To utilize the Kaggle API, ensure you have a Kaggle account and a Kaggle API token saved as a JSON file named kaggle.json
. Run the script with the --json_path
argument pointing to the location of your kaggle.json
file for authentication. For a step-by-step tutorial on obtaining your Kaggle API token, click here.
To separately download the ASL Dataset, you can click here.
Alternatively, you can use the provided dataset_downloader.py
script to automatically download the dataset for you. Simply run the script with the appropriate arguments, and it will handle the download and extraction process for you.
To begin, navigate to your Colab environment and ensure you have all the necessary Python scripts from the neuralnet
directory and the script
directory. Additionally, make sure you have your Kaggle API JSON file ready for uploading.
-
Accessing Colab Environment: Open your Colab notebook and ensure you are connected to the runtime.
-
Upload Python Scripts: Click on the "Files" icon on the left sidebar and select "Upload". Navigate to the directories containing your Python scripts (
neuralnet
andscript
), select all relevant files, and upload them to your Colab environment. -
Upload Kaggle API JSON File: Follow the same process to upload your Kaggle API JSON file. This file is named
kaggle.json
and contains your Kaggle API token for authentication.
Once all files are uploaded, you can proceed to execute your Python scripts within the Colab environment, utilizing the uploaded scripts and the Kaggle API token for accessing datasets.
All the required python commands to run the scripts are in notebook ASL_Alphabet_Classification.ipynb
.
- It may take around 1 hour to train.
- After training is completed, download model file named
efficientnet_model.pth
undermodels
directory.
If you wish to train the model on your local device, please ensure that
- You have at least 8GB of VRAM on a CUDA-supported GPU device.
- Kaggle API token (
kaggle.json
) is moved to the directory"C:/Users/{your_username}/.kaggle/"
on Windows, or"~/.kaggle/"
on Linux or macOS. - To provide the absolute path of the new
kaggle.json
location when running the script with the--json_path
argument for `dataset_downloader.py Pytorch Cuda
version is installed with appropriate version ofNvidia CUDA Toolkit
.
To run the demo:
-
Install Dependencies:
pip3 install torch torchvision --index-url https://download.pytorch.org/whl/cpu
pip install -r requirements.txt
-
Run Demo:
-
Run
main.py
for image classification using streamlit demo:cd demo streamlit run main.py
-
Run
detect.py
for live detection using opencv/mediapipe demo:python3 detect.py
Note: If you have no webcam you can use application named droidcam instead to turn mobile phone as webcam and logs are created on the file named
action_handler.log
. -
This project utilizes the EfficientNetB0
CNN architecture model for image classification. The pre-trained model is available in the model/
directory. You can load the model file efficientnet_model.pth
on Colab or a local device to perform inference on American Sign Language images.
Feel free to report any issues you encounter.
Don't forget to star the repo :)