Semantic-Aware Fine-Grained Correspondence

Establishing visual correspondence across images is a challenging and essential task. Recently, an influx of self-supervised methods have been proposed to better learn representations for correspondence. However, we find that these often fail leverage semantic information over-rely on the matching low-level features. In contrast, human vision capable distinguishing between distinct objects as pretext tracking. Inspired by this paradigm, propose semantic-aware fine-grained Firstly, demonstrate implicitly available through rich set image-level methods. We further design pixel-level learning objective which specifically targets For downstream tasks, fuse two kinds complementary together, demonstrating they boost performance synergistically. Our method surpasses previous state-of-the-art using convolutional networks variety including video object segmentation, pose tracking, part