Neural networks from scratch in CUDA C++

In this project, I tried to make performance similar to PyTorch. but there is still a long way to go.
I was using rtx 4070ti gpu in my local computer.

Algo that I have written in this project.

Layers

• BatchNorm
• CNN2D
• flatten_layer
• linear
• MaxPool2d
• ReLU
• Sigmoid

costs

• MSE(mean squared error)

Models

• MLP(multilayer perceptron)
• Image_CF(image classification). Used opencv lib for image processing.

Image classification Model architecture

How to run this?

For windows

I have used Visual Studio 2019.
MSVC = 14.29.30133 (v142)
CUDA = v12.2

After installing all those tools. It will run without any problem.

For Linux

Copy linux_cuda_nn to your any dir.

cd linux_cuda_nn
mkdir build
cd build
cmake ..
make
./cuda_nn