Tiny Video GAN

Key Insight

This project trains a small video GAN on face crops from UCF-101 so you can watch mode collapse happen with your own eyes — the generator discovers a couple of clips that reliably fool the discriminator and then keeps producing only those, ignoring the variety in the real data. GANs extend awkwardly to video because the discriminator must judge not just whether each frame looks real but whether the motion between frames does, which makes the adversarial game even less stable than it is for images. Living through that instability is the point: it is the concrete reason the field abandoned GANs for diffusion once diffusion proved both sharper and far more stable to train.