Agreement-Driven Multi-View 3D Reconstruction for Live Cattle Weight Estimation

TL;DR

This work tackles non-contact live cattle weight estimation by presenting a RGB-only 3D reconstruction pipeline that leverages SAM 3D with an agreement-guided multi-view fusion to produce accurate cattle geometries. The 3D representations are then fed into a stacked ensemble of 11 regression models trained on rich geometric features, achieving a best reported $R^2$ of $0.69 \\pm \\ 0.10$, $MAE = 9.16 \\pm \\ 2.32$ kg, and $MAPE = 2.22 \\pm \\ 0.56$% for weight prediction. In low-data farm settings, classical ensemble methods outperform DL-based reconstructions, highlighting the value of reconstruction quality and robust feature-based learning over deeper models. The proposed approach reduces hardware costs and handling requirements, enabling practical, on-farm deployment for improved cattle management and welfare.

Abstract

Accurate cattle live weight estimation is vital for livestock management, welfare, and productivity. Traditional methods, such as manual weighing using a walk-over weighing system or proximate measurements using body condition scoring, involve manual handling of stock and can impact productivity from both a stock and economic perspective. To address these issues, this study investigated a cost-effective, non-contact method for live weight calculation in cattle using 3D reconstruction. The proposed pipeline utilized multi-view RGB images with SAM 3D-based agreement-guided fusion, followed by ensemble regression. Our approach generates a single 3D point cloud per animal and compares classical ensemble models with deep learning models under low-data conditions. Results show that SAM 3D with multi-view agreement fusion outperforms other 3D generation methods, while classical ensemble models provide the most consistent performance for practical farm scenarios (R$^2$ = 0.69 $\pm$ 0.10, MAPE = 2.22 $\pm$ 0.56 \%), making this practical for on-farm implementation. These findings demonstrate that improving reconstruction quality is more critical than increasing model complexity for scalable deployment on farms where producing a large volume of 3D data is challenging.

Figures (6)

• Figure 1: MAPE and R$^2$ of weight estimation across 3D cattle reconstruction methods.
• Figure 2: Pipeline of agreement-driven multi-view 3D reconstruction for non-contact live cattle weight estimation.
• Figure 3: Visualization of multi-view agreement fusion in SAM 3D coarse geometry stage latent space. (a)--(f) shows the spatial distribution of the mean 3-view agreement score as the iterations progress. (g) is the per-view mean agreement score. (h) is the visualization of the final multi-view fused 3D Gaussian point cloud.
• Figure 4: Visual Comparison of different point clouds. Obvious structural anomalies are marked with red boxes. SAM 3D outputs Gaussian point clouds; however, we ignore the Gaussian parameters and visualise them as uncoloured point clouds for comparison.
• Figure 5: Effect of ensemble size on MAPE (%) for different 3D dataset. Performance improves with increasing top ML models but stabilizes after Top-11 for both RGB+D and DL-based 3D methods.