A convolutional neural network for traffic sign image classification.
Traffic Sign Dataset is an image classification dataset. It was made available in 2011 as part of a competition at the International Joint Conference on Neural Networks (IJCNN). The dataset contains 43 classes and over 50,000 images in total.
Note: the dataset was not uploaded in this repository.
A convolution neural network having the following architecture:
CONV2D -> CONV2D -> MAXPOOL -> DROPOUT -> CONV2D -> CONV2D -> MAXPOOL -> DROPOUT --> FLATTEN -> FULLYCONNECTED -> DROPOUT -> FULLYCONNECTED
It was trained from scratch on the training data using PyTorch.
The model achieved 93% accuracy on the validation set (random 20% subset of the training dataset).