Attention-Based Multi-Context Guiding for Few-Shot Semantic Segmentation

• DOI: 10.1609/aaai.v33i01.33018441
• Publication Date: 2019-08-21T07:38:33Z
• Authors: Tao Hu; Pengwan Yang; Chiliang Zhang; Gang Yu; Yadong Mu; Cees G. M. Snoek
• Keywords: Pascal (unit); Feature Learning; Feature (linguistics)

Few-shot learning is a nascent research topic, motivated by the fact that traditional deep methods require tremendous amounts of data. The scarcity annotated data becomes even more challenging in semantic segmentation since pixellevel annotation task labor-intensive to acquire. To tackle this issue, we propose an Attentionbased Multi-Context Guiding (A-MCG) network, which consists three branches: support branch, query feature fusion branch. A key differentiator A-MCG integration multi-scale context features between and branches, enforcing better guidance from set. In addition, also adopt spatial attention along branch highlight information several scales, enhancing self-supervision one-shot learning. address problem multi-shot learning, Conv-LSTM adopted collaboratively integrate sequential elevate final accuracy. Our architecture obtains state-of-the-art on unseen classes variant PASCAL VOC12 dataset performs favorably against previous work with large gains 1.1%, 1.4% measured mIoU 1-shot 5-shot setting.