README.md

metadata

license: other
license_name: worldrover-research
pretty_name: WorldRover style variants
task_categories:
  - depth-estimation
  - robotics
tags:
  - video
  - depth
  - camera-pose
  - embodied-ai
  - world-model
  - synthetic
  - relighting
  - domain-randomization
size_categories:
  - 100B<n<1T

WorldRover-styles

The same camera path rendered under six different treatments. 200 first-person trajectories across four scenes, each re-rendered as up to six variants — so the geometry, motion and action labels are held fixed while lighting, weather and material appearance change. 1000 clips, ~48 h of video at 30 fps, ~818 GB.

Variant	Scenes	Clips	What changes	Per clip
first_person	all 4	200	the reference render	rgb + depth + pose + actions
white_model	all 4	200	untextured white shading — pure geometry	rgb only
goldenhour	3	150	low warm sun	rgb only
night	3	150	night lighting	rgb only
snow	3	150	snow cover + overcast	rgb only
storm	3	150	storm sky + wet surfaces	rgb only

goldenhour / night / snow / storm cover med_village, paris and venice; art_nouveau is an interior set and ships first_person + white_model only.

Scene	Base trajectories	Median clip	Range	Video per variant
med_village	50	3 min 29 s	41 s – 6 min 45 s	3.01 h
venice	50	2 min 54 s	1 min 07 s – 6 min 43 s	2.66 h
paris	50	1 min 54 s	49 s – 5 min 15 s	1.80 h
art_nouveau	50	1 min 32 s	14 s – 5 min 00 s	1.39 h

Layout

<scene>/first_person/<clip_id>/     pinhole 1280x720, 30 fps
    rgb.mp4  depth/depth.mkv  depth/depth.meta.json
    camera_trajectory.csv  description.json  gamepad_format/  trajectory.png
<scene>/<variant>/<clip_id>/rgb.mp4  same path, same frame count, different look
style_grids/batch_grid_<scene>_<variant>.mp4   contact-sheet previews (64 videos)

A variant clip shares its <clip_id> with the first_person clip, so the pose, depth and action labels from first_person/<clip_id> apply to every variant of it unchanged.

Conventions

• Read the depth pixel format from each clip's depth/depth.meta.json — this batch is not uniform. 197 of the 200 clips are FFV1 / gbrp16le with the code in the R channel (G = B = 0); three art_nouveau clips — an725_navmesh_000005, an_react_000022, an_react_000037 — are single-plane FFV1 / gray16le instead. Every clip's depth.meta.json states its own format correctly (verified against the files), so read it rather than assuming. WorldRover-6scenes is uniformly gray16le. The quantization is identical either way: depth_m = exp(code/65535 * (log 200 − log 0.1) + log 0.1), radial distance.
• camera_trajectory.csv has one row more than the video has frames; the authoritative count is depth/depth.meta.json.
• Poses are Unreal-style: left-handed, centimetres, X-forward / Y-right / Z-up.

Tools

git clone https://github.com/AlayaLab/WorldRover
pip install -r WorldRover/tools/requirements.txt   # numpy, opencv-python; ffmpeg/ffprobe on PATH
cd WorldRover/tools                                # the package is not pip-installable yet

from worldrover import Clip

clip    = Clip("/data/WorldRover-styles/venice/first_person/fix_venice_000000")
rgb     = clip.rgb_frame(100)     # uint8 (H, W, 3), sRGB
depth_m = clip.depth_frame(100)   # planar depth in metres
pts     = clip.points_world(100)  # world points, centimetres
poses   = clip.poses              # per-frame pose + intrinsics

Clip takes a filesystem path to a clip directory. The reader takes the depth pixel format from that clip's own depth/depth.meta.json, so it handles both encodings used across these repositories, and it converts radial depth and the off-by-one trajectory row for you.

The style variants hold only rgb.mp4; point Clip at the matching first_person clip for that trajectory's depth, pose and action labels.

Related

• WorldRover-preview — the original release as a lite subset (no depth)
• Collection — every part in one place
• WorldRover-6scenes — paired 360° panoramic + first-person, six scenes, 7.6 TB

License

Rendered video, depth, camera pose and action labels are released for research use. The underlying 3D environments are third-party commercial assets, are not redistributed here, and each clip's description.json records its asset pack and licence. Tools are MIT.