Closed JohnCraigPublic closed 4 years ago
OK, sorry for the dumb questions - I see that pixel values are being expressed as floats, so as long as I watch the details of the image manipulations, it works naturally with 16-bit-per-pixel images. I think its working for me. I did have to use OpenCV to write out a 16-bit TIFF, but that's fine. Thanks. Closing....
For training all images are converted to float-32bit images. This is necessary since most GPUs only work on float32 type.
I have single channel (grayscale) images that have 16-bits-per-pixel. What do you think about using your technique to denoise such images? Will it be possible for me to use your technique and retain 16-bits-per-pixel? I don't want to smash the image down to 8-bits-per-pixel.