Unsupervised Dense Shape Correspondence using Heat Kernels
This work addresses shape correspondence for computer graphics and vision researchers, offering an incremental improvement by replacing geodesic distances with heat kernels for faster training.
The paper tackles the problem of learning dense shape correspondences without supervision by using heat kernels as a supervisor signal within a deep functional map framework, achieving results on benchmarks with challenges like partiality and topological noise.
In this work, we propose an unsupervised method for learning dense correspondences between shapes using a recent deep functional map framework. Instead of depending on ground-truth correspondences or the computationally expensive geodesic distances, we use heat kernels. These can be computed quickly during training as the supervisor signal. Moreover, we propose a curriculum learning strategy using different heat diffusion times which provide different levels of difficulty during optimization without any sampling mechanism or hard example mining. We present the results of our method on different benchmarks which have various challenges like partiality, topological noise and different connectivity.