Learn to implement a Deep Convolutional Generative Adversarial Network (DCGAN) from scratch in this comprehensive tutorial video. Explore the DCGAN architecture based on convolutional neural networks and train it on the CelebA dataset, representing a significant advancement from simpler fully connected GANs. Dive into the DCGAN paper, understand the implementation details of both the discriminator and generator, and learn about proper weight initialization. Follow along as the model is trained on MNIST and then adapted for the CelebA dataset. Gain insights into the setup process, training procedures, and necessary modifications for different datasets.
Overview
Syllabus
- Introduction
- Quick Paper Recap
- Implementation of Discriminator
- Implementation of Generator
- Weight initialization and test model
- Setup of training
- Training on MNIST
- Modifications to CelebA dataset
- Training on CelebA and ending
Taught by
Aladdin Persson
