mini-dnn-cpp is a C++ demo of deep neural networks. It is implemented purely in C++, whose only dependency, Eigen, is header-only.
Download and unzip MNIST dataset in mini-dnn-cpp/data/mnist/
.
mkdir build
cd build
cmake ..
make
Run ./demo
.
Result:
simple neural network with 3 FC layers can obtain 0.97+
accuracy on MNIST testset.
LeNet-like CNN can obtain 0.98+
accuracy on MNIST testset.