This repository contains an adaptation of the LeNet architecture applied to traffic sign classification. Specifically, the goal was to design, train, validate, and test a convolutional neural network (CNN) architecture while experimenting with various image processing techniques to achieve reasonable results.
The code for this project is in the IPython Notebook and you can dive into the details of what I did by reading the write-up.
The final model (built using Tensorflow) was based on the LeNet architecture with additional dropouts added. The final model achieved:
- Training Set Accuracy: ~99.8%
- Validation Set Accuracy: ~95%
- Test Set Accuracy: ~94%
The model architecture is as follows:
| Layer | Description |
|---|---|
| Input | 32x32x1 Grayscale/ Normalized image |
| Convolution 5x5 | 1x1 stride, valid padding, outputs 28x28x6 |
| RELU activation | |
| Max pooling | 2x2 stride, valid padding, outputs 14x14x6 |
| Convolution 5x5 | 1x1 stride, valid padding, outputs 10x10x16 |
| RELU activation | |
| Max pooling | 2x2 stride, valid padding, outputs 5x5x16 |
| Fully connected | 400 input, 120 output |
| RELU activation | |
| Dropout | 0.6 keep probablility (training) |
| Fully connected | 120 input, 84 output |
| RELU activation | |
| Dropout | 0.6 keep probablility (training) |
| Fully connected | 84 input, 43 output |