BECA: Beef Cattle Images for Long-Term Recognition and Diversity

A computer vision dataset for precision livestock farming, created by Zhang Yuqi and last updated in April 2026. It contains 29,061 total images from two sub-datasets: BECA-D for visual diversity with 16,889 images from 5,661 cattle, and BECA-L for long-term recognition with 12,172 images from 103 cattle tracked over up to five months.

Use Cases

• Long-term cattle re-identification based on dorsal view images tracked over months.
• Developing models for cattle recognition across multiple breeds based on visual diversity.
• Training object detection models using the provided annotation subsets.
• Conducting pose estimation for behavioral analysis in livestock.
• Benchmarking vision-based systems for automated livestock management.

Strengths

• Contains 29,061 total images, providing substantial scale for model training.
• Includes a long-term tracking subset (BECA-L) with images from 103 cattle over a continuous period of up to five months.
• Offers annotation subsets for key pipeline tasks like object detection and pose estimation.
• The dataset is 18.9 GB in size, indicating high-resolution or numerous image files.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• The geographic origin and specific breeds of the cattle are not specified, which may limit assessment of bias.
• Row count is unknown, which may limit suitability assessment for certain tasks.

Provenance

Source

figshare

Collection Method

Likely collected from real-world agricultural settings for computer vision research.

Time Range

The long-term tracking subset covers up to five months. Specific collection dates are not provided.

Freshness

Last updated 2026-04-22 12:38:02; freshness should be verified.