DiffBIR icon indicating copy to clipboard operation
DiffBIR copied to clipboard

Official codes of DiffBIR: Towards Blind Image Restoration with Generative Diffusion Prior

Results 86 DiffBIR issues
Sort by recently updated
recently updated
newest added

当我输入 `python gradio_diffbir.py --ckpt weights/general_full_v1.ckpt --config configs/model/cldm.yaml --reload_swinir --swinir_ckpt weights/general_swinir_v1.ckpt --device cuda` 会报错: `Traceback (most recent call last): File "C:\GitCloneProject\DiffBIR\gradio_diffbir.py", line 148, in input_image = gr.Image(source="upload", type="pil") File "C:\Anaconda3\envs\diffbir\lib\site-packages\gradio\component_meta.py", line...

Hello, how much is the minimum required vram for the stage2 training process? I train it on an A100 80G with batch size 16, image size 512*1024, and it nearly...

OSError: /usr/local/lib/python3.10/dist-packages/torchaudio/lib/libtorchaudio.so: undefined symbol: _ZN3c104cuda9SetDeviceEi

It is in the Powershsell and after i use the code: pip install -r requirements.txt, i have two errors ERROR: Could not find a version that satisfies the requirement triton==2.0.0...

感谢您的工作,我在阅读源码过程中发现在inference时对图像进行了pad,然后在生成结果中再移除了pad的部分,请问这个操作是用来处理什么问题的。我测试了几个样例,即使移除了pad,结果好像也不受影响。我在其他论文的相关程序中也看到会有pad的操作,所以我想咨询一下您,这里的pad是用来避免一些可能出现的问题吗?

I think maybe a better existing restoration module can be used to improve the performance?

Hi, great work! The paper mentioned that "finetune the diffusion model for 25k iterations with a batch size of 192", however, I see that batch size is 16 per single...

Now it only support to 1024, any way to higher resolution?

I try to train the stage 1 model under codeformer degradation for CT-imagery SR using grayscale CT images. Specifying in_chans = 1 in swinir.yaml configuration file gives me _"Given groups=1,...