Datasetpublic

sakifkhan98/cattle-vision-data

Cattle Vision Data — Pre-computed Tracking Results Pre-computed OC-SORT tracking outputs (with SAM2 mask RLEs) for the Cattle Vision Framework. Downloading these files skips scripts 06–08 (several hours of GPU inference on a V100). Files File Uncompressed Contents tracking_v2_cbvd5.tar.gz ~20 MB Per-video OC-SORT + SAM2 mask RLE JSONs for CBVD-5 tracking_v2_cvb.tar.gz ~641 MB Per-video OC-SORT + SAM2 mask RLE JSONs for CVB Usage… See the full description on the dataset page: https://huggingface.co/datasets/sakifkhan98/cattle-vision-data.

sourceHugging Facemitupdated 4mo agoView on Hugging Face

0likes5downloads

Dataset Card

Cattle Vision Data — Pre-computed Tracking Results

Pre-computed OC-SORT tracking outputs (with SAM2 mask RLEs) for the Cattle Vision Framework.

Downloading these files skips scripts 06–08 (several hours of GPU inference on a V100).

Files

File	Uncompressed	Contents
`tracking_v2_cbvd5.tar.gz`	~20 MB	Per-video OC-SORT + SAM2 mask RLE JSONs for CBVD-5
`tracking_v2_cvb.tar.gz`	~641 MB	Per-video OC-SORT + SAM2 mask RLE JSONs for CVB

Usage (Quickstart Path)

bash

# Download
huggingface-cli download sakifkhan98/cattle-vision-data \
  tracking_v2_cbvd5.tar.gz tracking_v2_cvb.tar.gz \
  --repo-type dataset --local-dir /tmp/tracking_v2/

# Extract into processed data directory
mkdir -p data/processed/tracking_v2/
tar -xf /tmp/tracking_v2/tracking_v2_cbvd5.tar.gz -C data/processed/tracking_v2/
tar -xf /tmp/tracking_v2/tracking_v2_cvb.tar.gz   -C data/processed/tracking_v2/

After extraction, data/processed/tracking_v2/{cbvd5,cvb}/ will be populated. Continue with scripts/09_generate_tubelets.sh.

JSON Schema

Each {video_id}_tracks.json file contains per-frame tracking results:

json

{
  "frame_idx": 0,
  "track_id": 1,
  "bbox": [x1, y1, x2, y2],
  "score": 0.95,
  "mask_rle": {"size": [H, W], "counts": "..."}
}

Citation

bibtex

@mastersthesis{khan2026cattle,
  author = {Sakif Khan},
  title  = {Multi-Behavior Recognition in Dairy Cattle from Surveillance Video},
  school = {Texas State University},
  year   = {2026},
}