CCNet:Criss-Cross Attention for Semantic Segmentation
1 Introduction
We propose a novel criss-cross attention module in this work,which can be leveraged to capture contextual information from long-range dependencis in a more efficent and effective way.
We propose a CCNet by taking advantanges of two recurrent criss-cross attention modules, achieving leading performance on segmrntation-based benchmarks,including Cityscapes,ADE20K and MSCOCO.
2. Related work
Semantic segmentation
Attention model
3. Approach
3.1 Overall
3.2 Criss-cross Attention
3.3 Recurrent criss-cross Attention
4 Experiments
问题在于这种对底层message passing的改变是否符合图像自身的结构特征，文章中采用的策略的隐含的假定是，相关性更多出现在cross上，然后通过循环可以传播这种相关性，就是把欧氏距离变成city block distance来计算。