/MSG-U-Net

High resolution image to image translation using multi-scale gradients

Primary LanguageJupyter Notebook

Efficient High-Resolution Image-to-Image Translation using Multi-Scale Gradient U-Net

This repo provides the code implementation of the paper Efficient High-Resolution Image-to-Image Translation using Multi-Scale Gradient U-Net. More specificallly to train on a specific dataset, include that dataset in the datasets folder.

Requirements

  1. python 3.6
  2. Tensorflow 1.12
  3. Keras
  4. Cuda toolkit 9.0
  5. Cudnn

Instructions to run the code

  1. Download any of the datasets mentioned in the paper Link to Paper
  2. Make sure you have all the requirements installed, then after run train.ipynb
  3. To replicate the results in the paper, run test.ipynb by making appropriate changes by using weights produced by train.ipynb
  4. Uplift the poses from 2d to 3d from the below link 3D Pose estimation Baseline
  5. Place output of the step 4 in unity assets folder and run it to watch the animation