ChickenVerse is a large-scale, openly released dataset for computer vision in broiler welfare monitoring. It covers imagery and video from five poultry facilities and includes two complementary sub-datasets, ChickenDet and ChickenAct, each with recommended train/val/test splits and evaluation protocols.
Paper: ChickenVerse: An Open Large-Scale Multi-Task Dataset for Chicken Detection, Segmentation, and Behavior Recognition | A. Amirivojdan et al., University of Tennessee.
ChickenDet is a COCO-format dataset for object detection and instance segmentation. It contains 6,539 images with 153,764 annotated bird instances, each carrying a mutually consistent bounding box and pixel-level mask. Frames were sampled from overhead-view recordings at five-minute intervals to maximize visual diversity, and annotations were produced with CVAT using SAM 2–assisted mask labeling. The dataset is partitioned by source facility: the training set spans multiple farms and camera geometries, the validation set comes from JARTU research pens, and the test set is a fully held-out commercial broiler house (F4), making test scores a measure of cross-facility transfer.
| Split | Images | Instances | Mean inst./img |
|---|---|---|---|
| Train | 5,222 | 116,329 | 22.3 |
| Val | 1,067 | 24,815 | 23.3 |
| Test | 250 | 12,620 | 50.5 |
| Total | 6,539 | 153,764 | 23.5 |
ChickenAct is a video classification dataset for broiler behavior recognition. It contains 15,250 short single-bird video clips spanning 15 welfare-relevant behavior classes (e.g., eating, drinking, walking, resting, preening, dust bathing). Clips were generated automatically via a detect-track-crop pipeline built on a ChickenDet-finetuned RT-DETRv2 detector and a Norfair tracker, then manually labeled through a lightweight annotation interface. Each clip is stored as an MP4 file at the source frame rate, organized into per-class folders.
ChickenAct.mp4
| Split | Clips | Classes |
|---|---|---|
| Train | 11,278 | 15 |
| Val | 1,829 | 15 |
| Test | 2,143 | 15 |
| Total | 15,250 | 15 |
This dataset is licensed under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License (CC BY-NC-SA 4.0).
You are free to share and adapt the data for non-commercial purposes, provided you give appropriate credit and distribute any derivatives under the same license.