Vol. 10 No. 2 (2026): Few-shot

This issue publishes the original research paper "Few-shot Semantic Segmentation Network with Prototype Enhancement and Prior Guidance (PEPGNet)".   Aiming at the problems of weak category prototype representation and low prior mask utilization in few-shot semantic segmentation, this paper proposes two core modules:  

1. Focused Attention Prototype Module (FAP): Generates enhanced prototypes with global semantic consistency and local discriminability by fusing multi-scale context and self-attention.
2. Prior Mask Guidance Module (PMG): Deeply encodes prior masks to capture spatial structure information, providing accurate geometric guidance for segmentation.

  Experiments on PASCAL and COCO datasets show that PEPGNet achieves significant mIoU improvements over the baseline ProtoFormer, outperforming existing state-of-the-art methods. This work is an original research achievement in computer vision, suitable for publication in this journal.