The dataset viewer is not available because its heuristics could not detect any supported data files. You can try uploading some data files, or configuring the data files location manually.
OpenScene LTF training cache (3 cameras)
Pre-processed training samples for a Latent TransFuser (LTF) driving policy, built with py123d and py123d_garage (pre-v0.2 layout) from OpenScene logs, which are derived from the nuPlan dataset.
Contents
Everything is under thispc-ltf-cache/:
| Path | What it holds |
|---|---|
py123d_garage_cache/ltf_openscene_<city>/<window>/data.mdb |
One LMDB store per log window (cities: boston, pittsburgh, singapore, vegas, plus validation) |
py123d_garage_cache/ltf_openscene_<city>/manifest.yaml |
Tensor encodings and the cache signature of that store |
logs/<window>/{sync,ego_state_se3,route_position,box_detections_se3}.arrow |
Per-window logs the trainer reads to list samples and build navigation targets |
sources.json |
Training and validation windows per city, and the number of training samples |
manifest.json |
Size and SHA-256 of every file, sample counts and the software versions used |
- 791 log windows and 372,412 training samples across Boston, Pittsburgh, Singapore and Las Vegas, plus a fixed validation set.
- Per sample: the left, front and right cameras stitched into one 1024×256 JPEG (quality 95); bird's-eye-view semantic labels (64 m × 64 m, 4 px/m); CenterNet box targets; and a 4 s ego trajectory at 2 Hz with heading.
Not compatible with py123d_garage v0.2.x, whose policies use the front camera only at 352×512 with a frame from 1 s earlier, and 6.4 s trajectories.
Verifying a download
manifest.json lists the size and SHA-256 of every file; compare them after download.
License and attribution
This cache is derived from nuPlan (Motional), redistributed through OpenScene (OpenDriveLab) under CC BY-NC-SA 4.0. It is shared under the same license: non-commercial use only, with attribution, and any derivative must use the same license. Users must also follow the nuPlan terms of use.
- nuPlan: H. Caesar et al., "nuPlan: A closed-loop ML-based planning benchmark for autonomous vehicles", 2021.
- OpenScene: OpenDriveLab, "OpenScene: The Largest Up-to-Date 3D Occupancy Prediction Benchmark in Autonomous Driving", 2023.
- Downloads last month
- 48