princeton-computational-imaging / dirtypixels Goto Github PK

I'm sorry to bother you．you change the data range to [-1.0, 1.0],and add noisy directly in the file simulate_raw_images.py.why？is there any problem?

When I look at your sensor_model.py file, I find that the raw image is actually a rgb image generated by a mask. Honestly, this toy quality image leads me to think that the experimental results of the paper are not fair.

To save time, for people who plan to use this dataset to validate raw images, I suggest you look elsewhere.

I am sorry to bother you. When I retrain the network (isp + mobilenet) following your advice, I got loss=nan from the first step. I didn't modify your code at all.Do you have any advice to make this code work?

How can I trained your model on SID dataset. I see you mentioned this experiment on section 5.3 but dont find any guide on how to run it in README