by ACVLab
ABot World Explorer 500h is a dataset of 30,969 action-conditioned video episodes from ACVLab. Each episode pairs an MP4 with keyboard actions, captions and a COLMAP camera model.
DatasetFree
Step 1: Browse episodes in the Hugging Face Dataset Viewer to find a sample ID.
Step 2: Pick a full commit ID if you need reproducible downloads.
Step 3: Use snapshot_download from huggingface_hub with allow_patterns for that sample's video.mp4 and annotations.tar.
Step 4: Check the archive's members and paths before extracting, and do not call extractall without checks.
Step 5: Read action.json, caption.json and the sparse COLMAP files together for each episode.
Apache-2.0.
An MP4 video and an annotations archive with action.json, caption.json and a COLMAP sparse model of cameras, images and 3D points.
Not necessarily. The card says the name is an identifier, not an audited duration claim.
Yes. The card shows how to fetch the files for one sample ID with snapshot_download and allow_patterns.