Beyond Spatio-Temporal Priors: A Generalizable Approach for Dense Correspondence Matching
FreeMatching is built to match identity across edited or generated images where physical continuity no longer holds.
The paper says classical dense matching assumptions, such as smooth motion and rigid geometry, fail in image editing and reference-guided generation. Its proposed model combines generative and semantic foundation representations, trained with supervision from classical datasets, tracked videos, and synthetic scenes. The authors report better correspondence quality on difficult IEG pairs while keeping competitive results on classical benchmarks. They also present FreeMatching as a metric for identity preservation, with scores said to correlate with human judgment. ArXiv · AI/CL/LG's note
The paper says classical dense matching assumptions, such as smooth motion and rigid geometry, fail in image editing and reference-guided generation. Its proposed model combines generative and semantic foundation representations, trained with supervision from classical datasets, tracked videos, and synthetic scenes. The authors report better correspondence quality on difficult IEG pairs while keeping competitive results on classical benchmarks. They also present FreeMatching as a metric for identity preservation, with scores said to correlate with human judgment. ArXiv · AI/CL/LG's note
score 4