Restormer
Restormer copied to clipboard
swz30
[CVPR 2022--Oral] Restormer: Efficient Transformer for High-Resolution Image Restoration. SOTA for motion deblurring, image deraining, denoising (Gaussian/real data), and defocus deblurring.
Great job! Have you tried working on image super-resolution task and how does it compare to existing work (e.g. SwinIR) ?
I wrote code in Python to test PSNR and SSIM, why is the PSNR and SSIM I tested in Python lower than those in the paper? PSNR is about 1.2dB...
When performing defocus blur training, do you pre-crop the image to 10 patches？
Hi.Thank you so much for your open source work. When I trained motion_deblur, I found that the effect in the paper could not be achieved. 1. I followed the dependency...
Hello, I would like to ask how the global features are obtained implicitly in restormer when calculating transposed attention on channel dimension?
I'm really impressed in inferencing on HuggingFace. I had run Motion Deblur task on my own CPU(i7) and got ~50s inference runtime. May I convert from pth model to onnx...
For DnCNN and FFDNet, the images in Kodak24 are center cropped into 500x500 for testing (Those images are from https://github.com/cszn/FFDNet/tree/master/testsets/Kodak24). However, the Kodak24 testing set provided in `download_data.py` does not...
Thanks for your great works which inspire me so much. In your paper(page8), compared to swinIR, your model has fewer Flops and runs 13x faster. And it comes true in...
Hi, I've been trying to train the deraining model on your datasets for the last one week, but every time I run the train.sh script, the data loaders get created...
Hello. In ''Restormer-main\basicsr\data\paired_image_dataset.py''  These functions are not defined. Why?
