Learning to Generate Realistic Noisy Images via Pixel-level Noise-aware Adversarial Training

Existing deep learning real denoising methods require a large amount of noisy-clean image pairs for supervision. Nonetheless, capturing a real noisy-clean dataset is an unacceptable expensive and cumbersome procedure.