DeepPotato

Artificial neural network

Build

mkdir build
cd build
cmake .. -DBUILD_XOR=ON -DBUILD_IMAGE_COMPRESSION=OFF -DBUILD_DIGITS=OFF
make

Windows

mkdir build
cd build
cmake .. -G "NMake Makefiles" -DBUILD_XOR=ON -DBUILD_IMAGE_COMPRESSION=OFF -DBUILD_DIGITS=OFF
nmake

Demo description

XOR

XOR demo is a simple neural network that learns XOR function.

Image compression

Image compression demo is a simple neural network that reproduces input image.

Digit recognition

Artificial neural network that recognizes handwritten digits.
Three types of tests are available:

• Test on MNIST dataset - slowly loops through test images by displaying single digit at a time and its prediction. (#define AUTO_TEST)
• Manual test - allows to draw digit on the screen and see the prediction. (both #define AUTO_TEST and #define WEBCAM NOT defined)
  
• Web cam test - use camera to find and recognize digits (#define AUTO_TEST NOT defined and #define WEBCAM defined, requires OpenCV installed and -DUSE_OPENCV=ON CMake option set)

Tests can be selected by adding or removing #define TEST and #define WEBCAM in demo/digits/main.cpp in the configuration section at the top of the file.
Pretrained model is available in demo/digits/models directory.

Note

Additional dependencies are required for image compression and digit recognition demos.

cd demo/external
git clone https://github.com/nothings/stb
git clone --depth 1 --branch SDL2 https://github.com/libsdl-org/SDL.git