OmniEgoCap: Device-Agnostic Sequence-Level Egocentric Motion Reconstruction
OmniEgoCap is a diffusion framework that reconstructs full-body motion from egocentric videos across different cameras without needing specific hardware retraining. It achieves this by analyzing entire sequences rather than short clips and using geometry-aware augmentation to handle varying camera views and occlusions. This allows the model to infer consistent body shapes and coherent movement even with sparse head and hand data.