Source code of Lenet 5 model implemented by using C/C++.

In this final project, we will implement and optimize the forward-pass of a convolutional layer utilizing CUDA. Convolutional layers are fundamental components of convolutional neural networks (CNNs).

Our optimized CUDA implementation of the convolutional layer will specifically cater to the inference process for layers C1 and C3.

To kickstart this endeavor, the Mini-DNN-CPP (Mini-DNN) framework will serve as a foundational platform for implementing a modified version of LeNet-5.

The chosen dataset for this project is Fashion MNIST, a collection of single-channel images with dimensions of 28 x 28 pixels. The output layer comprises 10 nodes, with each node representing the likelihood of the input belonging to one of the 10 classes, such as T-shirt, dress, sneaker, boot, and others.

git clone https://[email protected]/luong1409/Lenet-CUDA.git
cd Lenet-CUDA
make clean
make setup
make train_model

git clone https://[email protected]/luong1409/Lenet-CUDA.git
cd Lenet-CUDA
make clean
make setup
make test_model_cpu

git clone https://[email protected]/luong1409/Lenet-CUDA.git
cd Lenet-CUDA
make clean
make setup
make basic
make test_model_gpu

git clone https://[email protected]/luong1409/Lenet-CUDA.git
cd Lenet-CUDA
make clean
make setup
make constant_mem
make test_model_gpu

git clone https://[email protected]/luong1409/Lenet-CUDA.git
cd Lenet-CUDA
make clean
make setup
make shared_mem
make test_model_gpu

git clone https://[email protected]/luong1409/Lenet-CUDA.git
cd Lenet-CUDA
make clean
make setup
make combine
make test_model_gpu

MNIST dataset