niopeng / HyperRIM

HyperRIM: Hyper-Resolution Implicit Model

Project Page | Paper | Pre-trained Models

PyTorch implementation of HyperRIM: a conditional deep generative model that avoids mode collapse and can generate multiple outputs for the same input. The model is trained with Implicit Maximum Likelihood Estimation (IMLE). HyperRIM is able to:

• Increase the width and height of images by a factor of 16x
• Recover a plausible image from a badly compressed image
• Possibly do other things yet to be explored 😃

Installation

Please refer to this page.

Organization

The repository consists of the following components:

• code/: Code for training and testing the model
• experiments/: Directory for checkpoints and plots
• website/: Resources for the project page
• index.html: Project page