amanbind007 / Sign-Language-to-Text-Covertor

Real-time true action-detection based Sign Language to Text Covertor

Sign-Language-to-Text-Covertor

Real-time true action-detection based Sign Language to Text Covertor

Mediapipe Dataset is available at Kaggle

https://www.kaggle.com/datasets/amanbind/mp-data-no-face
To know how to use this dataset read the comments in the jupyter files from 1 to 6
You can also this video, I created this project from here itself.

Videos Dataset is Available at kaggle

https://www.kaggle.com/datasets/amanbind/smallwlasl/
To know how to use video dataset read the comments in 2 and 3 jupiter files. I have tried to explain there.
If You are using Video Dataset use 3rd script (3 Flip videos horizontally.ipynb) to increase the number of videos and workable for left hand users.

If you want to train the model yourself you can run the 5th jupyter file or use this file on kaggle i have created. I recommend you use kaggle with free GPU for processing if you do not have any PC with GPU.

The requirements.txt file is for conda on M1 Silicon
If You are using M1 based mac, you have to download download the following libraries

1. tensorflow-macos
2. mediapipe-silicon
3. opencv-python
4. jupyter
5. numpy
6. matplotlib

You can also use leverage apple metal gpu by using tensorflow-metal, but then you will have to first download and install miniconda.
Steps of installation are given here

If you are using any other system then you can simply install the following libraries through pip

1. tensorflow
2. tensorflow-gpu (optional, ie., if you have cuda GPU)
3. mediapipe
4. opencv-python
5. jupyter
6. numpy
7. matplotlib