This repository is the official PyTorch implementation of "Mosaic Convolution-Attention Network for Demosaicing Multispectral Filter Array Images" (TCI 2021)
I am facing an issue when finetuning your model over the ARAD-1K dataset. In the training process, I am having great results (about 80 psnr each epoch). But when I take theses new weights to the "eval.py" I am having terrible results (around 5 psnr).
Do you have any idea? also am I using the good python script for testing ?