Open catch-n-release opened 5 years ago
In that case, it is better to create dedicated dataset for your task to train standard denoising model instead of using noise2noise framework as you mentioned (can also get clean counter-parts of the same for noisy-clean pair). For this, this implementation is not suitable so please use the other implementations of standard denosing models. I think there would be difficulty in creating noisy-clean pairs; calibrating them in pixel level. 100> pairs would be required.
I have document scans to be cleaned. How do I go about training a new model for it. I can also get clean counter-parts of the same for noisy-clean pair, but dont know how to initiate training for them. And how many of them would I need for training to get significant results?