CVPR2025
EchoMatch: Partial-to-Partial Shape Matching via Correspondence Reflection
Yizheng Xie, Viktoria Ehm, Paul Roetzer, Nafie El Amrani, Maolin Gao, Florian Bernard, Daniel Cremers
Abstract
We present EchoMatch, a novel framework for partial-to-partial shape matching. We formalise a correspondence reflection approach, where we ensure forward-backward-consistency for overlapping points. If a point returns to its original neighbourhood after sending it from shape X to shape Y and back to shape X , we consider it to be an overlapping point, otherwise it is a non-overlapping point. Middle: We show improved overlap prediction for partial-to-partial shape matching compared to the previous SOTA method DPFM [3] . Right: We achieve SOTA matching results in the partial-to-partial setting.