ARXIV PREPRINT · CS.CV
Agreement-Driven Multi-View 3D Reconstruction for Live Cattle Weight Estimation
Publication status. This page describes the arXiv preprint. It does not assert publication in the IEEE ICIP proceedings. Conference and DOI metadata will be added only when an official publication record becomes available.
PLAIN-LANGUAGE SUMMARY
Three ordinary RGB views, inspected as one reconstruction.
The paper asks whether matched left, right, and top RGB views can support useful 3D cattle reconstruction without depth, LiDAR, or stereo hardware at inference. Agreement weighting gives more influence to view updates that align with cross-view consensus and less to conflicting updates. Geometric features from the resulting point cloud then support live-weight estimation.
REPORTED DATASET RESULT
103 CATTLE · 5-FOLD CV
2.22 ± 0.56% MAPE
R² 0.69 ± 0.10
These are dataset-level cross-validation results, not a prediction for the interactive Case 001 animal.