Code for the final project
To build docker image for Jetson
./Edge/build_docker_image.sh
To run the docker container
./Edge/run_container.sh
Once inside the container, run the hand capture and predict image by
python3 hand_detect_pb.py
Supplemental information on other options with the container will be found in
./Edge/bashcode
- VGG-16 so named for its 16 trainable layers ( 13 convolution + 3 dense layers)
- The model was imported Keras applications
- Pre-trained model weights was obtained from F.Chollet's github: F.Chollet's github
- The images from ASL dataset was sized down into 50X50 and trained for 10 epochs
- Notebook and standalone python script is available under Cloud/Models/VGG16/
- ResNet-50 model pretrained on ImageNet and imported from the Keras applications.
- Creates the ResNet-50 model and downloads the weights pretrained on the ImageNet dataset.
- ResNet-50 expects the images to be 224 x 224 pixels in size so we used the tf.image.resize() function to resize our images
- ImageDataGenerator to load the images and augment them in various ways.
