Generative-Adversarial-Networks-GANs-Specialization-Coursera

Overview

Prerequisites:

• Basic Mathematics
• Deep learning and CNN concepts (Andrew deep learning specialization course 4 is recommended)
• Segmentation and U-net architecture
• PyTorch

Covered concepts:

• GANs Basics (discriminator and generator)
• Loss function impact (BCE, and wasserstein) W-Gan-GP, SN-Gan, and Protein-Gan.
• Conditional GANs Vs Controllable Generation.
• GANs evaluation metric (FId, VGG-Loss, Inception score, SSIM, Precision, Recall, and F1 Score)
• Gans weakness and different generative models such as variational autoencoders (VAEs)
• StyleGan, and BigGan components and architecture methodology
• Gans application:
  • Data augmentation and privacy concerns.
  • Image-to-Image Translation models such as Pix2Pix, SRGan, PatchGan, and GauGan.
  • Unpaired image translation models such as CycleGan.

This specialization consists of three courses as follows:

• Build Basic Generative Adversarial Networks (GANs)

  • Week 1: Intro to GANs.
  • Week 2: Deep Convolutional GANs .
  • Week 3: W-GAN with gredient penalty.
  • Week 4: Conditional GAN and Controllable Generation.

• Build Better Generative Adversarial Networks (GANs)

  • Week 1: Evaluation methods of GANs such as FID and IS.
  • Week 2: GANs disadvantages and Bias.
  • Week 3: StyleGAN architecture.

• Apply Generative Adversarial Networks (GANs)

  • Week 1: GANs Applications such as Augmentation.
  • Week 2: Image-to-image translation with Pix2Pix and U-Net architecture.
  • Week 3: Unpaired translation with CycleGAN.

Repo Content

• Lecture Notes
• Assignments
• Extra Labs
• All papers discussed in this specialization