This repository is a successor to the original CVPR 2019 Paper : RL-GAN-Net: A Reinforcement Learning Agent Controlled GAN Network for Real-Time Point Cloud Shape Completion.

We propose a novel approach of replacing backpropogation of an Autoencoder by a Reinforcement Learning agent, for Monocular Depth Estimation from RGB Images.

• Python 3.6
• Linux / Windows
• Anaconda (optional)

The packages for the project are listed in requirements_conda.txt and requirements_pip.txt files. Only install the ones needed or you can clone the whole environment.

1. Download depth data from Link
2. Process Depth Images to remove noise
3. Train the MobileNetV2 model on depth data using trainMobileNetV2.ipynb
4. Train RL by using pre-trained MobileNetV2 model, by running trainRL.py
5. Test with new RGB images by running testRL.py

1. https://github.com/sfujim/TD3

If you use this work for your projects, please take the time to cite the original CVPR paper:

@InProceedings{Sarmad_2019_CVPR,
author = {Sarmad, Muhammad and Lee, Hyunjoo Jenny and Kim, Young Min},
title = {RL-GAN-Net: A Reinforcement Learning Agent Controlled GAN Network for Real-Time Point Cloud Shape Completion},
booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
month = {June},
year = {2019}
}