RL Final Project (2020-1)

This project is based on the following 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}
}

Link for the original paper: https://arxiv.org/abs/1904.12304
Refer to the report "Reinforcement_learning_for_point_cloud_completion.pdf" in this repo for detailed explanation.

Requirements:

• conda create -n <env_name> --file requirements_conda.txt python=3.6
• pip install -r requirements_pip.txt
• conda install pytorch==1.2.0 torchvision==0.4.0 cudatoolkit=XX.X -c pytorch (XX.X: cuda version)
• mkdir data && ln -s <directory of train,test> shape_net_core_uniform_samples_2048_split

Steps:

1. Download data from https://github.com/optas/latent_3d_points.
2. Process Data with Processdata2.m to get complete point cloud (not incomplete!!)
3. Train the autoencoder using main.py and save the model
   • link data paths (train, test). see #TODO
   • open visdom server with port 8102 python -m visdom.server -port 8102
4. Generate GFV using pretrained AE using GFVgen.py and store data
   • link pretrained model & train data path. see #TODO
5. Train GAN on the generated GFV data by by going into the GAN folder (trainer.py) and save model
6. Train RL by using pre-trained GAN and AE by running trainRL.py
   • First, process data with Processdata.m to get incomplete point cloud
   • link data paths (incomplete training dataset). see #TODO in RL_params.py
7. Test with Incomplete data by running testRL.py
   • link pretrained RL network paths

Credits:

1. https://github.com/optas/latent_3d_points
2. https://github.com/heykeetae/Self-Attention-GAN
3. https://github.com/lijx10/SO-Net (for chamfer distance)
4. https://github.com/sfujim/TD3
5. https://spinningup.openai.com/en/latest/algorithms/sac.html
6. https://github.com/wentaoyuan/pcn (for visualization)