-
Notifications
You must be signed in to change notification settings - Fork 188
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
The right way of inference #17
Comments
Yes, it will output different results due to Guassian random noise. |
Thanks! |
I have encountered this same problem. In my experience, colorization in complex scenes requires a lot of data, unlike faces which can be trained well with a small amount of data. But I haven't had time to verify it yet. |
Can you possibly share the inference script? Thank you! |
Hi @AdelZakirovRZ |
Hey guys, so about the inference example.
|
I tried the above code with some changes to it. I was able to get it to spit out an image, but the image is black. Any suggestions on what my error is?
|
Oh also, I did normalize the script by changing line 64 to:
But it is only spitting out very strange images.. |
Hey, the output is in [-1; 1] - I think you miss that part |
So then should it be this? I'm still not getting anything that looks right:
|
Ok going through your code again, I had to hunt a bit but I found in Util, your tensor2img function. I ran my prediction through that and it generated this: Is this correct? The original image was this: ./misc/image/Mask_Places365_test_00144085.jpg |
hey @richard-schwab
that part does not look right. Also, FYI - I am not the author of the code |
Thanks for sharing. I think this test code is correct. Did you get better experimental results later? |
hey, Code of @AdelZakirovRZ is used for colorization where y_cond is grayscale image. In inpainting task, y_cond is masked image contains noises, like You need to munually replace blank masked regions with random noises. |
Hey @Janspiry |
Hey, thanks for your contributions for this good question and I will keep it for others to add more details |
@AdelZakirovRZ Hello, Could you share your code for colorization using pytorch-lightning? I am also using pl, but it is difficult to translate this repo to pl. And I am also having problem in reproduction of colorization results, too. |
Hi, may I ask if this has happened to you? I used CelebaA to train the model, but just get some strange result strange. It seems to color a grayscale image with the same color. The same situation occurs in the sketch coloring task. |
@Nyoko74 , I did not, maybe you need to adjust learning rate and train more iterations. |
Thank you, I will try it. |
Hi @Nyoko74, I am having this same issue. Did you manage to fix it? |
Hey @AdelZakirovRZ |
Hey @Nyoko74 Did you solve this problem? And actually in the results shared by @AdelZakirovRZ , I find that the image in the bottom right, the man's hand is blue. I'm wonder that this problem could be solved by increasing training time? If not, how to solve this problem? |
In my inpainting case, during the inference only the In the
|
Hi everyone, for inpainting task I took a deeper look into the
Input: |
Hey guys,
can you please direct me on how to properly inference the trained model?
I wrote a small script for it, but not sure that I am doing everything right.
One of the conceptual questions for me is the robustness of the results. I tried to train the colorization model, and during inference it gives me different results for the same image. Is that because of randomness in a noise scheduling? Or is it something else?
Thanks in advance!
The text was updated successfully, but these errors were encountered: