The object of the project is to distinguish 43 different types of traffic sign used is Germany. Number of Output layers in LeNet has changed to 43. Then, the network has retrained using Traffic sign data
Images of German Traffic Sign Dataset have used for training and test. Pickled datasets are available below.
- Training data (102MB)
- Validation data (12.9MB)
- Test data (37MB)
- Training data (431MB)
or you can download all data Traffic_Signs_data.zip (395MB)
LeNet with Batch Normalization before each activation layer. No dropout!
Convolutional weights were initialized by 'He' method.
- Python 3.5.2
- Tensorflow 1.0.1
- Optimizer:
Adam - Learning rate:
10e-3 - Loss:
Cross entropy - Batch Size:
1024 - Epoch:
100
