[CVPR 2022 Oral] Official repository for "MAXIM: Multi-Axis MLP for Image Processing". SOTA for denoising, deblurring, deraining, dehazing, and enhancement.
Hello everyone who participated in this project!
First of all I want to say that this is a wonderful project which compresses several of mine favorite low level image processing algorithms.
Nevertheless I've witnessed severe quality differences in the results of the deblurring inference.
I've downloaded and used your pretrained checkpoints based on the GoPro dataset.
Using your sample images in the deblurring folder working perfectly fine!
However using samples with motion blur which are encoded as JPG the quality drops massively.
Here are the examples:
ORIGINAL JPG:
INFERRED PNG:
INFERRED PNG BASED ON PNG:
INFERRED PNG BASED ON JPG:
There were also changes needed to convert the images prior before inference to PNG because otherwise the saved image will be corrupted.
Would somebody smarter than me please elaborate why this is happening and how I could fix it?
Hello everyone who participated in this project! First of all I want to say that this is a wonderful project which compresses several of mine favorite low level image processing algorithms.
Nevertheless I've witnessed severe quality differences in the results of the deblurring inference. I've downloaded and used your pretrained checkpoints based on the GoPro dataset. Using your sample images in the deblurring folder working perfectly fine! However using samples with motion blur which are encoded as JPG the quality drops massively. Here are the examples:
ORIGINAL JPG:
INFERRED PNG:
INFERRED PNG BASED ON PNG:
INFERRED PNG BASED ON JPG:
There were also changes needed to convert the images prior before inference to PNG because otherwise the saved image will be corrupted. Would somebody smarter than me please elaborate why this is happening and how I could fix it?
Thank you very much in advance!