This project aims to recognize finger spelling from hand poses presented as RGB images.
model.lua provides a sample run of the model on a pre-selected sequence of training and testing images chosen for the purpose of this demonstration. With everything else in place, this file can be run as simply as "th model.lua"
We use Google Incpetion architecture to achieve this. The main library used here is TensorFlow
We use a residual network for this. This is implemented in TensorFlow again.
This will do a sample run for a chunked data set consisting of a sequence length of 5. Make sure to unzip the hand_images.zip file before running this command.