Question on denoising training data generation

[Calvin-Pang]

Hi @SkyeLu @IVRL,

Thanks for sharing the great work. When running the code, I have a question on denoising training data generation. I noticed the code below in the NoiseImageGenerationDataset class:

    if len(in_paths) >= 20:
        print('Number of clean images is larger than 20!!!')
        sys.exit()
    else:
        clean_img_names = sid_train_clean_info[str(iso_value) + '_' + str(ratio_value)]
        all_clean_img_names = os.listdir(os.path.join(data_folder, 'Sony/long'))
        clean_pool = [item for item in all_clean_img_names if item not in clean_img_names]
        clean_selected = random.sample(clean_pool, 30 - len(in_paths))

Does it mean that for some ISO&ratio pairs, the denoising training will face fewer or even no data samples? For example, if in a specific ISO&ratio combination, there are already more than 20 real data points, it will just 'sys.exit()' and be excluded in the noise generation and following denoising training?
I would appreciate it a lot if you could help clarify it!