SideCow-VSS: A Video Semantic Segmentation Dataset and Benchmark for Intelligent Monitoring of Dairy Cows Health in Smart Ranch Environments
View abstract on PubMed
Summary
This summary is machine-generated.A new dataset, SideCow-VSS, aids precision livestock farming by enabling AI-driven monitoring of dairy cows. This resource supports developing computer vision tools for early disease detection and improved herd health management.
Area Of Science
- Computer Vision
- Animal Science
- Machine Learning
Background
- Precision livestock farming relies on non-invasive dairy cow monitoring for health management.
- AI development is limited by a lack of realistic video datasets and standardized benchmarks for deep learning models.
Purpose Of The Study
- Introduce SideCow-VSS, a novel video semantic segmentation dataset for dairy cows.
- Evaluate deep learning model performance for automated cow monitoring tasks.
- Provide a benchmark for future computer vision research in livestock farming.
Main Methods
- Collected 921 side-view video clips of dairy cows under various on-farm conditions.
- Annotated data with dense, pixel-level segmentation masks.
- Systematically evaluated eight deep learning architectures, including CNNs and Transformers.
Main Results
- Mask2Former with Swin-L backbone achieved the highest mIoU (97.32%) for morphological analysis.
- PIDNet-s demonstrated the fastest inference speed (59.5 FPS) for real-time applications.
- Performance varied across models, highlighting a trade-off between accuracy and speed.
Conclusions
- SideCow-VSS serves as a foundational resource for advancing computer vision in dairy farming.
- The study provides a quantitative framework for selecting appropriate deep learning models.
- This work facilitates the development of systems for automated health monitoring and disease prevention in dairy production.

