Nir Ben Haim
David Levit
deep learning project - GAN which receives as input a letter in English and produces an image containing the letter as it is indicated in sign language (American Sign Language).
| File name | Purpose |
|---|---|
main.py |
general purpose main application |
model.py |
generator amd discriminator models |
train.py |
main application for training the model |
util.py |
contain print results function and loss function |
data.py |
data loader |
import_dataset.py |
download data |
| Arguments | Purpose |
|---|---|
--eval |
Model evaluation if true. Model training and evaluating if false |
--model_path |
Model save/load directory |
--data_path |
Dataset save/load directorys |
--chosen_letter |
Letter chosen to be printed, only when purely evaluating |
- In order to use the GAN without traning eval = True, model_path and data_path contains the wanted directory, and in order to get specific letter implement it in chosen_letter.
- In order to train the GAN eval = False, model_path and data_path contains the wanted directory.
- https://www.kaggle.com/datasets/datamunge/sign-language-mnist.
- https://github.com/soumith/ganhacks.
- Wasserstein GAN - Martin Arjovsky et AL, 2017
- Conditional Generative Adversarial Nets - Mehdi Mirza et AL, 2014