In this project, I create several Generative Adversatial Network models for various Image to Image translation tasks. The models were first trained using tensorflow. Then, I use Flask to create a web-based implementation for uploading images and getting the augmented image.
After watching a video about CycleGAN on Two Minute Papers I was immediately fascinated with the idea of GANs. Mingled with a desire to improve my computer vision knowledge as well as my ability to read research papers and implement them, I decided to create this project. Not only have I grown more comfortable in reading research papers(which at one time looked scary math jiggerish!), I have also had a lot of fun with this project.All in all this project served as way to :
- Help me get comfortable with reading Research Papers
- Increase my Computer Vision knowledge
- Improve my tensorflow coding skills
- Image Deraining - Implemented Pix2Pix architecure (Paper)
- Animate Me - Trainined CycleGAN model (Paper) on the selfie to anime dataset used in U-GAT-IT paper (Github Repo)
- Image Superresolution - Used SRGAN architecture (Paper) and used the Div2k dataset in Tensorflow Datasets
-- Me frustated and tired of training GANs