Clothes Color Segmentation Based on Channel Attention Module

This paper studies a color segmentation network for mixed-color clothes based on BiSeNet V2. Aiming at the color category segmentation error in BiSeNet V2, the CAM_BiSeNet V2 network model is designed, and the channel attention module is added before the context embedding block in the semantic branch of BiSeNet V2. The selection of channels is integrated to achieve a better representation of the features in the channel dimension. The experimental results show that this algorithm can effectively improve the segmentation accuracy of the model, and the segmentation accuracy rate is increased by 10.61% (91.19% vs 80.58
^%
)