Issue in Inference Mode about dehaze-gan HOT 3 CLOSED

Hey @jeffin07,

In the README file, I had mentioned the following:

3. For recreating the results reported in the paper, use the repository legacy (for more details refer below). 
This repository is the refactored version of the final model, but it uses newer versions of some TensorFlow operations. 
Those operations are not available in the old saved checkpoints

The checkpoint files should work with the code in the legacy folder. Do try that out and let me know if it works.

from dehaze-gan.

@thatbrguy I've tried replicate.py in the legacy folder but in replicate.py we can't give our own test images right, the images are taken from A.npy so that's why I tried main.py file. replicate.py runs fine. The problem with that is I cannot use my own test images

from dehaze-gan.

Hey, sorry for the very late reply. As explained in the README file (and in the previous comment), the checkpoint files may not be compatible with the code outside the legacy folder.

If you want to perform inference on custom images using the checkpoint files, an easy way to do that is to modify replicate.py such that it reads in images rather those .npy files. Apologies for the inconvenience.

from dehaze-gan.