Learning Skip Map For Efficient Ultra-High Resolution Image Segmentation

The pattern distribution of ultra-high resolution images is usually unbalanced. While part of an image contains complex and fine-grained patterns such as boundaries, most areas are composed of simple and repeated patterns. In this work, we propose to learn a skip map, which can guide a segmentation network to skip simple patterns and hence reduce computational complexity. Specifically, the skip map highlights simple-pattern areas that can be down-sampled for processing at a lower resolution, while the remaining complex part is still segmented at the original resolution. We empirically show that the skip map can also reflect the quality of visual representations. Applied on the state-of-the-art ultra-high resolution image segmentation network, our proposed skip map saves more than 30% computation while maintaining comparable segmentation performance.