search

daifeng2016 / End-to-end-CD-for-VHR-satellite-image

The project aims to contribute to geoscience community
65 stars 22 forks source link

the fuse method #7

Open scscualex opened 4 years ago

scscualex commented 4 years ago

Hello,I'm quite fused about the fuse method. It seems that the method in your article and code have some differences. In article, the final change map is obtained from for γ. In code, the final map is obtained from feature maps could you please confirm it?

daifeng2016 commented 4 years ago

Hi, the output of each scale (in our article it is 4) are fused to generate a new output, then a sigmoid layer is followed. In total, five outputs are generated during training, while only the fused output is utilized during test phase. The training code is consistent with that. Maybe you are confused with deep supervison.

Daisy-7 commented 3 years ago

Hi, the output of each scale (in our article it is 4) are fused to generate a new output, then a sigmoid layer is followed. In total, five outputs are generated during training, while only the fused output is utilized during test phase. The training code is consistent with that. Maybe you are confused with deep supervison.

Hi! I have a question about the No.5 output result. In your paper, it generated by the following equation: image but in your code, it generated by the following code: image

image is not equal to the conv1_2 , but equal to the image

Is there any conflict in your paper and code? Which one is correct?