Image Restoration(Inpainting) using DCGAN (Apr '20)
- Implemented based on the IEEE paper "Semantic Image Inpainting using Deep Generative Models
- Auto generated incomplete patches in an image.
- Designed a DCGAN network in python using tensorflow.
- Trained and tested using CelebA_HQ dataset
- Preprocessed and trained 10k images over the network for 8000 iterations using Cuda and CuDnn
- Obtained a reduced loss when compared to other conventional image restoration techniques like auto encoders.