Learning from the Target: Dual Prototype Network for Few Shot Semantic Segmentation

Due to the scarcity of annotated samples, diversity between support set and query becomes main obstacle for few shot semantic segmentation. Most existing prototype-based approaches only exploit prototype from feature ignore information sample, failing remove this obstacle.In paper, we proposes a dual network (DPNet) dispose segmentation new perspective. Along with extracted set, propose build pseudo-prototype based on foreground features in image. To achieve goal, cycle comparison module is developed select reliable generate them. Then, interaction utilized integrate their underlying correlation. Finally, multi-scale fusion introduced capture contextual during dense (pseudo-prototype) feature. Extensive experiments conducted two benchmarks demonstrate that our method exceeds previous state-of-the-arts sizable margin, verifying effectiveness proposed method.