Jianyi Wang
Jianyi Wang
Hi, if you want to adopt your own data, you may refer to [here](https://github.com/IceClear/StableSR/blob/main/basicsr/data/realesrgan_dataset.py#L18) and [here](https://github.com/IceClear/StableSR/blob/main/ldm/models/diffusion/ddpm.py#L1923) and modify the input to your desired input format.
Hi~ thanks for your interest of our work. I think you may need several days to obtain the stable performance, e.g., 5 days to one week.
Hi. The original resolution of this image is 352x224. I tested on 4x and did not find anything wrong.
> > Hi. The original resolution of this image is 352x224. I tested on 4x and did not find anything wrong. > > can you provide the original LR image?...
Hi. SNR is used as a metric to analyze our time-aware encoder, not a conditioned input. So we did not add the measuring code inside the repo.
1. No other solutions. Sry. 2. We actually finetune CFW individually, very similar to your description but first directly generating training data instead of loading stablesr_000117,ckpt to avoid unnecessary memory...
It seems that your CPU RAM is not enough. You need at least 18G for running.
Hi. I mean **CPU RAM**, not GPU :)
1. Correct. 2. We use x4 data for 128 to 512. 3. (1) We use all the test images. (2) Correct.