search

dliu5812 / PDAM

【CVPR 2020 & IEEE Transactions on Medical Imaging 2021】PDAM: Unsupervised Domain Adaptive Instance Segmentation in Microscopy Images
MIT License
21 stars 4 forks source link

CycleGAN #6

Open Xujingkk opened 4 months ago

Xujingkk commented 4 months ago

Hi, Dongnan Liu When using CycleGAN to generate Kumar-like style images from BBBC images, it seems that the results are learned in reverse, although the style is learned. That is, the region of cell nuclei in BBBC, is shown as background in the generated data, while the region of background in BBBC, has cell nuclei in the generated data.

I'm not sure what the problem is and would appreciate your help.

I found that the code uses "img = cv2.imread(file_path, 1)" to read in the data without converting it to RGB channels, does this have any effect?

Xujingkk commented 4 months ago

Hi Dongnan, According to your reply, we have tried image inverse, the code for inverse is as follows, the result is as shown in the picture, but nothing seems to work, so we have to ask you for help again!

  1. You replied earlier that the image should be inverted to a white background and black nuclei, the image we inverted seems to have gray nuclei, is that the problem here?
  2. May I ask if we have written the code wrong? Or is there some other unnoticed data processing? 967c2e5b8ca7531a5755d78f56713a3 f35edea0d007c731950f60f0a82bcf7