Making Faces: Conditional Generation of Faces using GANs...
summary
Claims a conditional GAN that flips binary face attributes like smiling or bangs using spectral normalization and a versatile auxiliary classifier, implemented in Keras/TensorFlow and trained on CelebA. The author provides runnable Colab notebooks, a few‑GB cropped dataset, and code to condition on multi‑label vectors; training examples use cheap AWS GPUs (10 epochs shown) and generation runs in ~2 seconds. Results produce recognizable faces with artifacts but demonstrate attribute control and improved stability from the normalization.
AI-written summary · about
chapters (15)
- 0:00Introduction
- 5:00Deep learning fundamentals
- 13:05Convolutional classifiers
- 20:37From classification to generation
- 22:46How GANs work
- 33:39Learning image distributions
- 34:00Lipschitz continuity and spectral normalization
- 40:50Conditional GANs and multi-label data
- 43:04Generated face results
- 52:53Hands-on notebooks
- 1:00:05Convolutional classifier in Keras
- 1:06:45GAN generator and discriminator
- 1:15:21Training GAN step-by-step
- 1:21:11Adding spectral normalization and classifier
- 1:26:31Final results and Q&A
related talks
37mHarnessing the power of Generative Adversarial Networks (GANS) for supervised learning
03mBuilding artificial people: Endless possibilities and the dark side
43mHitchhiker’s Guide to Generative Adversarial Networks (GANs)
15mDrag Makeovers, via AI!
47mUnmasking Generative Adversarial Networks
29mThe Mustache Movement
40mOverview of Generative Adversarial Networks (GANs)
24mA Journey through Generative AI's Frontier Python Libraries