ssv2
Datasets
All datasets matching “ssv2”ssv2-annotations-v1
ssv2-annotations-v1
Annotations only — no images, no video.
VITRA-style hand episodes for Something-Something V2, with per-hand instructions and paraphrases.
episodes
52,706
training samples (index_frame_pair rows)
1,124,722
annotation
MANO pose + world/camera joints + per-frame extrinsics
text
one instruction per episode + 2.09 paraphrases on average
images / video
not included — see Getting the frames below
What we did
Episodes are NOT… See the full description on the dataset page: https://huggingface.co/datasets/MIT-Media-Lab/ssv2-annotations-v1.ssv2-frame-lance
SSv2 frame-Lance (deterministic 20% prefix)
This repository currently contains only the exact deterministic 20% subset
used by the LeVJEPA scaling-law pretraining configuration. It is not the full
Something-Something v2 dataset.
Field
Value
Source episode space
168,913
Available episodes
33,783
Available shards
8 (00000–00007)
Frame rows
1,905,090
Lance payload
65,348,568,212 bytes
Clips per video
1
The subset is the first 33,783 entries of the seed-0… See the full description on the dataset page: https://huggingface.co/datasets/hs272/ssv2-frame-lance.ssv2_testThis dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v3.0",
"robot_type": "unknown",
"total_episodes": 27157,
"total_frames": 1271386,
"total_tasks": 174,
"chunks_size": 1000,
"data_files_size_in_mb": 100,
"video_files_size_in_mb": 200,
"fps": 12,
"splits": {
"train": "0:27157"
},
"data_path": "data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet",
"video_path":… See the full description on the dataset page: https://huggingface.co/datasets/n3puiol/ssv2_test.ssv2ssv2-3x3
SSV2 3x3
A multi-frame action dataset for teaching a model to look across several frames and describe the action it sees.
Each image is a 3x3 collage built from frames sampled from the original Something-Something V2 videos.
Splits
Split
Samples
train
150000
test
1000
Columns
image: 3x3 frame collage
label_text: action label text
annotation_text: action template text
video_id: original video identifier
visual_cues_ssv2