The full dataset viewer is not available (click to read why). Only showing a preview of the rows.
Error code: DatasetGenerationError
Exception: IndexError
Message: list index out of range
Traceback: Traceback (most recent call last):
File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1859, in _prepare_split_single
original_shard_lengths[original_shard_id] += len(table)
~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^
IndexError: list index out of range
The above exception was the direct cause of the following exception:
Traceback (most recent call last):
File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 1369, in compute_config_parquet_and_info_response
parquet_operations, partial, estimated_dataset_info = stream_convert_to_parquet(
~~~~~~~~~~~~~~~~~~~~~~~~~^
builder, max_dataset_size_bytes=max_dataset_size_bytes
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
)
^
File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 948, in stream_convert_to_parquet
builder._prepare_split(split_generator=splits_generators[split], file_format="parquet")
~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1694, in _prepare_split
for job_id, done, content in self._prepare_split_single(
~~~~~~~~~~~~~~~~~~~~~~~~~~^
gen_kwargs=gen_kwargs, job_id=job_id, **_prepare_split_args
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
):
^
File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1880, in _prepare_split_single
raise DatasetGenerationError("An error occurred while generating the dataset") from e
datasets.exceptions.DatasetGenerationError: An error occurred while generating the datasetNeed help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
text string |
|---|
2026-01-01 00:00:00.000: axis_0,d,0.239 |
2026-01-01 00:00:00.000: axis_1,d,-0.36 |
2026-01-01 00:00:00.000: axis_2,d,-0.085 |
2026-01-01 00:00:00.000: axis_3,d,-0.023 |
2026-01-01 00:00:00.033: axis_0,d,0.258 |
2026-01-01 00:00:00.033: axis_1,d,-0.388 |
2026-01-01 00:00:00.033: axis_2,d,-0.088 |
2026-01-01 00:00:00.033: axis_3,d,-0.024 |
2026-01-01 00:00:00.066: axis_0,d,0.277 |
2026-01-01 00:00:00.066: axis_1,d,-0.415 |
2026-01-01 00:00:00.066: axis_2,d,-0.09 |
2026-01-01 00:00:00.066: axis_3,d,-0.026 |
2026-01-01 00:00:00.099: axis_0,d,0.295 |
2026-01-01 00:00:00.099: axis_1,d,-0.441 |
2026-01-01 00:00:00.099: axis_2,d,-0.092 |
2026-01-01 00:00:00.099: axis_3,d,-0.028 |
2026-01-01 00:00:00.133: axis_0,d,0.313 |
2026-01-01 00:00:00.133: axis_1,d,-0.467 |
2026-01-01 00:00:00.133: axis_2,d,-0.095 |
2026-01-01 00:00:00.133: axis_3,d,-0.029 |
2026-01-01 00:00:00.166: axis_0,d,0.33 |
2026-01-01 00:00:00.166: axis_1,d,-0.491 |
2026-01-01 00:00:00.166: axis_2,d,-0.097 |
2026-01-01 00:00:00.166: axis_3,d,-0.031 |
2026-01-01 00:00:00.199: axis_0,d,0.346 |
2026-01-01 00:00:00.199: axis_1,d,-0.514 |
2026-01-01 00:00:00.199: axis_2,d,-0.098 |
2026-01-01 00:00:00.199: axis_3,d,-0.032 |
2026-01-01 00:00:00.233: axis_0,d,0.361 |
2026-01-01 00:00:00.233: axis_1,d,-0.535 |
2026-01-01 00:00:00.233: axis_2,d,-0.1 |
2026-01-01 00:00:00.233: axis_3,d,-0.033 |
2026-01-01 00:00:00.266: axis_0,d,0.375 |
2026-01-01 00:00:00.266: axis_1,d,-0.555 |
2026-01-01 00:00:00.266: axis_2,d,-0.102 |
2026-01-01 00:00:00.266: axis_3,d,-0.034 |
2026-01-01 00:00:00.299: axis_0,d,0.388 |
2026-01-01 00:00:00.299: axis_1,d,-0.574 |
2026-01-01 00:00:00.299: axis_2,d,-0.103 |
2026-01-01 00:00:00.299: axis_3,d,-0.035 |
2026-01-01 00:00:00.333: axis_0,d,0.399 |
2026-01-01 00:00:00.333: axis_1,d,-0.591 |
2026-01-01 00:00:00.333: axis_2,d,-0.105 |
2026-01-01 00:00:00.333: axis_3,d,-0.036 |
2026-01-01 00:00:00.366: axis_0,d,0.409 |
2026-01-01 00:00:00.366: axis_1,d,-0.606 |
2026-01-01 00:00:00.366: axis_2,d,-0.106 |
2026-01-01 00:00:00.366: axis_3,d,-0.037 |
2026-01-01 00:00:00.399: axis_0,d,0.417 |
2026-01-01 00:00:00.399: axis_1,d,-0.619 |
2026-01-01 00:00:00.399: axis_2,d,-0.107 |
2026-01-01 00:00:00.399: axis_3,d,-0.038 |
2026-01-01 00:00:00.433: axis_0,d,0.423 |
2026-01-01 00:00:00.433: axis_1,d,-0.631 |
2026-01-01 00:00:00.433: axis_2,d,-0.108 |
2026-01-01 00:00:00.466: axis_0,d,0.429 |
2026-01-01 00:00:00.466: axis_1,d,-0.642 |
2026-01-01 00:00:00.499: axis_0,d,0.432 |
2026-01-01 00:00:00.499: axis_1,d,-0.651 |
2026-01-01 00:00:00.499: axis_2,d,-0.109 |
2026-01-01 00:00:00.499: axis_3,d,-0.039 |
2026-01-01 00:00:00.533: axis_0,d,0.435 |
2026-01-01 00:00:00.533: axis_1,d,-0.659 |
2026-01-01 00:00:00.566: axis_0,d,0.436 |
2026-01-01 00:00:00.566: axis_1,d,-0.666 |
2026-01-01 00:00:00.599: axis_0,d,0.435 |
2026-01-01 00:00:00.599: axis_1,d,-0.672 |
2026-01-01 00:00:00.599: axis_3,d,-0.038 |
2026-01-01 00:00:00.633: axis_0,d,0.434 |
2026-01-01 00:00:00.633: axis_1,d,-0.677 |
2026-01-01 00:00:00.666: axis_0,d,0.431 |
2026-01-01 00:00:00.666: axis_1,d,-0.681 |
2026-01-01 00:00:00.666: axis_2,d,-0.108 |
2026-01-01 00:00:00.699: axis_0,d,0.427 |
2026-01-01 00:00:00.699: axis_1,d,-0.685 |
2026-01-01 00:00:00.699: axis_3,d,-0.037 |
2026-01-01 00:00:00.733: axis_0,d,0.422 |
2026-01-01 00:00:00.733: axis_1,d,-0.688 |
2026-01-01 00:00:00.733: axis_2,d,-0.107 |
2026-01-01 00:00:00.766: axis_0,d,0.415 |
2026-01-01 00:00:00.766: axis_1,d,-0.691 |
2026-01-01 00:00:00.766: axis_2,d,-0.106 |
2026-01-01 00:00:00.766: axis_3,d,-0.036 |
2026-01-01 00:00:00.799: axis_0,d,0.408 |
2026-01-01 00:00:00.799: axis_1,d,-0.693 |
2026-01-01 00:00:00.799: axis_2,d,-0.105 |
2026-01-01 00:00:00.799: axis_3,d,-0.035 |
2026-01-01 00:00:00.833: axis_0,d,0.4 |
2026-01-01 00:00:00.833: axis_1,d,-0.695 |
2026-01-01 00:00:00.833: axis_2,d,-0.104 |
2026-01-01 00:00:00.833: axis_3,d,-0.034 |
2026-01-01 00:00:00.866: axis_0,d,0.391 |
2026-01-01 00:00:00.866: axis_1,d,-0.697 |
2026-01-01 00:00:00.866: axis_2,d,-0.102 |
2026-01-01 00:00:00.899: axis_0,d,0.381 |
2026-01-01 00:00:00.899: axis_1,d,-0.699 |
2026-01-01 00:00:00.899: axis_2,d,-0.101 |
2026-01-01 00:00:00.899: axis_3,d,-0.033 |
2026-01-01 00:00:00.933: axis_0,d,0.37 |
2026-01-01 00:00:00.933: axis_1,d,-0.701 |
WorldRover-preview
Long, continuous camera paths through photoreal 3D environments, rendered with per-frame metric depth, camera pose and action labels — and, where it matters, rendered more than once along the same path, so you can hold geometry and motion fixed while the projection or the lighting changes.
This repository is the original preview release, as a lite subset: RGB, camera pose, actions and metadata for four scenes, without depth (depth is 87% of the bytes). The rest of the data lives in the repositories below — see the collection for all parts in one place.
Parts
| Repository | What it is | Scenes | Clips | Video | Size |
|---|---|---|---|---|---|
this repo (WorldRover-preview) |
lite: paired panoramic + first-person, RGB + pose + actions, no depth | 4 | 129 per view | 4.1 h per view | 103 GB |
| WorldRover-6scenes | paired 360° panoramic + first-person, with lossless depth | 6 | 600 per view | 18.9 h per view | 7.6 TB |
| WorldRover-styles | one trajectory set re-rendered under 6 lighting/style treatments | 4 | 1000 | 48 h | 818 GB |
| med_village · paris · venice · art_nouveau | the original release, full depth, one repo per scene | 1 each | 13–47 per view | ~30 min per view | 169–334 GB |
Start with WorldRover-6scenes unless you specifically want the original four-scene
release or the style variants — it is the largest, newest and most complete part.
The two kinds of pairing
Same path, two projections (WorldRover-6scenes, and this preview repo). pano/<clip_id> and
fp/<clip_id> are the same camera path: the first-person clip is rendered from the panoramic
clip's per-frame trajectory, so the two camera_trajectory.csv files agree row for row and only
the intrinsics differ (360°/0 mm equirect vs 65.5°/28 mm pinhole). No interpolated alignment is
involved.
Same path, six appearances (WorldRover-styles). Each trajectory is rendered as a reference
first-person clip plus up to five variants — untextured white model, golden hour, night, snow,
storm — frame-aligned, so the depth, pose and action labels of the reference clip apply to every
variant of it unchanged.
What a clip contains
rgb.mp4 H.264, 30 fps
depth/depth.mkv FFV1 16-bit, lossless, log-quantized radial depth
depth/depth.meta.json near/far, authoritative frame count, decode formula
camera_trajectory.csv per-frame world pose + intrinsics
description.json scene, asset pack, licence, trajectory summary, render settings
gamepad_format/ action labels (axis events + timeline)
trajectory.png top-down path plot
Clips are 11 s to 8.5 min of continuous motion — no cuts, no teleports.
Conventions that are easy to get wrong
- Depth is log-quantized and radial:
depth_m = exp(code/65535 * (log 200 − log 0.1) + log 0.1), and the value is distance along the ray, not along the optical axis. Convert before unprojecting. The pixel format is not the same everywhere:WorldRover-6scenesis uniformlygray16le, whileWorldRover-stylesis mostlygbrp16lewith the code in the R channel (three art_nouveau clips excepted). Each clip'sdepth/depth.meta.jsonnames its own format — read it instead of assuming. - Frame counts:
camera_trajectory.csvhas one row more than the video has frames (the last row is the closing keyframe). The authoritative count isdepth/depth.meta.json. - Poses are Unreal-style: left-handed, centimetres, X-forward / Y-right / Z-up, camera looking down its own +X.
- Panoramic frames are equirectangular: a rectangular crop is not a perspective view. Reproject before comparing one against a first-person clip.
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-6scenes/venice/fp/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.
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.
- Downloads last month
- 914