BasketballGAN
Generate the ghosting defensive strategies given offensive sketch.
Paper | CGVLab
Video | Supplemental
BasketballGAN: Generating Basketball Play Simulation through Sketching
Hsin-Ying Hsieh1, Chieh-Yu Chen2, Yu-Shuen Wang1 and Jung-Hong Chuang1
1National Chiao Tung University,
2NVIDIA Corporation
Accepted paper in ACMMM 2019.
Prerequisites
- OS: Linux
- NVIDIA Dokcer
- NVIDIA NGC Tensorflow Docker Image
- NVIDIA GPU (V100 16GB)
Getting Stated
~$ git clone https://github.com/chychen/BasketballGAN.git
~$ cd BasketballGAN
BasketballGAN$ docker login nvcr.io
BasketballGAN$ docker pull nvcr.io/nvidia/tensorflow:19.06-py2
BasketballGAN$ docker run --runtime=nvidia -it --rm -v $PWD:$PWD --net host nvcr.io/nvidia/tensorflow:19.06-py2 bash
root@c63207c81408:~/BasketballGAN$ apt update
root@c63207c81408:~/BasketballGAN$ apt install ffmpegDownload Dataset
- create 'data' folder
- save dataset under folder 'data'
BasketballGAN$ mkdir dataTraining
BasketballGAN$ cd src
BasketballGAN/src$ python Train_Triple.py --folder_path='tmp' --data_path='data'Logs/Samples/Checkpoints
- "BasketballGAN/src/tmp/Log": training summary for tensorboard.
- "BasketballGAN/src/tmp/Samples": generated videos sampled on different epoches.
- "BasketballGAN/src/tmp/Checkpoints": tensorflow checkpoints on different iterations.Monitoring
- Sampled Videos
- Using Simple HTTP Server to monitor sampled videos while training.
- Simple HTTP Server (http://127.0.0.1:8000)
BasketballGAN/src$ python -m http.server 8000- Training Logs
BasketballGAN/src$ tensorboard --logdir='tmp/Log'Public Relations
Citation
If you find this useful for your research, please use the following.
@article{hsieh2019basketballgan,
title={BasketballGAN: Generating Basketball Play Simulation Through Sketching},
author={Hsieh, Hsin-Ying and Chen, Chieh-Yu and Wang, Yu-Shuen and Chuang, Jung-Hong},
journal={arXiv preprint arXiv:1909.07088},
year={2019}
}