Liangwei Jiang comments

Results 53 comments of


                                            Liangwei Jiang

Hue deviation and noise : 6464 -> 256256

Hey, thanks for your contributions for this good question and I will keep it for others to add more details

What params should I use for the 128 to 1024 task, and for the 512 to 1024 task? How should I choose the Unet architecture?

Feel free to reopen the issue if there is any question.

ddim sampling

Sorry for late reply, I am only reproducing SR3 and have no plans to use DDIM.

When the condition is null, x_in = (b, c, h, w) as show in https://github.com/Janspiry/Image-Super-Resolution-via-Iterative-Refinement/blob/52af4cb2ec0c09ab7de3ca5ffb5c77432dce0fc6/model/sr3_modules/diffusion.py#L206

Question about test set size

1. Theoretically, you need to test all images, which wil takes a lot of time. 2. You need to first make inferences with the model, save the output, and then...

Train a new model

I think this is not normal. Could you put up the log and configuration?

inference a new test image is too slow？

Feel free to reopen the issue if there is any question.

Unet input channels number for colorization

I think this is the better choice, the 6 channels are just to follow the paper elaboration. ![image](https://user-images.githubusercontent.com/32922494/175776459-1e0c4209-53dc-4cf6-8fcc-cb5a93ea9a68.png)

Unet input channels number for colorization

> I see, thank you for your comment! Would you mind commenting on other possible approaches for conditioning that you think are valid. The paper is not really clear on...

Unet input channels number for colorization

Feel free to reopen the issue if there is any question.