traffic_light_classifier

Dataset

Copied over from another Udacity repo. As confirmed with Udacity mentors, this dataset is OK to use for the capstone project. The data could have been provided as a git submodule, but it was copied over since there was little data, and so that users don't have to deal with submodules, and in case the other repo gets deleted.

Call to run

python tf_classifier.py

Sample Output

_________________________________________________________________
Layer (type)                 Output Shape              Param #  
=================================================================
batch_normalization_1 (Batch (None, 128, 64, 3)        12        
_________________________________________________________________
conv2d_1 (Conv2D)            (None, 126, 62, 8)        224      
_________________________________________________________________
batch_normalization_2 (Batch (None, 126, 62, 8)        32        
_________________________________________________________________
activation_1 (Activation)    (None, 126, 62, 8)        0        
_________________________________________________________________
dropout_1 (Dropout)          (None, 126, 62, 8)        0        
_________________________________________________________________
average_pooling2d_1 (Average (None, 63, 31, 8)         0        
_________________________________________________________________
conv2d_2 (Conv2D)            (None, 61, 29, 16)        1168      
_________________________________________________________________
batch_normalization_3 (Batch (None, 61, 29, 16)        64        
_________________________________________________________________
activation_2 (Activation)    (None, 61, 29, 16)        0        
_________________________________________________________________
dropout_2 (Dropout)          (None, 61, 29, 16)        0        
_________________________________________________________________
average_pooling2d_2 (Average (None, 30, 14, 16)        0        
_________________________________________________________________
flatten_1 (Flatten)          (None, 6720)              0        
_________________________________________________________________
dense_1 (Dense)              (None, 120)               806520    
_________________________________________________________________
batch_normalization_4 (Batch (None, 120)               480      
_________________________________________________________________
activation_3 (Activation)    (None, 120)               0        
_________________________________________________________________
dropout_3 (Dropout)          (None, 120)               0        
_________________________________________________________________
dense_2 (Dense)              (None, 84)                10164    
_________________________________________________________________
dense_3 (Dense)              (None, 3)                 255      
=================================================================
Total params: 818,919
Trainable params: 818,625
Non-trainable params: 294
_________________________________________________________________
None
Epoch 1/3
74/74 [==============================] - 3s - loss: 0.2232 - acc: 0.9181 - val_loss: 0.3395 - val_acc: 0.9192
Epoch 2/3
74/74 [==============================] - 2s - loss: 0.0963 - acc: 0.9738 - val_loss: 0.0887 - val_acc: 0.9630
Epoch 3/3
74/74 [==============================] - 1s - loss: 0.0764 - acc: 0.9730 - val_loss: 0.0427 - val_acc: 0.9899
Saved trained model at /home/workspace/traffic_light_classifier/tl_net
224/297 [=====================>........] - ETA: 0s('Test loss:', 0.042701885737604167)
('Test accuracy:', 0.98989898989898994)
Predicting one image took 3.249149 ms