Vision Transformer Implementation

---

You can import this module via

pip install vtransformer

How to use it?

1. Configure config yaml file to work with model
2. Module has 2 methods:
   1. run - use for training and testing model
   2. __call__ - use to classify images and utilize model Example of the code:

import VisionTransformer as ViT
import hydra
from pathlib import Path


@hydra.main(config_path="path_to_your_config", config_name="name_of_your_config")
def main(cfg):
   abspath = str(Path(__file__).parent.resolve()) + "/"  # Get absolute path, if your config works with relative paths
   model = ViT.Program(cfg, abspath)  # Create object of class Program to work with transformer

   model.run()  # Run config, that you wrote to train and test model

   arr = ["path1", "path2", "path3"]
   print(model(arr, False))  # Utilize model


if __name__ == "__main__":
   main()

---

Credits

Thanks for the idea and pretrained models to:

• A huge credit to my teacher, Alexandr Korchemnyj
• Idea is based on an article An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale
• I have used pretrained model from SWAG, check the license