The generative adversarial networks seem to work very effectively for training generative deep neural networks. Our proposed generative adversarial model that works on 36 class, each having 10,000 images, generate the Nepali Handwritten Letters that are similar to the real-life data-set of size 360,000 images. This research paper shows how a generative adversarial framework is used to tell the computer to generate Nepali Handwritten Letter. Our objective is to synthesis of Nepali Handwritten Letter using a learning framework called generative adversarial networks. We construct discriminator networks and generator networks both of five convolution layers and choose the activation function such that generator networks generate the image and discriminator networks check the generated image is how similar to real-life image. To measure the quantitative performance, we use the Frechet inception distance (FID) methodology. The FID value we find of 18 random samples, generated by our networks is 38413677.145. For a qualitative measure of our model let the reader judge the quality of the image generated by our generator model.
Following things must be required to use and learn more about the project.
- python 3
- PyTorch
- pickle
- matplotlib
- numpy
- Frechet inception distance(FID)
- visit this repo to calculate FID distance of our project
- Dataset link download from here
- contact: basantbhandari2074@gmail.com
Use the git to clone the project.
git clone https://github.com/basantbhandari/Generation-Of-Nepali-Hand-written-letter-using-Generative-Adverserial-Network.git
python nepali_letter_gen_using_dcgan_on_nepali_letter_final.py
Pull requests are welcome. For major changes, please open an issue first to discuss what you would like to change.
Please make sure to update tests as appropriate.