Hi,
I have a question regarding the training data. It seems that both the original and denoised images are of jpeg format, and thus the ground truth images have compression artifacts. This would affect the results as these artifacts are inconsistent across images. Also, it implies that the model is required to learn to reproduce such artifacts. Shouldn't the ground truth images be clean of these artifacts?
Best regards,
Evelyn
Hi evelynchee,
Can you please post your question here: https://ai-benchmark.net/index.php?threads/real-image-denoising-challenge.5/
Thanks,
Andrey