Datasets:
Modalities:
Text
Formats:
json
Languages:
English
Size:
< 1K
ArXiv:
Tags:
autonomous-driving
vision-language-action
instruction-following
counterfactual-evaluation
carla
realsim
License:
|
Download language2drive.OpenloopTrack.nScenes/README.md from Mike087/Language2Drive: direct link, hf CLI and curl.
- Browser
- Download file 900 Bytes
-
https://huggingface.co/datasets/Mike087/Language2Drive/resolve/main/language2drive.OpenloopTrack.nScenes/README.md
- Command line
-
hf download hf://datasets/Mike087/Language2Drive/language2drive.OpenloopTrack.nScenes/README.md
-
curl -L -o README.md https://huggingface.co/datasets/Mike087/Language2Drive/resolve/main/language2drive.OpenloopTrack.nScenes/README.md
900 Bytes
| # Language2Drive nuScenes open-loop replay assets | |
| Private replay assets for `language2drive.OpenloopTrack.nScenes` in the Language2Drive source repository. Authorized downloads require a read-scoped Hugging Face token. Other dataset directories are unchanged. | |
| The archive contains 75 frozen public PCUs, evaluator-only contracts, three nuScenes map expansion files, and 720 heldout predictions per model for four models. It supports CPU replay of paper Table 9 without raw sensor files or model weights. Evaluate contracts only in the scorer, never in a model adapter. | |
| Hashes and file inventory are in `assets.json`. Original maps and nuScenes-derived data follow CC BY-NC-SA 4.0. Credit Holger Caesar et al., *nuScenes: A multimodal dataset for autonomous driving*, CVPR 2020, https://www.nuscenes.org/ . See `assets/DATA_LICENSE.md` in the archive. Source code retains its separate MIT license. | |