search

gndlwch2w / msvm-unet

The official codes for the work "MSVM-UNet: Multi-Scale Vision Mamba UNet for Medical Image Segmentation".
https://arxiv.org/abs/2408.13735
29 stars 2 forks source link

Four diagonal directions to aggregate information #2

Closed Assioncreed closed 1 month ago

Assioncreed commented 1 month ago

Hi, I would like to know where exactly is the code that implements the aggregation of information from the four diagonal directions proposed in the paper?

gndlwch2w commented 1 month ago

Thank you for your attention to our work. In the paper, we mention the introduction of convolution operations in the feed-forward network to aggregate information from four diagonal directions. Specifically, for any feature point on the feature map, the convolution operation is able to aggregate features from all directions around it (including the diagonal directions). Through multiple stacked layers, the receptive field can be expanded, enabling the aggregation of diagonal features across the entire feature map. Moreover, to handle organs of various sizes and shapes, we employ multi-scale parallel convolutions on top of this. For specific code implementation, please refer to: https://github.com/gndlwch2w/msvm-unet/blob/abb1a8ed5cf6f211a9c7b375feb7a40db0775960/model/decoder.py#L13

Assioncreed commented 1 month ago

Thank you very much. I understand.