HumanRecon: Neural reconstruction of dynamic human using geometric cues and physical priors

Recent methods for dynamic human reconstruction have achieved promising reconstruction results. Most of these methods rely only on RGB color supervision without considering explicit geometric constraints or physical priors. This makes existing human reconstruction techniques more prone to overfitting to color and causes inherent geometric ambiguities, especially in sparse multi-view setups. Motivated by recent advances in monocular geometry prediction, we consider the geometric constraints of estimated depth and normals in learning neural implicit representation for dynamic human reconstruction. As a geometric regularization, this provides reliable yet explicit supervision information and improves reconstruction quality. We also exploit several beneficial physical priors, such as adding noise to the view direction and maximizing density on the human surface. These priors ensure the color rendered along rays is robust to view direction and reduce the inherent ambiguities of density estimated along rays. Experimental results demonstrate that depth and normal cues, predicted by human-specific monocular estimators, can provide effective supervision signals and render more accurate images. Finally, we also show that the proposed physical priors significantly reduce overfitting and improve the quality of novel view synthesis. Our code is available at: https://github.com/PRIS-CV/HumanRecon.