Testing code does not work well about a-little-bit-more

[abhijithpunnappurath]

The output images are being saved as 16-bit images. See this line of testing code. In this case above, the 8 LSB bits hold the result, and the 8 MSB bits are zeros. This is why you are seeing black images.

Can you please try the following:

1. Try a 16-bit image instead of 8-bit. You should not be seeing a black image now as output.
   python test.py --set_names Sintel_sample --type_8_or_16 1 --quant 6 --quant_end 16 --dep 4 --save_result 1

2. For your 8-bit recovery test case above, convert the output 16-bit images to 8-bit as
   img = cv2.imread('kodim01_output.png',-1).astype(np.uint8)
cv2.imwrite('kodim01_output_8bit.png',img)
   The images should be correctly visualized now.

3. I am not sure why the PSNR and SSIM are not being calculated correctly. Can you confirm that your conda environment is using scikit-image=0.15.0?

[xunmeibuyue]

@abhijithpunnappurath Thanks very much for the anwser!

I find it works well on Google colab, as well as when I just use a single GPU.

This is really a beautiful work, simple but effective. I would like to learn from it!

[abhijithpunnappurath]

I see, thanks for letting me know.

Very happy to hear that you found our work interesting!
Thank you.