neXuralNetwork (neXt neural network)

This toolbox has been written as a part of my bachelor degree project and contains the implementation of a convolitional neural network that it's written in C++ and exported to Python as a module. neXuralNetwork is and always will be an open source project where anyone can contribute to it.

Compiling neXuralNet on Windows

Firstly you need to install the following softwares:

• CMake (version 3.7.0 or newer)
• Visual Studio - only Visual Studio 2015 and newer are supported since pybind11 relies on various C++11 language features that break older versions of Visual Studio;
• Python - needed for building the python module (we recomand Anaconda environment);

neXuralNet uses following libraries as its dependencies:

• OpenCV (version 3.0.0 or newer) - because this library is huge, it isn't included with this project (you need to download/build it). For VS2017 you can download and use the OpenCV from here
• RapidJSON - included as submodule (no action required)
• pybind11 - included as submodule (no action required)
• GTest - included as submodule (no action required) - used only for tests

Type the following to compile the neXuralNet Python module:

git clone --recursive https://github.com/nitnelav1101/nexuralnetwork.git
cd nexuralnetwork
mkdir build
cd build
SET OpenCV_DIR=<YOUR_PATH_HERE>
cmake -G "Visual Studio 15 2017 Win64" ..
cmake --build . --config Release --target INSTALL

When building process is complete, the final .pyd file (Python module) will be located in <PROJECT_ROOT_DIR>/build/install/py_module. Copy the content of this folder to the Python's site-packages directory.

Usage at a glance

This is a minimal Python script of running neXural module (in order to run this script, you need to install also the OpenCV and Numpy Python modules):

import nexuralnet
import numpy as np
import cv2

# Run the trainer
trainer = nexuralnet.trainer('network.json', 'trainer.json')
trainer.train('<TRAINING_DATA>', '<TARGET_DATA>', 'trained_file.json', '<INFO_FOLDER_PATH>', <TRAINING_DATA_SOURCE_TYPE>, <TARGET_DATA_SOURCE_TYPE>);

# Run the network
net = nexuralnet.network('network.json')
net.deserialize('c:/Users/Alexandru Musat/Google Drive/neXuralNet/nexural-data/mnist_softmax/mnist.json')

image = cv2.imread('image.jpg', 0) # 0 - read as a gray image, for color use 1
net.run(image)
net.saveFiltersImages('trained_file.json')
result = net.getResultJSON()
print(result)

License

neXuralNetwork (neXt neural network) is licensed under the MIT License. See the LICENSE for the specific language governing permissions and limitations under the License.