4 Categories
There are mainly 4 Image generation methods.
1. GAN
Train a GENERATOR that generates an image from “z”, and have a DISCRIMINATOR discriminate whether the generated image is real or not. The GENERATOR tries to learn to be able to generate an image that is not distinguishable from a real image.
2. VAE
Encode an input while constraining it within a normal distribution, and decode it back from “z” which is sampled from THAT distribution. The model tries to learn the distribution so that they can generate images with additional conditions.
3. Flow-based
After applying the input to a function f(x), use an inverse function of f(x) to generate the image.
4. Diffusion-based
Define a Markov chain of diffusion steps to slowly add random noise to data and then learn to reverse the diffusion process to construct desired data samples from the noise.
Reference:What are Diffusion Models?
