Triplet GAN

This is the code for Training Triplet Networks with GAN (arXiv).

The code is based on Improved Techniques with GAN (arXiv) (code)

Official Code Repo - Github

Installation

Use the package manager pip to install all required libraries.

pip install requirements.txt 

or

conda create --name <env> --file requirements.txt 

Pretraining

For pretraining the Triplet GAN we have used Improved Techniques with GAN for without the triplet loss function.

For Pretraining for MNIST data use pretrain.py --config_file configs/pretrain_gan.yml

Training Triplet GAN

For training the triplet GAN after pretraining use main.py --config_file configs/triplet_gan.yaml

Results

Pretraining

Triplet Traininig

Accuracy and mAP

	m = 16		m = 32
	N = 100	N = 200	N = 100	N = 200
Accuracy	0.9813	0.9819	0.9826	0.9833
mAP	0.9345	0.9501	0.9330	0.9473

Tsne Plots

Generator Output

Contributing

Pull requests are welcome. For major changes, please open an issue first to discuss what you would like to change.