imageInpainting

In this project, I applied the algorithm described in the paper titled 'Context Encoders: Feature Learning by Inpainting.' The objective is to fill in missing portions of images using the Places365 dataset. Due to time constraints, I focused on a subset of this dataset, condensing it from 365 categories to just four: corn_field, farm, highway, and lake-natural, each comprising 5,000 images. These images were distributed into two sets, with 500 images per category in the validation set and 4,500 images in the training set. In total, our dataset consisted of 20,000 photos.