Search

Few-shot semantic segmentation tackles the problem of recognizing novel class objects from images with only a few annotated exemplars. The key problem in few-shot semantic segmentation is how to effectively model the correspondences between support and query features. Previous works propose to tackle the problem by prototype matching or distance based metric learning. In this work, we introduce a kernel-based similarity matching model, enforcing robust guidance from both foreground and background semantics. In addition, guidance sorting and allocation modules are presented to better explore the guidance from support set. Specifically, guidance sorting module calibrates the most similar semantic patterns on query maps for each support pixel and produces the index vectors. While the allocation module is able to select the most representative correspondences on similarity maps based on index vectors. To integrate the insights of kernel-based similarity features we define a pyramidal paradigm, which progressively integrates guidance signal, query features and mask priors. In this way, the relationships between support and query features are dynamically explored in both foreground and background semantics. Extensive qualitative and quantitative evaluations on PASCAL-5i, COCO-20i and FSS-1000 are conducted to prove the efficiency and advantage of our proposed method. Experimental results demonstrate that our method performs favorably against state-of-the-art methods with reasonable computational cost. [ABSTRACT FROM AUTHOR]

Few-shot semantic segmentation tackles the problem of recognizing novel class objects from images with only a few annotated exemplars. In this work, a multi-scale attentional similarity guidance neural network is proposed to tackle the challenging few-shot semantic segmentation task. This work attempts to build a pyramidal paradigm, which analyzes multi-scale correspondence from diverse levels of semantic features. To achieve this goal, the attentional guided network is introduced to exploit similarity maps, attention vectors as well as query features to segment unseen class objects using only a few image samples. Specifically, the proposed network aims to extract the similarity maps with both foreground and background semantics, which enable more stable correspondences between query and support features. Meanwhile, the object-aware correlation module is introduced to construct attention vector, which is able to guide the network focusing on interested objects and suppress irrelevant background regions. The final segmentation mask is obtained by cross-scale guidance module reinforced by attention mechanism, which takes advantage of both similarity maps and attention vectors. In this way, the proposed network is capable of handling the diversity of object instances, which leads to robust segmentation results. Extensive qualitative and quantitative evaluations on PASCAL-5i, COCO-20i and FSS-1000 verify the effectiveness of the proposed method. The proposed method achieves favorable performance with much less trainable parameters (i.e., 4.2 M parameters with ResNet-50 backbone). [ABSTRACT FROM AUTHOR]