Emergent Correspondence from Image Diffusion
This addresses the fundamental computer vision problem of image correspondence for researchers and practitioners, offering a novel unsupervised approach that is competitive with supervised methods.
The paper tackled the problem of finding correspondences between images by showing that correspondence emerges in image diffusion models without explicit supervision, and they proposed DIFT to extract these features, achieving a 19-point accuracy improvement over DINO on the SPair-71k benchmark.
Finding correspondences between images is a fundamental problem in computer vision. In this paper, we show that correspondence emerges in image diffusion models without any explicit supervision. We propose a simple strategy to extract this implicit knowledge out of diffusion networks as image features, namely DIffusion FeaTures (DIFT), and use them to establish correspondences between real images. Without any additional fine-tuning or supervision on the task-specific data or annotations, DIFT is able to outperform both weakly-supervised methods and competitive off-the-shelf features in identifying semantic, geometric, and temporal correspondences. Particularly for semantic correspondence, DIFT from Stable Diffusion is able to outperform DINO and OpenCLIP by 19 and 14 accuracy points respectively on the challenging SPair-71k benchmark. It even outperforms the state-of-the-art supervised methods on 9 out of 18 categories while remaining on par for the overall performance. Project page: https://diffusionfeatures.github.io