- Implemented and trained generative adversarial networks (GANs) for generating realistic cat face images
- Gained practical experience with generative models and recurrent neural networks using PyTorch
- Applied techniques to enhance image quality in GAN-generated images
- Modeled generator and discriminator network architectures based on DCGAN
- Implemented data augmentation using PyTorch’s built-in transforms
- Developed a dedicated debugging notebook, “GAN_debugging.ipynb,” featuring a smaller network trained on the MNIST dataset for validation and testing.
- Demonstrated attention to detail by verifying loss functions and training code on the smaller dataset before transitioning to the main project
- Completed the debugging GAN training on MNIST in approximately five minutes and the main project GAN training on the cat dataset in about one hour