How to Start:
- Download the dataset from roboflow - https://universe.roboflow.com/majorproject-25tao/american-sign-language-v36cz/dataset/2/images?split=train in Pascal-Voc(.xml) format
- Run the feature Extraction code first and please replace the path directories according to your path.
- I created train_images, test_images and valid_images folders which are for the images and extracted features for objects saved in txt.
- Point path in code to these above directories.(Not the starting cells).
- save the output.csv for later Evaluation.
Training the YOLOv5
- upload the train_images,valid_images in drive and open the collaboratory file and run the following commands.
- It took me 11 hrs for training, you can use collab pro to have a smooth experience
Testing the Model
- Download the exported model in ONXX format and point the file in Test_images.ipynb file.
- Don't forget to point the output.csv file in code.
- Second from last is to generate the text from the labels created either by images or videos.
- Last cell will convert genrated text to audio.
NOTE Please refer the attached ppt to know more in detail and any inputs to improve the code can be appreciated.