Given partially visible objects within images, Amodal3R reconstructs semantically meaningful 3D assets with reasonable geometry and plausible appearance.

Arxiv: https://arxiv.org/abs/2503.13439

Project Page: https://sm0kywu.github.io/Amodal3R/

Our model is built upon the "foundation" model TRELLIS. The pre-trained model weights are fetched from https://huggingface.co/JeffreyXiang/TRELLIS-image-large. Thanks to their impressive work!!!