Pre-requisite:
- Download the dataset from the following link:
WLASL dataset - Extract the datset and transfer the videos folder to dataset folder in current base directory containing these code files.
- Create python virtual environment:
conda create --name <env_name> --file requirements.txt - Activate virual env:
conda activate <env_name> - Download the trained model from the link and move to trained_model folder:
Trained Model
Files Walkthrough with order of operations:
-
dataset_splitter.py: This file read the original dataset and create folder specific num_classes containing training, val and test data. And copies files from the original dataset folder to num_classes data folder.
-
data_augment.py: This file parse through training and validation data and creates copies of files with differnet crop ratio and scaling rate.
-
extract_features.py: This file uses media_pip framework to extract 3d features from the video files
-
spatio_temporal_conv.py: Code containing class of R(2+1)D CNN model.
-
sign_language_real_time.py: Code to do real time sign language detection.