Русская версия

Easy_sign

Easy_sign is an open source russian sign language recognition project that uses small CPU model for predictions and is designed for easy deployment via Streamlit.

About the project

Easy_sign uses a machine learning model to recognize Russian sign language gestures. The model was trained on ~180,000 examples of Russian sign language gestures. Approximately 20,000 of which are taken from the Slovo dataset. The model recognizes 1598 Russian sign language gestures and can process 3-3.5 gestures per second on an Intel(R) Core(TM) i5-6600 CPU @3.30GHz. The list of recognized gestures is available in the file RSL_class_list.txt.

To learn more about the project - visit the link to our article on habr.

Installation

conda create --name fleury-env python=3.10
conda activate fleury-env
pip install -r requirements.txt

Usage

streamlit run app.py

Links

The PINCODE team has released a training simulator based on easy_sign to study Russian sign language. Habr (Russian), github

S3D models trained on the Slovo dataset for different input lengths. Weights: Hugging Face.

Frames	Link	Mean accuracy, %
32	https://huggingface.co/maximazzik/easy_sign/resolve/main/slovo/S3Dx32x1x1001.onnx	44.22
48	https://huggingface.co/maximazzik/easy_sign/resolve/main/slovo/S3Dx48x1x1001.onnx	52.28
64	https://huggingface.co/maximazzik/easy_sign/resolve/main/slovo/S3Dx64x1x1001.onnx	55.86

License

This work is licensed under the Creative Commons Attribution-ShareAlike 4.0 International License.