{
 "schema": "vpm-worldbench/sample/1.0",
 "uid": "Simul-General/clip_00029",
 "subset": "Simul-General",
 "id": "clip_00029",
 "source": {
  "dataset": "HyperWorld-Bench-General",
  "repo": "hf:rooty2020/HyperWorld-Bench-General",
  "revision": "ca94a48704dc230bf39202f3c672f356aacb21c7",
  "id": "clip_00029",
  "path": "bench_people&car/clip_00029"
 },
 "n_views": 4,
 "n_frames": 81,
 "fps": 15.0,
 "duration_s": 5.4,
 "views": [
  {
   "name": "view0",
   "video": "videos/view0.mp4",
   "width": 1280,
   "height": 1280,
   "source_name": "cam01",
   "role": null,
   "static": false,
   "hfov_deg": 51.85,
   "motion_kind_rel_anchor": "pan_tilt",
   "max_travel_m": 8.412,
   "max_rotation_deg": 11.276
  },
  {
   "name": "view1",
   "video": "videos/view1.mp4",
   "width": 1280,
   "height": 1280,
   "source_name": "cam02",
   "role": null,
   "static": false,
   "hfov_deg": 38.357,
   "motion_kind_rel_anchor": "static",
   "max_travel_m": 8.412,
   "max_rotation_deg": 0.0
  },
  {
   "name": "view2",
   "video": "videos/view2.mp4",
   "width": 1280,
   "height": 1280,
   "source_name": "cam03",
   "role": null,
   "static": false,
   "hfov_deg": 26.003,
   "motion_kind_rel_anchor": "random_polyline",
   "max_travel_m": 9.2369,
   "max_rotation_deg": 11.358
  },
  {
   "name": "view3",
   "video": "videos/view3.mp4",
   "width": 1280,
   "height": 1280,
   "source_name": "cam04",
   "role": null,
   "static": false,
   "hfov_deg": 28.049,
   "motion_kind_rel_anchor": "translation",
   "max_travel_m": 7.9206,
   "max_rotation_deg": 28.84
  }
 ],
 "world": {
  "description": "CARLA Town10HD_x6 map frame as delivered by the vendor in right-handed metres (X_ue/100, -Y_ue/100, Z_ue/100), +z up, ground at z ~ 0, recentred per sample: X = X_source - o with o = (x, y, 0) of the target surface-point centroid at frame 0.",
  "units": "m",
  "handedness": "right",
  "up": "+z",
  "origin_in_source": [
   91.96752106282209,
   -88.4270934664185,
   0.0
  ],
  "to_source": {
   "note": "X_source = A @ X + b; source = vendor right-handed metre world of cameras.json extrinsic_w2c / c2w",
   "A": [
    [
     1.0,
     0.0,
     0.0
    ],
    [
     0.0,
     1.0,
     0.0
    ],
    [
     0.0,
     0.0,
     1.0
    ]
   ],
   "b": [
    91.96752106282209,
    -88.4270934664185,
    0.0
   ],
   "source_units": "m"
  },
  "to_ue_cm": {
   "note": "X_ue = A @ X + b; Unreal/CARLA left-handed centimetre world of ue_pose.location_cm and visibility.json *_cm",
   "A": [
    [
     100.0,
     0.0,
     0.0
    ],
    [
     0.0,
     -100.0,
     0.0
    ],
    [
     0.0,
     0.0,
     100.0
    ]
   ],
   "b": [
    9196.752106282209,
    8842.70934664185,
    0.0
   ],
   "source_units": "cm"
  }
 },
 "files": {
  "cameras": "cameras.npz",
  "tracks3d": "tracks3d.npz",
  "captions": "captions.json",
  "depth": "depth/depth.zip",
  "labels": {
   "target": "labels/target.npz",
   "oos_segments": "labels/oos_segments.json"
  }
 },
 "tracks": {
  "n_points": 256,
  "groups": [
   "target"
  ],
  "visible_outside_frustum_dropped": 0,
  "default_report_group": "target",
  "visibility_rule": "visible = in_frustum AND depth test against depth.zip of that view and frame at the nearest pixel (rint(u), rint(v)) of the native 1280x1280 grid, (u, v) projected with the stored K (OpenCV pixel centres, cx = cy = 639.5; = floor(u), floor(v) with K_delivered): |D - z| <= clip(0.02 z, 0.05 m, 0.15 m), D = planar depth in metres, 65535 (sky) never passes. Strict rule: area-sampled points on the far side or inside the target (vehicle interior/glass, eyes) fail it; the tolerance is 2 % of the distance, at least 5 cm and at most 15 cm. in_frustum = recomputed with the stored K (valid, z > 0.05 m and -0.5 <= u, v < 1279.5, i.e. the vendor in_frustum_cam rule z > 0 and 0 <= u, v < 1280 with the delivered K; agreement with in_frustum_cam in stats). The legacy pre-release protocol used the vendor in_frustum_cam alone (frustum-only).",
  "accuracy": "exact simulator GT: 256 triangle-area-weighted surface points moved by linear blend skinning of the CARLA bone transforms (rigid propagation for unskinned vehicle parts); no cloth/hair secondary motion; other traffic is untracked. Stored as float32 (< 4e-6 m rounding within 100 m of the origin). Depth is quantised to 1 cm. K = delivered intrinsics with cx = cy = 640 - 0.5 = 639.5: the vendor uses the CARLA/UE pixel-corner convention (depth-vs-track residuals are lowest with the corrected K); cameras.npz K_delivered keeps the delivered matrix (pre-release numbers used 640, docs/DATA_FORMAT.md §1).",
  "depth_test": {
   "pixel": "rint",
   "K": "cameras.npz K (OpenCV pixel centres)",
   "tol_rel": 0.02,
   "tol_min_m": 0.05,
   "tol_max_m": 0.15,
   "grid": [
    1280,
    1280
   ],
   "invalid_values": [
    65535
   ]
  },
  "stats": {
   "n_in_frustum": 82944,
   "n_visible": 56914,
   "visible_frac_of_in_frustum": 0.6862,
   "in_frustum_sky_hits": 0,
   "n_visible_other_pixel_rule": {
    "rint_K_delivered": 56427
   },
   "visible_abs_dz_m": {
    "median": 0.012316,
    "p90": 0.105674
   },
   "in_frustum_recomputed_agreement": 1.0,
   "in_frustum_recomputed_mismatches": 0
  },
  "part_note": "part = vendor point_section (mesh section / material slot, '<mesh>/<material>'; a name that occurs more than once gets '#<section index>' appended, codes unchanged); component = mesh component (doors, glass, lights for vehicles); point_skinned = driven by bone skinning (else rigid).",
  "extras": {
   "component": "(N,) int16 index into component_names: mesh component of the point (e.g. body, doors, glass, lights for vehicles)",
   "component_names": "(C,) str component names",
   "component_mesh_names": "(C,) str mesh file of each component",
   "point_skinned": "(N,) bool, moved by bone skinning (else rigid with its component)"
  }
 },
 "depth": {
  "file": "depth/depth.zip",
  "member": "{source_view}/frame_{t:04d}.png",
  "view_member_names": [
   "cam01",
   "cam02",
   "cam03",
   "cam04"
  ],
  "dtype": "uint16",
  "unit_scale": 0.01,
  "invalid_values": [
   65535
  ],
  "valid_max_m": 655.34,
  "width": 1280,
  "height": 1280,
  "kind": "rendered (synthetic GT)",
  "planar_z": true,
  "note": "CARLA sensor.camera.depth decoded by the vendor to linear planar z in cm, lossless 16-bit grayscale PNG, pixel-aligned with the RGB video; 65535 = sky/far clamp (no zeros occur). The pre-release loader additionally discarded depth > 80 m."
 },
 "caption": "A bald man in a dark long-sleeved jacket with light lettering, gray pants, and dark shoes jogs across a shaded pedestrian plaza toward a crosswalk and street, passing benches, planters, and lampposts under overcast daylight.",
 "attributes": {
  "category": "pedestrian",
  "subcategory": "pedestrian",
  "semantic_tag": 12,
  "blueprint": "walker.pedestrian.0045",
  "difficulty": "normal",
  "hard_relay_k": null,
  "camera_motion_kinds": [
   "pan_tilt",
   "static",
   "random_polyline",
   "translation"
  ],
  "all_cameras_moving": true,
  "map": "Town10HD_x6",
  "tile": 0,
  "rotation_only_views": [],
  "oos_departed_views": [],
  "n_frames_target_invisible_all_views": 0,
  "n_default_queries_at_ref_frame": 235,
  "visible_frac_of_in_frustum": 0.6862,
  "quality_ok": true,
  "abort": null,
  "source_frame_range": [
   5167,
   5247
  ],
  "flag_rules": {
   "quality_ok_false": "the vendor capture gate .capture_done quality_ok is false",
   "target_not_visible_in_view_at_ref_frame": "at frame 16 the target is not visible in that view by either trigger: (a) the vendor target state (labels/target.npz state_strict) is not 'visible' (instance mask < 300 px, out of frustum or absent), or (b) the view has zero visible GT track points (tracks3d visible, depth test)",
   "target_invisible_all_views_some_frames": "at some frame the vendor target state (state_strict) is not 'visible' in any view",
   "caption_mentions_leaving_view": "the caption says the target leaves the view (out of/from view|frame|sight, leaves the view|frame, off-screen, disappear*)",
   "few_default_queries_at_ref_frame": "fewer than 20 points with default_gt and visible in >= 1 view at frame 16 (exact count in attributes.n_default_queries_at_ref_frame)"
  }
 },
 "flags": [],
 "files_sha256": {
  "cameras.npz": "b45129d4ec31337d57a73cc89a1d19ccee743c66e27f9054530c98428b248f7a",
  "captions.json": "42ad1f8916c33dfcf74c89b2ddb80b3b81ce75ea0ae770d392976a181c4d8bcd",
  "depth/depth.zip": "acb668a837903184e8309b0f57890384d882cd02b73148640a9517a1d95bfb6f",
  "labels/oos_segments.json": "b947cf6462023fd22b02443f58ec8d86d38f33f770cdb40be418a0dc6a47e490",
  "labels/target.npz": "f02bfbfd52f1a1120c2d2d62d025a3a84ec182d8a75340a742a4f0479d3466c8",
  "source/cameras.json": "a2f38ed02a74edb0f25ead8d4dc104bfc996cc411068767f08f48685834500ae",
  "source/caption.txt": "5eef5d1d8418da00a1e6fe887df69dfd6efd58083294f87364c4722e25601da1",
  "source/capture_done.json": "181e09aaa031f3da0328df3854c6c18190b22c89f737eb4df4e744581f2bf8d2",
  "source/meta.json": "913f12aff41ca49b90a4c4adacedd934e5cbaa45e3fff78055150ac576c8ae04",
  "source/points_target.npz": "fe3a1918f976ec68dcca4d460e84df77a2b5947f8d90a31be8da539c4d6e147f",
  "source/visibility.json": "85d03278c3c72fa7b150e65ee9284638ba8e19c7ed943a2116323ff6a5dc2439",
  "tracks3d.npz": "ca74b581f7f146b2b16b5c933ab06d83f8e6fb1c3e56fb5cc689773bfe100cbf",
  "videos/view0.mp4": "b7990f88fa8227dbf15b72ac09b24e6cfa156b319779ab69be8824077f5406bd",
  "videos/view1.mp4": "a7b31e8b34247512f8abed776c0f4dd4ff6eee808b42dd1661a2f270ecd4d019",
  "videos/view2.mp4": "63d11f63e04f31dbec6da6d4a834c894c05babc0b7fa6dd74e7f67fdf13dd6eb",
  "videos/view3.mp4": "92d32c185d7050a934c7ef08b1daad8e15eedacc7f681e277238af513120a6ca"
 }
}