Sign language translator model

This model is designed to translate sign language into letters according to the American system. In addition, a telegram bot has been written to recognize, oddly enough, sign language. The bot is ready to use, the weights of the model in the file: bot/saved.pt

Bot functionality:

• Recognition of letters in the photo
• Recognition of a text on video
• Recognition of a text in circles

Model training

Training datasets are not stored in the repository, so read data/README.md and create a dataset. The training can be started with the command:

python3 train.py

If you want to put the completed model in its place saved.pt , then in evaluate.py there is a commented-out line for saving the model in .pt format, then the new weights are moved to bot/ and that's it!

Initializing telegram bot

If you want to use telegram bot with the model you've trained, do the following:

• Create a bot using BotFather
• Go to bot/main.py
• Change YOUR_NGROK_TOKEN and YOUR_TELEGRAM_BOT_TOKEN to the corresponding tokens
• Run bot/main.py
• Enjoy!