Generative Image Models

Generative Image Models

1. Texture Synthesis by Non-parametric Sampling (Alexei Efros)

2. Scene Completion Using Millions of Photographs (Alexei Efros)

   • initially, it figures out a set of similar images from a extreme large dataset
   • then, images regions are molten using Poison Blending
     • Poisson Image Editing (Microsoft research UK)

3. variety of deep learning approaches

   • Deep Generative Image Models using a Laplacian Pyramid of Adversarial Networks (Facebook AI Research)
     • fundamental problem whithin computer vision
     • how?
       • building a series of Generative Models, each of which captures image structure at a particular scale of a Laplacian pyramid
         • The Laplacian Pyramid as a Compact Image Code (Peter Burt)
           • creates an image codification to remove the high correlation between neighboor pixels
     • why?
       • Deep learning approaches have proven highly effective at discriminative tasks in vision
         • the same level of success has not been obtained for generative tasks, despite numerous efforts
   • useful architectures
     • Restricted Boltzman Machine
     • Deep Boltzman Machines
     • Denoising Auto-encoders
   • proved to work fine only for very simple datasets

4. Learning to Generate Chairs with Convolutional Neural Networks (Alexey Dosovitskiy)

5. DRAW: A Recurrent Neural Network For Image Generation (Google DeepMind)

6. Generative Image Modeling Using Spatial LSTMs (Theis)