Research map / Label-efficient and robust learning / Recognition and long-tail
Digital Twin-Driven Teat Localization and Shape Identification for Dairy Cow (Student Abstract)
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Applies DINO, YOLO-F and Faster R-CNN object detectors to localize dairy cow teats and classify teat-end shape from a small camera-collected dataset, with DINO reaching 0.783 mAP.
Problem
Automating dairy cow teat localization and teat-end shape classification, which are predictors of animal health (mastitis, hyperkeratosis) and milk quality, since manual assessment is burdensome on large farms.
Approach
Applies existing transformer and CNN object detectors to a newly collected, veterinarian-labeled teat dataset from a rotary milking parlor, treating teat localization and shape classification as one detection task and comparing detectors.
Key results
- DINO achieved the highest validation mAPs of 0.783, versus 0.634 for YOLO-F and 0.573 for Faster R-CNN.
- DINO's average inference time was 628 ms, about 110% of the baselines' runtime per the paper's claim.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use DINO instead of YOLO-F | teat localization and shape classification · validation set | mAP: 0.634 -> 0.783 | Helped |
| Use DINO instead of YOLO-F | teat localization and shape classification · validation set | avg inference time: 598 ms -> 628 ms | Hurt |
| Use DINO instead of Faster RCNN | teat localization and shape classification · validation set | mAP: 0.573 -> 0.783 | Helped |
| Use DINO instead of Faster RCNN | teat localization and shape classification · validation set | avg inference time: 576 ms -> 628 ms | Hurt |
Limitations
- Small preliminary dataset from a single farm and one camera setup
- Only small objects; evaluation limited to three detectors without broader architecture coverage
Experimental setup
Datasets: in-house dairy cow teat video keyframes (348 images, 968 objects)
Benchmarks: in-house teat validation split
Metrics: mean average precision (mAP) average inference time
Baselines: YOLO-F Faster R-CNN
Builds on: DINO Faster R-CNN YOLO
Closest papers by meaning
- MmCows: A Multimodal Dataset for Dairy Cattle Monitoring (NeurIPS 2024 · similarity 0.84)
- AgMMU: A Comprehensive Agricultural Multimodal Understanding Benchmark (NeurIPS 2025 · similarity 0.78)
- Can Large Multimodal Models Understand Agricultural Scenes? Benchmarking with AgroMind (NeurIPS 2025 · similarity 0.77)
- Animal-Bench: Benchmarking Multimodal Video Models for Animal-centric Video Understanding (NeurIPS 2024 · similarity 0.77)
- BioTrove: A Large Curated Image Dataset Enabling AI for Biodiversity (NeurIPS 2024 · similarity 0.76)
- Combining Deep Learning and Street View Imagery to Map Smallholder Crop Types (AAAI 2024 · similarity 0.76)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.