TRAINING
The network was trained using 15 images representing counting gesture(1 to 5) for 3 different hands.By different hands I mean that there 3 different people involved.Thus giving different shades of the skin color.The data folder in the code I have uploaded has these images.
Network training was achieved as follows:-
1. Take an image and process it on the GPU, i.e generate the 1's and 0's matrix.
2. Store these values in a .txt file and give them as input to the Kohone SOM neural network, I implemented in MATLAB.
3. Then store the kohonen weights (weights before the kohonen layer) generated in MATLAB in a .txt file.
4. Thus, once I have the weights I read them into my GPU application and do the necessary processing to generate the final output.
Here's the MATLAB code which was used for the above.
