Mk Bashar

Sow body condition is an important indicator for growers as it has a large impact on lactation performance and piglet survival. However, body condition measures used during production, such as visual scoring and calipers, correlate poorly with underlying tissue composition. Ultrasound scans can provide direct measurements of subcutaneous backfat thickness and loin muscle depth, but their operation is labor intensive and not scalable for production. To address this, we present PigFormer, a fully automated pipeline that estimates subcutaneous backfat thickness, loin muscle depth, and total tissue thickness at the 12th rib from a ceiling-mounted RGB-D camera. Our method converts raw depth frames into standardized height maps via SAM3-to-MaskDINO segmentation distillation, ground-plane removal, and orientation normalization. PigFormer then processes each height map as a sequence of cross-sectional slices using a transformer encoder with Rotary Position Embeddings, capturing spatial relationships along the full dorsal surface. On a multi-site dataset of 319 sow and gilt instances (6,705 frames) from two U.S. facilities, PigFormer achieves a backfat mean absolute error of 2.35 mm and overall error of 3.91 mm, improving over two proposed baseline approaches.

Animal posture is a key indicator of health, behavior, and welfare in precision livestock farming. However, manual posture monitoring in large-scale swine production systems is labor-intensive, costly, and often impractical. While recent computer vision approaches have shown promise for automated pig posture recognition, they are typically trained on limited or homogeneous camera viewpoints, leading to poor performance on novel datasets. In this work, a multi-view pig posture benchmark dataset with five posture classes and per-instance camera viewpoint angle information is introduced. Then, a viewpoint-aware posture recognition framework that integrates geometric context into deep visual representations and incorporates camera domain-invariant learning via an adversarial layer is designed. The proposed viewpoint-aware approach enables the method to better generalize to unseen viewpoints, achieving a +2% macro F1-score gain over the baseline and reaching 86.13%.

We present a camera-based method for estimating the respiratory rate of naturally sleeping pigs from video. Our pipeline leverages dense point tracking with CoTracker to capture subtle belly motion, followed by Butterworth bandpass filtering and correlation-based trajectory grouping to isolate breathing signals. The trajectories are analyzed using Fast Fourier Transform (FFT) to extract the dominant respiratory frequency. We evaluate the method on a dataset of 55 nighttime video samples captured of group-housed pigs, achieving a mean absolute error of 1.48 breaths per minute. Compared to an intensity-based baseline, our approach shows significantly higher accuracy and robustness under low-light, cluttered farm conditions. Ablation studies further demonstrate the contribution of each component and the impact of query region size. This work offers a non-invasive solution for automated respiratory monitoring in livestock environments.

Multiple Object Tracking (MOT) is a core capability in modern computer vision, essential to autonomous driving, surveillance, sports analytics, robotics, and biomedical imaging. Persistent identity assignment across frames remains challenging in real scenes because of occlusion, dense crowds, appearance ambiguity, scale variation, camera motion, and identity switching. In this survey we synthesize recent progress by organizing methods around the problems they target and the paradigms they adopt. We cover the historical progression from tracking-by-detection to hybrid and end-to-end designs, and we summarize major architectural directions including transformer-based trackers, generative/diffusion formulations, state-space predictors, Siamese and graph-based models, and the growing impact of foundation models for detection and representation. We review benchmark trends that motivate method design, documenting the shift from saturated pedestrian benchmarks to challenge-driven and domain-specific datasets and we analyze evaluation practice by comparing classic and newer motion- and safety-centric metrics. Finally, we connect algorithmic trends to practical deployment constraints and outline emerging directions, foundation-model integration, open-vocabulary and multimodal tracking, unified evaluation, and domain-adaptive methods, that we believe will shape MOT research and real-world adoption.