4D-HOF: Hand-Object Flow Matching for Feed-Forward 4D Interaction Reconstruction
The paper’s claim is a feed-forward system that refines rough hand-object motion estimates into stable 4D reconstructions without per-sequence optimization.
4D-HOF uses conditional flow matching to move foundation-model-derived hand and object states toward a learned interaction manifold. The authors say this lets the model correct translation, rotation, and alignment errors during generation. Test-time guidance is built into that process, using physical interaction constraints and observed 2D evidence rather than a separate cleanup step. They report state-of-the-art results on out-of-domain benchmarks and better stability in in-the-wild scenes. ArXiv · AI/CL/LG's note
4D-HOF uses conditional flow matching to move foundation-model-derived hand and object states toward a learned interaction manifold. The authors say this lets the model correct translation, rotation, and alignment errors during generation. Test-time guidance is built into that process, using physical interaction constraints and observed 2D evidence rather than a separate cleanup step. They report state-of-the-art results on out-of-domain benchmarks and better stability in in-the-wild scenes. ArXiv · AI/CL/LG's note
score 5