CANet(CANet: Class-Agnostic Segmentation Networks with Iterative Refinement and Attentive Few-Shot Learning)[1] consists of a two-branch dense comparison module which performs multi-level feature comparison, and an iterative optimization module which iteratively refines the predicted results. There are some details of reading and implementing it.

Contents

• Paper & Code & note
• Paper understanding
• Code understanding
• Note
• References

Paper & Code & note

Paper: CANet: Class-Agnostic Segmentation Networks with Iterative Refinement and Attentive Few-Shot Learning(CVPR 2019 paper)
Code: PyTorch
Note: Mendeley

Paper

Abstract

Problem Solution

Conceptual Understanding

Dense Comparison Module(DCM)

• Feature Extractor: features in lower layers often relate to low-level cues, e.g., edges and colors while features in higher layers relate to object-level concepts such as object categories. Instead, we focus on middle-level features that may constitute object parts shared by unseen classes.
• Dense Comparison: Here, they use global average pooling over the foreground area to squeeze the feature maps to a feature vector, global image features turn out to be useful in segmentation tasks. After obtained the global feature vector from the support set, they concatenate the vector with all spatial locations in the feature map generated by the query branch.

Iterative Optimization Module(IOM)

They propose to incorporate the predicted masks in a residual form: $$M_t = x+F(x,y_{t-1})$$
They use Atrous Spatial Pyramid Pooling module (ASPP) proposed to capture multi-scale information.

Core Conception

Experiments

Code

[Updating]

Note

• Well described the latent details in feature map and its process.

References

[1] Zhang C, Lin G, Liu F, et al. Canet: Class-agnostic segmentation networks with iterative refinement and attentive few-shot learning[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2019: 5217-5226.
[2] CaNet. https://github.com/icoz69/CaNet.