SAM 3D Animal: Promptable Animal 3D Reconstruction from Images in the Wild

Researchers introduced SAM 3D Animal, a promptable framework for reconstructing multiple animals in 3D from a single image.

The method uses SMAL+ along with keypoint and mask prompts to resolve instance ambiguity in crowded, occluded scenes.

It was trained on a new 5K+ image dataset, Herd3D, and achieves state-of-the-art results on several benchmarks.

The work improves 3D animal understanding in realistic multi-animal settings and outperforms prior model-based and model-free approaches.