SAM3-Assisted Training of Lightweight YOLO Models for Precision Pig Farming

2026-05-25

AI Summary

This study addresses the challenges of deploying high-performance object detection models on edge devices for precision pig farming, where lightweight models typically rely on labor-intensive manual annotations while powerful foundation models are computationally prohibitive. To overcome this, the work proposes an efficient, annotation-free training and deployment framework that leverages SAM 3 as an offline zero-shot pseudo-label generator, followed by knowledge distillation to train a compact YOLOv8 model. This is the first effort to employ SAM 3 for fully automatic pseudo-labeling, enabling end-to-end training without any human intervention. Evaluated on the PigLife dataset, the distilled YOLOv8m achieves 79.4% mAP, with inference speed approximately 200 times faster than the teacher model and AP50 exceeding 99% under low-occlusion conditions—performance nearly on par with models trained using manually annotated data.

Abstract

Deep learning-based object detection has revolutionized Precision Livestock Farming (PLF), yet a critical barrier remains: high-performance Foundation Models (such as SAM 3) are too computationally intensive for edge deployment, while lightweight models (like YOLO) require prohibitive manual annotation efforts. This work proposes a fully automated knowledge distillation pipeline that leverages the Segment Anything Model 3 (SAM 3) to generate zero-shot pseudo-labels for training efficient YOLOv8 detectors. By treating SAM 3 as an offline auto-annotator, we eliminate the manual labeling bottleneck, producing models capable of real-time inference on resource-constrained hardware. We systematically evaluate this approach on the PigLife dataset, comparing SAM 3-supervised models against human-annotated baselines. Results demonstrate that a SAM 3-trained YOLOv8m achieves a mean Average Precision (mAP) of 79.4% without human intervention, while reducing inference latency by approximately 200$\times$ compared to the teacher model. Furthermore, stratified analysis reveals that in low-occlusion scenarios, the automated pipeline achieves detection rates comparable to human benchmarks ($AP_{50} > 99\%$). These findings indicate that foundation models can serve as effective, zero-annotation-cost supervisors, enabling scalable edge computing solutions for smart agriculture.

Problem

• Precision Livestock Farming
• Foundation Models
• Lightweight Models
• Manual Annotation
• Edge Deployment

Innovation

• knowledge distillation
• zero-shot pseudo-labeling
• lightweight YOLO
• foundation model supervision
• edge deployment