π€ AI cloning the behavior of a driving simulator π
Our data is skewed toward the middle:
Let's balance the data:
Crop the landscape and hood of the car:
Switch to YUV space:
Apply Gaussian Blur:
Resize:
Model: "sequential_4"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
conv2d_5 (Conv2D) (None, 31, 98, 24) 1824
_________________________________________________________________
conv2d_6 (Conv2D) (None, 14, 47, 36) 21636
_________________________________________________________________
conv2d_7 (Conv2D) (None, 5, 22, 48) 43248
_________________________________________________________________
conv2d_8 (Conv2D) (None, 3, 20, 64) 27712
_________________________________________________________________
conv2d_9 (Conv2D) (None, 1, 18, 64) 36928
_________________________________________________________________
flatten (Flatten) (None, 1152) 0
_________________________________________________________________
dense (Dense) (None, 100) 115300
_________________________________________________________________
dense_1 (Dense) (None, 50) 5050
_________________________________________________________________
dense_2 (Dense) (None, 10) 510
_________________________________________________________________
dense_3 (Dense) (None, 1) 11
=================================================================
Total params: 252,219
Trainable params: 252,219
Non-trainable params: 0
Predicted angle:
Actual angle:
Need to downgrade engineio & socketio:
pip install python-socketio==4.6.1
pip install python-engineio==3.13.2