Image-Super-Resolution-via-Iterative-Refinement

Image-Super-Resolution-via-Iterative-Refinement copied to clipboard

inference a new test image is too slow？

1

Thanks for your sharing. The result is amazing. But the inference time is too long? about 4 minitus for one image?

tengshaofeng

Train a new model

1

Hey, so I use custom datasets to train a new model. I had a problem here: [issue43](https://github.com/Janspiry/Image-Super-Resolution-via-Iterative-Refinement/issues/43) Also I set json file as you did, total number iteration is 1000000....

yrsunn

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?

3

Specifaclly, I'm asking about the following section in the config file: ``` "unet": { "in_channel": 6, "out_channel": 3, "inner_channel": 64, "channel_multiplier": [?????????], "attn_res": [?????????], "res_blocks": ?????????, "dropout": ????????? }, ```

roey1rg

ddim sampling

23

Dear SR3 team, Thank you for sharing this great implementation. I try the colab for 64->512 model, it works very good. The ddpm sampling for 2k step is very slow,...

betterze

fix_uncondition_shape

1

When the condition is null, a tensor is passed in instead of the shape

xiaoyuan1996

Question about test set size

1

Question 1: I find SR try to sample image one by one, and the test set size is set to 50. But there are 30,000 images in test set, Should...

z562

Hue deviation and noise : 64*64 -> 256*256

28

Hello, First of all, thank you very much for the implementation you have done. it has saved us a lot of time. it's really a great job I am trying...

uminaty

Now, n_workers is 0 by default if the platform is win32. One is able to receive out of memory error if the memory of the disk of output dir does...

harnvo

Support for upscaling 16-bit grayscale images

12

Hi, I would like to upscale PNG images. The format for these images is 16-bits for 1 channel (single channel grayscale). I would ideally like to upscale them from 512x512...

VladStojDev

The result is too noisy

9

Thank you for the code. I have trained the sr3 model on the images of different resolution, like 16->128, 64->512, 256->1024 on ffhq and celebahq. The iters are 50k, and...

timerobin