Download README.md from we-dream/pretraining-data: direct link, hf CLI and curl.
- Browser
- Download file 2.16 kB
-
https://huggingface.co/datasets/we-dream/pretraining-data/resolve/main/README.md
- Command line
-
hf download hf://datasets/we-dream/pretraining-data/README.md
-
curl -L -o README.md https://huggingface.co/datasets/we-dream/pretraining-data/resolve/main/README.md
pretty_name: WE-Dream Pretraining Data
viewer: false
tags:
- we-dream
- robotics
- egocentric-video
- world-action-model
WE-Dream Pretraining Data
EgoDex and EgoVerse data for WE-Dream pretraining, with latent-action and hand-pose annotations. Download links are provided below.
Downloads
| Dataset | Latent actions | Hand actions |
|---|---|---|
| EgoDex | Download on ModelScope | Download on ModelScope |
| EgoVerse 360p | Download on ModelScope | Download on ModelScope |
| EgoVerse 480p | Download on ModelScope | Download on ModelScope |
Latent-action pretraining uses the latent-action datasets. Action-free pretraining reuses their video and language. Hand-action pretraining uses the hand-action datasets.
Access and download
Open the ModelScope.ai links you need, sign in, and request access. Use a ModelScope.ai token from the same account for downloading.
Follow the data download guide to configure access. From the WE-Dream code repository root, download an individual dataset with:
python -m pip install modelscope-hub==0.4.6
python scripts/data/download_pretrain_dataset.py egodex_latentaction
Replace egodex_latentaction with the dataset name at the end of any download link above. Files are saved under data/pretrain/<dataset>; rerun the command to resume an interrupted download.
See the pretraining guide for the latent-action, hand-action, and action-free recipes. Dataset access and license terms are provided by the linked source repositories.
EgoWE is not included in the public data release.