to the ground."]}
-{"episode_index": 12, "length": 405, "tasks": ["Pick up the pan from the table using one hand. Take a few steps away from the table. Using your other hand, touch the bottom of the pan with your palm, and make some circular motions to wipe it. Walk back to the table and put the pan down."]}
-{"episode_index": 13, "length": 425, "tasks": ["Pick up the pan from the table using one hand. Take a few steps away from the table. Using your other hand, touch the bottom of the pan with your palm, and make some circular motions to wipe it. Walk back to the table and put the pan down."]}
-{"episode_index": 14, "length": 442, "tasks": ["Pick up the pan from the table using one hand. Take a few steps away from the table. Using your other hand, touch the bottom of the pan with your palm, and make some circular motions to wipe it. Walk back to the table and put the pan down."]}
-{"episode_index": 15, "length": 501, "tasks": ["Pick up the block from the table with both hands. Hold it by two opposite corners. Slowly rotate it several rotations while maintaining contact. Put it back on the table."]}
-{"episode_index": 16, "length": 360, "tasks": ["Pick up the block from the table with both hands. Hold it by two opposite corners. Slowly rotate it several rotations while maintaining contact. Put it back on the table."]}
-{"episode_index": 17, "length": 419, "tasks": ["Pick up the block from the table with both hands. Shift to holding it with one hand from the bottom. take a few steps back and face a different direction. Hold it with both hands again, and put it on the ground."]}
-{"episode_index": 18, "length": 535, "tasks": ["Pick up the roller on the table. Slowly roll the roller back and forth on the table surface a few times while moving around the table. Lay the roller back on the table and step back."]}
-{"episode_index": 19, "length": 443, "tasks": ["Pick up the roller on the table. Slowly roll the roller back and forth on the table surface a few times while moving around the table. Lay the roller back on the table and step back."]}
-{"episode_index": 20, "length": 549, "tasks": ["Pick up the roller on the table. Slowly roll the roller back and forth on the table surface a few times while moving around the table. Lay the roller back on the table and step back."]}
-{"episode_index": 21, "length": 563, "tasks": ["Walk over to the pink foam roll, and pick it up from the ground. Hold it from one end using one hand. Touch the other end of the roll on the ground a few times. Switch to the other hand and do the same. Put the roll back on the ground."]}
-{"episode_index": 22, "length": 533, "tasks": ["Pick up the foam roll on the table. Use your hands to slowly roll the roller back and forth on the table surface a few times while walking along with it. Lay the roller back on the table and step back."]}
-{"episode_index": 23, "length": 552, "tasks": ["Walk over to the pink foam roll, and pick it up from the ground. Hold it from one end using one hand. Touch the other end of the roll on the ground a few times. Switch to the other hand and do the same. Put the roll back on the ground."]}
-{"episode_index": 24, "length": 420, "tasks": ["Walk up to the wooden stool and sit on the stool. Rotate the direction you're facing while remaining seated. Stand up."]}
-{"episode_index": 25, "length": 365, "tasks": ["Pick up the wooden stool from the ground using both hands. Let go of one hand and take a few steps. hold the stool with both hands again and place it back on the ground."]}
-{"episode_index": 26, "length": 399, "tasks": ["Walk up to the wooden stool and sit on the stool. Rotate the direction you're facing while remaining seated. Stand up."]}
-{"episode_index": 27, "length": 440, "tasks": ["Use both hands to pick up the white laptop cart, then lay it down sideways on the ground. Release it and stand up."]}
-{"episode_index": 28, "length": 366, "tasks": ["Use both hands to pick up the white laptop cart, then lay it down sideways on the ground. Release it and stand up."]}
-{"episode_index": 29, "length": 419, "tasks": ["Step up to the white laptop cart, and use one hand to push it forward a few steps. Put both hands on the cart, and pull it side ways a few steps. Release the cart and step away."]}

track_1/meta/episodes_metadata.jsonl

@@ -1,30 +0,0 @@
-{"episode_index": 0, "sequence_id": "2026-03-16_15-04-26_hula_hoop_inside_ground_01", "camera": "back_stereo_camera_left", "object": "hula_hoop", "object_prompt": "a shiny skinny hula hoop.", "video_key": "observation.images.exo_camera"}
-{"episode_index": 1, "sequence_id": "2026-03-16_15-07-35_hula_hoop_through_01", "camera": "back_stereo_camera_left", "object": "hula_hoop", "object_prompt": "a shiny skinny hula hoop.", "video_key": "observation.images.exo_camera"}
-{"episode_index": 2, "sequence_id": "2026-04-25_14-46-14_hula_hoop_jump_rope_06", "camera": "back_stereo_camera_left", "object": "hula_hoop", "object_prompt": "a shiny skinny hula hoop.", "video_key": "observation.images.exo_camera"}
-{"episode_index": 3, "sequence_id": "2026-06-03_18-26-12_big_red_bowl_flip_03", "camera": "left_stereo_camera_left", "object": "big_red_bowl", "object_prompt": "A red plastic bowl.", "video_key": "observation.images.exo_camera"}
-{"episode_index": 4, "sequence_id": "2026-06-08_15-30-20_big_red_bowl_push_05", "camera": "front_stereo_camera_left", "object": "big_red_bowl", "object_prompt": "A red plastic bowl.", "video_key": "observation.images.exo_camera"}
-{"episode_index": 5, "sequence_id": "2026-06-09_11-19-34_big_red_bowl_skinny_wood_chair_05", "camera": "front_stereo_camera_left", "object": "big_red_bowl", "object_prompt": "A red plastic bowl.", "video_key": "observation.images.exo_camera"}
-{"episode_index": 6, "sequence_id": "2026-05-29_15-24-23_iron_pick_place_right_left_hand_02", "camera": "left_stereo_camera_left", "object": "iron", "object_prompt": "An metallic iron with blue accents.", "video_key": "observation.images.exo_camera"}
-{"episode_index": 7, "sequence_id": "2026-06-08_15-54-40_iron_rotate_mid_air_04", "camera": "left_stereo_camera_left", "object": "iron", "object_prompt": "An metallic iron with blue accents.", "video_key": "observation.images.exo_camera"}
-{"episode_index": 8, "sequence_id": "2026-06-09_11-34-07_iron_ground_05", "camera": "right_stereo_camera_left", "object": "iron", "object_prompt": "An metallic iron with blue accents.", "video_key": "observation.images.exo_camera"}
-{"episode_index": 9, "sequence_id": "2026-03-12_16-43-03_white_desk_side_to_side_01", "camera": "front_stereo_camera_left", "object": "white_desk", "object_prompt": "a medium sized white desk.", "video_key": "observation.images.exo_camera"}
-{"episode_index": 10, "sequence_id": "2026-03-13_11-18-20_white_desk_left_foot_push_04", "camera": "left_stereo_camera_left", "object": "white_desk", "object_prompt": "a medium sized white desk.", "video_key": "observation.images.exo_camera"}
-{"episode_index": 11, "sequence_id": "2026-04-29_14-21-44_white_desk_lower_to_ground_09", "camera": "left_stereo_camera_left", "object": "white_desk", "object_prompt": "a medium sized white desk.", "video_key": "observation.images.exo_camera"}
-{"episode_index": 12, "sequence_id": "2026-09-11_15-27-19_black_pan_wipe_06", "camera": "front_stereo_camera_left", "object": "black_pan", "object_prompt": "a black frying pan.", "video_key": "observation.images.exo_camera"}
-{"episode_index": 13, "sequence_id": "2026-09-11_15-28-23_black_pan_wipe_07", "camera": "left_stereo_camera_left", "object": "black_pan", "object_prompt": "a black frying pan.", "video_key": "observation.images.exo_camera"}
-{"episode_index": 14, "sequence_id": "2026-09-11_15-55-29_black_pan_wipe_08", "camera": "front_stereo_camera_left", "object": "black_pan", "object_prompt": "a black frying pan.", "video_key": "observation.images.exo_camera"}
-{"episode_index": 15, "sequence_id": "2026-09-10_16-14-06_foam_grass_block_rotate_02", "camera": "right_stereo_camera_left", "object": "foam_grass_block", "object_prompt": "a foam block with a pixalated design of various shades of green and various shades of brown.", "video_key": "observation.images.exo_camera"}
-{"episode_index": 16, "sequence_id": "2026-09-11_15-12-36_foam_grass_block_rotate_04", "camera": "front_stereo_camera_left", "object": "foam_grass_block", "object_prompt": "a foam block with a pixalated design of various shades of green and various shades of brown.", "video_key": "observation.images.exo_camera"}
-{"episode_index": 17, "sequence_id": "2026-09-11_15-53-19_foam_grass_block_pick_place_05", "camera": "left_stereo_camera_left", "object": "foam_grass_block", "object_prompt": "a foam block with a pixalated design of various shades of green and various shades of brown.", "video_key": "observation.images.exo_camera"}
-{"episode_index": 18, "sequence_id": "2026-09-10_15-45-17_paint_roller_table_roll_03", "camera": "back_stereo_camera_left", "object": "paint_roller", "object_prompt": "white with yellow stipped paint roller with a blue handle.", "video_key": "observation.images.exo_camera"}
-{"episode_index": 19, "sequence_id": "2026-09-11_15-00-57_paint_roller_table_roll_04", "camera": "left_stereo_camera_left", "object": "paint_roller", "object_prompt": "white with yellow stipped paint roller with a blue handle.", "video_key": "observation.images.exo_camera"}
-{"episode_index": 20, "sequence_id": "2026-09-11_15-45-25_paint_roller_table_roll_05", "camera": "left_stereo_camera_left", "object": "paint_roller", "object_prompt": "white with yellow stipped paint roller with a blue handle.", "video_key": "observation.images.exo_camera"}
-{"episode_index": 21, "sequence_id": "2026-09-10_15-52-02_pink_foam_roll_tap_03", "camera": "left_stereo_camera_left", "object": "pink_foam_roll", "object_prompt": "pink foam roll.", "video_key": "observation.images.exo_camera"}
-{"episode_index": 22, "sequence_id": "2026-09-10_15-57-50_pink_foam_roll_table_001", "camera": "front_stereo_camera_left", "object": "pink_foam_roll", "object_prompt": "pink foam roll.", "video_key": "observation.images.exo_camera"}
-{"episode_index": 23, "sequence_id": "2026-09-11_15-04-48_pink_foam_roll_tap_04", "camera": "front_stereo_camera_left", "object": "pink_foam_roll", "object_prompt": "pink foam roll.", "video_key": "observation.images.exo_camera"}
-{"episode_index": 24, "sequence_id": "2026-09-10_16-02-58_short_wood_stool_sit_rotate_001", "camera": "front_stereo_camera_left", "object": "short_wood_stool", "object_prompt": "short wooden stool with no back support.", "video_key": "observation.images.exo_camera"}
-{"episode_index": 25, "sequence_id": "2026-09-10_16-08-39_short_wood_stool_pick_place_02", "camera": "right_stereo_camera_left", "object": "short_wood_stool", "object_prompt": "short wooden stool with no back support.", "video_key": "observation.images.exo_camera"}
-{"episode_index": 26, "sequence_id": "2026-09-11_15-49-16_short_wood_stool_sit_rotate_05", "camera": "right_stereo_camera_left", "object": "short_wood_stool", "object_prompt": "short wooden stool with no back support.", "video_key": "observation.images.exo_camera"}
-{"episode_index": 27, "sequence_id": "2026-09-10_15-07-44_white_laptop_cart_place_down_01", "camera": "right_stereo_camera_left", "object": "white_laptop_cart", "object_prompt": "white_laptop_cart.", "video_key": "observation.images.exo_camera"}
-{"episode_index": 28, "sequence_id": "2026-09-10_15-36-27_white_laptop_cart_place_down_02", "camera": "back_stereo_camera_left", "object": "white_laptop_cart", "object_prompt": "rolling white laptop desk with cup holder and basket.", "video_key": "observation.images.exo_camera"}
-{"episode_index": 29, "sequence_id": "2026-09-10_15-38-09_white_laptop_cart_push_pull_003", "camera": "left_stereo_camera_left", "object": "white_laptop_cart", "object_prompt": "rolling white laptop desk with cup holder and basket.", "video_key": "observation.images.exo_camera"}

track_1/meta/episodes_stats.jsonl

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:ed9bff1cac70865d95f3d798d82f4aa1da37b032de2c149c03bbcfd539d0c0da
-size 57505

track_1/meta/info.json

@@ -1,91 +0,0 @@
-{
-  "codebase_version": "v2.1",
-  "robot_type": "video_only_object_tracking",
-  "fps": 30,
-  "total_episodes": 30,
-  "total_frames": 16563,
-  "total_tasks": 22,
-  "total_videos": 30,
-  "total_chunks": 1,
-  "chunks_size": 1000,
-  "splits": {
-    "train": "0:30"
-  },
-  "data_path": "data/chunk-{episode_chunk:03d}/episode_{episode_index:06d}.parquet",
-  "video_path": "videos/chunk-{episode_chunk:03d}/{video_key}/episode_{episode_index:06d}.mp4",
-  "episodes_metadata": "meta/episodes_metadata.jsonl",
-  "cameras": [
-    "back_stereo_camera_left",
-    "front_stereo_camera_left",
-    "left_stereo_camera_left",
-    "right_stereo_camera_left"
-  ],
-  "features": {
-    "observation.images.exo_camera": {
-      "dtype": "video",
-      "shape": [
-        1152,
-        1536,
-        3
-      ],
-      "names": [
-        "height",
-        "width",
-        "channel"
-      ],
-      "info": {
-        "video.fps": 30,
-        "video.height": 1152,
-        "video.width": 1536,
-        "video.channels": 3,
-        "video.codec": "h264",
-        "video.pix_fmt": "yuv420p",
-        "video.is_depth_map": false,
-        "has_audio": false
-      },
-      "note": "one static third-person camera per episode; the actual camera (front/back/left/right_stereo_camera_left) is in meta/episodes_metadata.jsonl['camera']"
-    },
-    "timestamp": {
-      "dtype": "float32",
-      "shape": [
-        1
-      ],
-      "names": null
-    },
-    "frame_index": {
-      "dtype": "int64",
-      "shape": [
-        1
-      ],
-      "names": null
-    },
-    "episode_index": {
-      "dtype": "int64",
-      "shape": [
-        1
-      ],
-      "names": null
-    },
-    "index": {
-      "dtype": "int64",
-      "shape": [
-        1
-      ],
-      "names": null
-    },
-    "task_index": {
-      "dtype": "int64",
-      "shape": [
-        1
-      ],
-      "names": null
-    },
-    "next.done": {
-      "dtype": "bool",
-      "shape": [
-        1
-      ],
-      "names": null
-    }
-  }
-}

track_1/meta/tasks.jsonl

@@ -1,22 +0,0 @@
-{"task_index": 0, "task": "Walk over to the hula hoop on the floor and step inside the hula hoop. Lift the hula hoop over your entire body and place back on the floor when done."}
-{"task_index": 1, "task": "Walk over to the hula hoop on the floor and grasp it with your right hand keeping the hula hoop vertical. Pass your right foot  through the hula hoop and then your head and torso while shifting the hula hoop from your right hand to the left hand. Finally step through the hula hoop using your left leg. Place the hula hoop on the floor."}
-{"task_index": 2, "task": "Walk over to the hula hoop on the floor and pick up the hula hoop using both hands. Jump through the hula hoop as if it was a jump rope 3x."}
-{"task_index": 3, "task": "Walk over to the big red bowl on the round table and flip the big red bowl onto the round table. Step away when complete."}
-{"task_index": 4, "task": "Walk over to the big red bowl on the round table and push the red bowl away you using both hands as you walk around the round table. Step away when complete."}
-{"task_index": 5, "task": "Walk over to the big red bowl on the floor and pick up the red bowl using your right hand and then handover to your left hand and then place the big red bowl on the skinny wood chair. Step away when complete."}
-{"task_index": 6, "task": "Walk over to the iron on the round table and pick it up with your right hand stand it up vertically. Pick up the iron with your left hand and set it down horizontally on the round table."}
-{"task_index": 7, "task": "Walk over to the iron on the floor and pick up the iron using both hands and rotate the iron as you walk within the bounding area and place it on the ground, step away when complete."}
-{"task_index": 8, "task": "Walk over to the iron on the ground and kneel near the iron and pick up the iron using your right hand and proceed to iron the ground infront of you. Place the iron on the ground when complete and stand up."}
-{"task_index": 9, "task": "Walk to the white desk that is lying on its side. Place both hands on it and use them to carefully flip the desk onto the other side."}
-{"task_index": 10, "task": "Walk to the white desk, use your left foot to push the desk."}
-{"task_index": 11, "task": "Walk to the white desk, use both your hands to grasp on the edge of the table closest to you, and lower the desk to the ground."}
-{"task_index": 12, "task": "Pick up the pan from the table using one hand. Take a few steps away from the table. Using your other hand, touch the bottom of the pan with your palm, and make some circular motions to wipe it. Walk back to the table and put the pan down."}
-{"task_index": 13, "task": "Pick up the block from the table with both hands. Hold it by two opposite corners. Slowly rotate it several rotations while maintaining contact. Put it back on the table."}
-{"task_index": 14, "task": "Pick up the block from the table with both hands. Shift to holding it with one hand from the bottom. take a few steps back and face a different direction. Hold it with both hands again, and put it on the ground."}
-{"task_index": 15, "task": "Pick up the roller on the table. Slowly roll the roller back and forth on the table surface a few times while moving around the table. Lay the roller back on the table and step back."}
-{"task_index": 16, "task": "Walk over to the pink foam roll, and pick it up from the ground. Hold it from one end using one hand. Touch the other end of the roll on the ground a few times. Switch to the other hand and do the same. Put the roll back on the ground."}
-{"task_index": 17, "task": "Pick up the foam roll on the table. Use your hands to slowly roll the roller back and forth on the table surface a few times while walking along with it. Lay the roller back on the table and step back."}
-{"task_index": 18, "task": "Walk up to the wooden stool and sit on the stool. Rotate the direction you're facing while remaining seated. Stand up."}
-{"task_index": 19, "task": "Pick up the wooden stool from the ground using both hands. Let go of one hand and take a few steps. hold the stool with both hands again and place it back on the ground."}
-{"task_index": 20, "task": "Use both hands to pick up the white laptop cart, then lay it down sideways on the ground. Release it and stand up."}
-{"task_index": 21, "task": "Step up to the white laptop cart, and use one hand to push it forward a few steps. Put both hands on the cart, and pull it side ways a few steps. Release the cart and step away."}

track_2/README.md

@@ -1,12 +1,84 @@
-# Track 2: trajectory variants
+# Track 2
 
-Track 2 contains the same 30 human-object interaction episodes in two challenge tiers. Both tiers use the same episode order, 22,990-frame timeline, tasks, exocentric RGB observation, fitted ground planes, and G1 support surfaces.
+Human-object interaction sequences reconstructed with SOMA-X from four
+frame-synchronized RGB views. Every episode includes a full-body pose and one tracked
+object with a textured 3D mesh. Object poses are expressed as **`world_T_object` in the
+source OpenCV world frame**, so placing the mesh at a logged pose recovers its
+reconstructed location.
 
-| Tier | Directory | Human trajectory | Object trajectory and mesh |
-|---|---|---|---|
-| Tier 1: multiview caption | [`tier_1_multiview_caption/`](tier_1_multiview_caption/) | Original SOMA-X reconstruction | Original pose and aligned mesh |
-| Tier 2: synthetic noise | [`tier_2_synthetic_noise/`](tier_2_synthetic_noise/) | Synthetically perturbed human trajectory | Synthetically perturbed object trajectory; original mesh |
+| | |
+|---|---|
+| Episodes | 30 |
+| Frames | 22,990 (440–1,003 per episode, mean 766) |
+| Rate | 30 fps nominal |
+| Tasks | 30 |
+| Cameras | 4 streams, 120 clips |
+| Objects | 29, each with a textured mesh |
 
-Each directory is a complete LeRobot v2.1 dataset with its own `README.md`, `meta/`, `data/`, `videos/`, and `mesh/`. Select one tier directory as the dataset root. Do not concatenate tiers: episode and global frame indices intentionally overlap so results can be compared frame by frame.
+## Layout
 
-The `observation.images.exo_camera` video stream is byte-identical across tiers and is the monocular input participants should use for reconstruction. “Multiview caption” is the challenge tier name; this staged participant release exposes only the selected exocentric stream. Reconstruction results are not published; participants generate them with their own method before retargeting and grounding.
+```
+track_2/
+├── data/chunk-000/episode_0000NN.parquet    human + object pose, one row per frame
+├── videos/chunk-000/observation.images.<camera>/episode_0000NN.mp4
+├── mesh/<object>/<object>.glb               textured mesh
+└── meta/
+    ├── info.json                            schema, counts, tasks, objects
+    ├── episodes_metadata.jsonl              episode → sequence, task, object, mesh
+    ├── episodes.jsonl                       episode lengths and tasks
+    ├── episodes_stats.jsonl                 per-episode feature statistics
+    └── tasks.jsonl                          task descriptions
+
+```
+
+## Pose data
+
+Each parquet has **13 columns**, one row per video frame:
+
+```
+observation.human.pose                   fixed_size_list<float32, 231>
+observation.human.translation            fixed_size_list<float32, 3>
+observation.human.identity_coeffs        fixed_size_list<float32, 45>
+observation.human.scale_params           fixed_size_list<float32, 68>
+observation.human.bone_length_flexibles  fixed_size_list<float32, 6>
+observation.object.pose                  fixed_size_list<float32, 7>
+observation.object.visible               bool
+timestamp  frame_index  episode_index  index  task_index  next.done
+```
+
+Each episode tracks exactly one object. Its name is stored in the parquet's schema
+metadata under `objects` and in `meta/episodes_metadata.jsonl` with the source sequence,
+task, and mesh path.
+
+An object's name **is** its mesh folder: `white_desk` →
+`mesh/white_desk/white_desk.glb`.
+
+### Conventions
+
+- **`observation.human.pose` contains 77 local rotation vectors in radians**, flattened
+  in the joint order declared by `meta/info.json`.
+- Human translation is in metres. Identity, scale, and flexible-bone values are the
+  corresponding SOMA-X body-model parameters.
+- **`observation.object.pose` is `[x, y, z, qw, qx, qy, qz]`** — translation in metres,
+  **quaternion w-first**.
+- **Object poses are `world_T_object`** in the source OpenCV world frame. Each parquet
+  declares this in schema metadata under `pose_convention`.
+- When `observation.object.visible` is false the object pose is all zeros — a zero
+  quaternion, not a rotation. **Filter on visibility before using a pose.**
+
+## Cameras
+
+| Stream | Resolution | Kind |
+|---|---|---|
+| `back_stereo_camera_left` | 1536×1152 | colour, static third-person |
+| `front_stereo_camera_left` | 1536×1152 | colour, static third-person |
+| `left_stereo_camera_left` | 1536×1152 | colour, static third-person |
+| `right_stereo_camera_left` | 1536×1152 | colour, static third-person |
+
+Video is **frame-exact with the parquets** — row *i* corresponds to frame *i* in every
+clip, verified across all 120. Files are byte-identical copies of the
+source H.264 videos; no concatenation or re-encoding was performed.
+
+The supplied export has **no camera intrinsics or extrinsics**, so meshes cannot be
+projected into the images from this dataset alone. The source calibration sequence
+identifier is preserved per episode, but it is not a calibration matrix.