Generating textured 3D dairy cow models from a few images

Automated individual re-identification of dairy cows from surveillance cameras is essential for the early detection of diseases and estrus.Training accurate re-identification models requires large-scale annotated datasets, which are labor-intensive to acquire. To address this issue, this paper proposes a system that generates textured 3D models of dairy cows from a few images. Our pipeline integrates state-of-the-art models to reconstruct full-body 3D models from just five images to synthesize multi-view data. We demonstrate the effectiveness of this 3D-based augmentation for training re-identification models through experiments on synthetic data.