Skip to content

GAN-based Remastering of Video Recordings of old StarCraft game

Notifications You must be signed in to change notification settings

SreeranjaniD/RESGAN

Repository files navigation

Modified Pix2Pix architecture for StarCraft video remastering in PyTorch

Contributors: Sreeranjani Didugu, Sandeep Kumar Remani

We would like to sincerely thank the authors of Pix2Pix for using their open-sourced source code in PyTorch.

The Pix2Pix code was written by Jun-Yan Zhu and Taesung Park, and supported by Tongzhou Wang.

Kindly refer, original Pix2Pix page for installing packages.

Our Contribution

Data Preprocessing:

The snapImage folder contains the automated python script to capture images while playing the game video from the StarCraft and the images are saved using unqiue file name each time it is. Kindly save these images in different set folder each time is it run over the reply video. Once, we have these images collected over several videos stored in the different set folder we can combine the SD and corresponding cartoon image together by creating two new subdirectories named cartoon and SD folder inside the combineData folder. These subdirectories should each have their own subdirectories train, val, test, etc. In ./combineImage/cartoon/train, put training images. In ./combineImage/SD/train, put the corresponding images. Repeat same for other data splits (val, test, etc).

Corresponding images in a pair {A,B} must be the same size and have the same filename, e.g., ./combineImage/cartoon/train/1.jpg is considered to correspond to /combineImage/SD/train/1.jpg.

Once the data is formatted this way, call:

python datasets/combine_A_and_B.py --fold_A ./combineImage/cartoon/train --fold_B ./combineImage/SD/train --fold_AB ./combineImage

Create a new folder inside the dataset folder named starcraft and paste all the train,test and val folder inside the starcraft folder.

Training

To train the model,

python train.py --dataroot ./ dataset/starcraft --name experimentname --model pix2pix --direction AtoB --no_flip --checkpoints_dir trained_model --netG resnet_9blocks --ngf 256 --no_dropout --init_type kaiming --beta1 0.8

To test the model,

python test.py --dataroot ./dataset/starcraft/ --model pix2pix --name experimentname.8  --ngf 256 --direction AtoB  --netG resnet_9blocks --beta1 0.8 --checkpoints_dir trained_model --init_type kaiming --no_dropout 

Experiment analysis

Once we have the train set and test set results, put the log file and the both images folder inside the testResults folder. Rename the test set result folder as testimage and train set result folder is trainimage. Run the following code, we generates the training loss curve, average L1 loss and excel file containing the L1 loss value and its recorreponding image name.

python testResults.py 

Testing on Youtube video

To convert Youtube video to images change the video file name inside the code and run

python video2frame.py

To convert the generated fake cartoon image back to video, put the fake images inside a folder name testResult and run,

python img2video.py 

NOTE: The research is still under work and the end to end model will be released soon.

Reference citation:

@inproceedings{CycleGAN2017,
  title={Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networkss},
  author={Zhu, Jun-Yan and Park, Taesung and Isola, Phillip and Efros, Alexei A},
  booktitle={Computer Vision (ICCV), 2017 IEEE International Conference on},
  year={2017}
}

About

GAN-based Remastering of Video Recordings of old StarCraft game

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published