Contest4 - StackGAN++

• StackGAN++ implemented by Tensorflow

  • CS565600 Deep Learning Contest4
  • For inclass contest

• Framework

<img src="framework.jpg" width="900px" height="350px"></img>

• Installation Guide

  • pip install -r requirement.txt
  • CUDA 8.0, cudnn 6.0 and Tensorflow 1.4
  • Download and unzip dataset besides main.py
  • Download and unzip testing tool besides main.py
  • Install and setup config kaggle cli
  • mkdir inference besides main.py

• Training

  • python main.py
  • If you occur memory error when training, you could resize the batch_size in main.py line 19

• Evaluate and Submit to Kaggle

  • cd testing && python inception_score.py ../inference ../score.csv && kg submit ../score.csv
  • If you have 2 GPUs, you can uncomment main.py line 806 to evaluate in loop. At the same time you should specified the testing/inception_score.py to run on different GPU with enough memory space.

• Reference

  • StackGAN++ PyTorch paper code