Dataset Viewer
Auto-converted to Parquet Duplicate
video_index
int32
9.68k
9.73k
frame_index
int16
0
1.44k
frame_bytes
unknown
video_key
stringclasses
43 values
9,683
0
[ 255, 216, 255, 224, 0, 16, 74, 70, 73, 70, 0, 1, 1, 0, 0, 1, 0, 1, 0, 0, 255, 219, 0, 67, 0, 3, 2, 2, 2, 2, 2, 3, 2, 2, 2, 3, 3, 3, 3, 4, 6, 4, 4, 4, 4, 4, 8, 6, 6, 5, 6, 9, 8, 10, 10, 9, 8, 9, 9, 1...
factory012_worker025_00015
9,683
1
[ 255, 216, 255, 224, 0, 16, 74, 70, 73, 70, 0, 1, 1, 0, 0, 1, 0, 1, 0, 0, 255, 219, 0, 67, 0, 3, 2, 2, 2, 2, 2, 3, 2, 2, 2, 3, 3, 3, 3, 4, 6, 4, 4, 4, 4, 4, 8, 6, 6, 5, 6, 9, 8, 10, 10, 9, 8, 9, 9, 1...
factory012_worker025_00015
9,683
2
[ 255, 216, 255, 224, 0, 16, 74, 70, 73, 70, 0, 1, 1, 0, 0, 1, 0, 1, 0, 0, 255, 219, 0, 67, 0, 3, 2, 2, 2, 2, 2, 3, 2, 2, 2, 3, 3, 3, 3, 4, 6, 4, 4, 4, 4, 4, 8, 6, 6, 5, 6, 9, 8, 10, 10, 9, 8, 9, 9, 1...
factory012_worker025_00015
9,683
3
"/9j/4AAQSkZJRgABAQAAAQABAAD/2wBDAAMCAgICAgMCAgIDAwMDBAYEBAQEBAgGBgUGCQgKCgkICQkKDA8MCgsOCwkJDRENDg8(...TRUNCATED)
factory012_worker025_00015
9,683
4
"/9j/4AAQSkZJRgABAQAAAQABAAD/2wBDAAMCAgICAgMCAgIDAwMDBAYEBAQEBAgGBgUGCQgKCgkICQkKDA8MCgsOCwkJDRENDg8(...TRUNCATED)
factory012_worker025_00015
9,683
5
"/9j/4AAQSkZJRgABAQAAAQABAAD/2wBDAAMCAgICAgMCAgIDAwMDBAYEBAQEBAgGBgUGCQgKCgkICQkKDA8MCgsOCwkJDRENDg8(...TRUNCATED)
factory012_worker025_00015
9,683
6
"/9j/4AAQSkZJRgABAQAAAQABAAD/2wBDAAMCAgICAgMCAgIDAwMDBAYEBAQEBAgGBgUGCQgKCgkICQkKDA8MCgsOCwkJDRENDg8(...TRUNCATED)
factory012_worker025_00015
9,683
7
"/9j/4AAQSkZJRgABAQAAAQABAAD/2wBDAAMCAgICAgMCAgIDAwMDBAYEBAQEBAgGBgUGCQgKCgkICQkKDA8MCgsOCwkJDRENDg8(...TRUNCATED)
factory012_worker025_00015
9,683
8
"/9j/4AAQSkZJRgABAQAAAQABAAD/2wBDAAMCAgICAgMCAgIDAwMDBAYEBAQEBAgGBgUGCQgKCgkICQkKDA8MCgsOCwkJDRENDg8(...TRUNCATED)
factory012_worker025_00015
9,683
9
"/9j/4AAQSkZJRgABAQAAAQABAAD/2wBDAAMCAgICAgMCAgIDAwMDBAYEBAQEBAgGBgUGCQgKCgkICQkKDA8MCgsOCwkJDRENDg8(...TRUNCATED)
factory012_worker025_00015
End of preview. Expand in Data Studio

V-JEPA Mamba Dataset (v2)

Preprocessed video frames from the Egocentric-10K dataset, formatted for V-JEPA Mamba pretraining.

Dataset Summary

Each video from Egocentric-10K is filtered and clipped:

  • Duration filter: Videos < 3 minutes are discarded. Videos > 3 minutes are clipped to the first 3 minutes.
  • Frame extraction: 8 fps using ffmpeg → 1,440 frames per video
  • Resolution: 384×384 RGB, shortest-side scaling + center crop
  • Format: Each video stored as a single Parquet file with columns:
Column Type Description
video_index int32 Unique video index (0-based)
frame_index int16 Frame position within the video (0-1439)
frame_bytes binary JPEG-encoded frame (quality 92)
video_key string Original video identifier from Egocentric-10K

Dataset Structure

data/
  video_00000.parquet       # Videos 0-9999
  ...
  _state.json                # Resume checkpoint
  shard_01/
    video_10000.parquet      # Videos 10000-19999
    ...

(Sharded at 9,500 files per directory to respect HF's 10k/dir limit)

Usage

import pyarrow.parquet as pq
from PIL import Image
import io

# Load a video's frames
table = pq.read_table("data/shard_01/video_10238.parquet")
for row in table.to_pylist():
    img = Image.open(io.BytesIO(row["frame_bytes"]))  # 384×384 RGB
    # ... training loop ...

Processing Pipeline

Built with preprocess-v3.py:

  1. Download + Extract: 8 parallel download threads + 8 extraction threads, decoupled via queue
  2. Process: All available CPU cores (30 reserved), ffmpeg decode → JPEG encode → Parquet
  3. Upload: Single commit per chunk via upload_folder (avoids rate limits)

Resume-safe: _state.json tracks tar-level progress. On restart, fully-done tars are skipped entirely (zero re-download).

Source

builddotai/Egocentric-10K — 19,495 tar shards across 85 factories of egocentric video recordings.

Citation

@misc{vjepa_mamba_dataset,
  title={V-JEPA Mamba Dataset: Preprocessed Egocentric Video Frames},
  author={Phi-9 Research},
  year={2026},
  url={https://huggingface.co/datasets/rookierufus/Vjepa_mamba_dataset_v2}
}
Downloads last month
3,568