This is an implementation of the 2018 paper Noise2Noise and 2023 paper Zero-Shot Noise2Noise, for our AIML course project. This was done in collaboration with Swarnava Chakraborty, Vipul Tejwani and Omar Muhammad.
- Report:
N2N_Review_Report.pdf - Presentation:
N2N_presentation.pdf
The dataset folders have been git ignored. Find the datasets here:
- Admiral : 0.jpg to 490.jpg for train_images and rest for test_images.
- Imagenet-256x256 : Randomly selected 500 images from the dataset for train_images_500 and 50 images for test_images_50. The exact images can be found here. We used all 24 images of the kodak dataset and 26 random images from BSD300 to make the test_images_50. (Cropped to make 256x256)