Generative Adversarial Networks for Real-robot Missions

Informations

• Author: Pin-Wei "David" Chen
• Thesis: Website, PDF

Development Environment

• Ubuntu 16.04
• Python3

Setup

$ cd
$ git clone https://github.com/championway/gan_rv_thesis

Semantic Segmentation

Train Semantic Segmentation on PST900 Dataset

• Download PST900 Dataset -- Reference: PST900: RGB-Thermal Calibration, Dataset and Segmentation Network

FCN-Pix2Pix

$ cd ~/gan_rv_thesis/Semantic_Segmentation/
$ jupyter-notebook PST_FCN_training.ipynb

UNet-Pix2Pix

$ cd ~/gan_rv_thesis/Semantic_Segmentation/
$ jupyter-notebook PST_UNet-Pix2Pix_training.ipynb

FCN-Pix2Pix

$ cd ~/gan_rv_thesis/Semantic_Segmentation/
$ jupyter-notebook PST_FCN-Pix2Pix_training.ipynb

Virtual Dataset from Simulation to Real

Download Sim2Real Dataset

• Sim2Real Sample Dataset (25.9MB)
• Sim2Real Dataset (19.8GB)

Training SSIM-CycleGAN

$ cd ~/gan_rv_thesis/Sim2Real/
$ python3 cyclegan_ssim.py

Create GAN Sim2Real Dataset

$ cd ~/gan_rv_thesis/Sim2Real/
$ jupyter-notebook Create_GAN_Sim2Real_Dataset.ipynb