search

jakeret / tf_unet

Generic U-Net Tensorflow implementation for image segmentation
GNU General Public License v3.0
1.9k stars 748 forks source link

Why do I change "VALID" to "SAME"? Why is the effect of network prediction very poor? Are there pixels not aligned? But I did not find this problem. #275

Open yeyuxmf opened 5 years ago

yeyuxmf commented 5 years ago

Why do I change "VALID" to "SAME"? Why is the effect of network prediction very poor? Are there pixels not aligned? But I did not find this problem.

yeyuxmf commented 5 years ago

@jakeret self.store_prediction()--->util2D.combine_img_prediction2D() -->torgb2D(data.reshape(-1, ny, ch).
In the to_rgb2D function. This function must have a deep copy of "data", otherwise the value of "data" will be changed, and the next test will produce errors.