fecnn

A fork from Marvin, and simplify this project for learning Deep Learning.

This is a neural network framework using c++11 and CUDA.

Dependences

Download CUDA 7.5 and [cuDNN 5.1](https://developer.nvidia.com/ cudnn).

Config you cuda path in compile.sh.

Compilation

sh compile.sh

Usage

Convolutional Neural Network on MNIST digits.

1. Prepare data: run examples/mnist/prepare_mnist.m in Matlab

2. Config your neural network model in /examples/mnist,just like this file lenet.json.

3. Visualize filters: config your json file path in /tools/viewer_matlab/plotNet.m,and run it in Matlab. See you train result like this：

4. Train a model: run ./examples/mnist/train.sh in shell. See you train result like this：

5. Visualize different model change the train result: use /tools/picture.py. Before run it,you need config:

	file_list:the different models tained result output in shell,and save as files.
	dirname:the dir of file_list.