APANet: Adaptive Prototypes Alignment Network for Few-Shot Semantic Segmentation

24 Nov 2021  ·  Jiacheng Chen, Bin-Bin Gao, Zongqing Lu, Jing-Hao Xue, Chengjie Wang, Qingmin Liao ·

Few-shot semantic segmentation aims to segment novel-class objects in a given query image with only a few labeled support images. Most advanced solutions exploit a metric learning framework that performs segmentation through matching each query feature to a learned class-specific prototype. However, this framework suffers from biased classification due to incomplete feature comparisons. To address this issue, we present an adaptive prototype representation by introducing class-specific and class-agnostic prototypes and thus construct complete sample pairs for learning semantic alignment with query features. The complementary features learning manner effectively enriches feature comparison and helps yield an unbiased segmentation model in the few-shot setting. It is implemented with a two-branch end-to-end network (i.e., a class-specific branch and a class-agnostic branch), which generates prototypes and then combines query features to perform comparisons. In addition, the proposed class-agnostic branch is simple yet effective. In practice, it can adaptively generate multiple class-agnostic prototypes for query images and learn feature alignment in a self-contrastive manner. Extensive experiments on PASCAL-5$^i$ and COCO-20$^i$ demonstrate the superiority of our method. At no expense of inference efficiency, our model achieves state-of-the-art results in both 1-shot and 5-shot settings for semantic segmentation.

PDF Abstract

Datasets


Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Few-Shot Semantic Segmentation COCO-20i (1-shot) APANet (ResNet-101) Mean IoU 41.9 # 45
Few-Shot Semantic Segmentation COCO-20i (1-shot) APANet (ResNet-50) Mean IoU 40.5 # 53
Few-Shot Semantic Segmentation COCO-20i (1-shot) APANet (VGG-16) Mean IoU 37.2 # 61
Few-Shot Semantic Segmentation COCO-20i (5-shot) APANet (ResNet-101) Mean IoU 46.4 # 51
Few-Shot Semantic Segmentation COCO-20i (5-shot) APANet (ResNet-50) Mean IoU 43 # 57
Few-Shot Semantic Segmentation COCO-20i (5-shot) APANet (VGG-16) Mean IoU 43.2 # 56
Few-Shot Semantic Segmentation PASCAL-5i (1-Shot) APANet (VGG-16) Mean IoU 59 # 81
Few-Shot Semantic Segmentation PASCAL-5i (1-Shot) APANet (ResNet-101) Mean IoU 64 # 56
Few-Shot Semantic Segmentation PASCAL-5i (1-Shot) APANet (ResNet-50) Mean IoU 63 # 63
Few-Shot Semantic Segmentation PASCAL-5i (5-Shot) APANet (ResNet-101) Mean IoU 68 # 53
Few-Shot Semantic Segmentation PASCAL-5i (5-Shot) APANet (VGG-16) Mean IoU 62.6 # 74
Few-Shot Semantic Segmentation PASCAL-5i (5-Shot) APANet (ResNet-50) Mean IoU 66 # 64

Methods


No methods listed for this paper. Add relevant methods here