Amodal Segmentation Based on Visible Region Segmentation and Shape Prior

Almost all existing amodal segmentation methods make the inferences of occluded regions by using features corresponding to whole image. This is against human's perception, where human uses visible part and shape prior knowledge target infer region. To mimic behavior solve ambiguity in learning, we propose a framework, it firstly estimates coarse mask mask. Then based on prediction, our model infers concentrating region utilizing memory. In this way, background occlusion can be suppressed for estimation. Consequently, would not affected what given same regions. The leverage makes estimation more robust reasonable. Our proposed evaluated three datasets. Experiments show that outperforms state-of-the-art methods. visualization indicates category-specific feature codebook has certain interpretability. code available at https://github.com/YutingXiao/Amodal-Segmentation-Based-on-Visible-Region-Segmentation-and-Shape-Prior.