Magic Numbers in Lane/Vehicle Detection Example

We're trying to understand the code in the example: "Code Generation for Deep Learning Simulink Model That Performs Lane and Vehicle Detection."

In the lane_detection_coordinates.m file there are some numbers that aren't explained: the persistent laneCoeffMeans and laneCoeffStds. Where did these come from? How would we produce these for this instance or, in general, any instance?

[Image Analyst]

Do you have a reference link for that example? (Make it easy for us - don't make us hunt around for it.)

[squared]

Sure thing.

https://www.mathworks.com/help/gpucoder/ug/lane-vehicle-detection-simulink-gpucoder.html

Answer 1

[Aishwarya]

Hi,

As per my understanding, you are seeking clarification regarding the variables "laneCoeffMeans" and "laneCoeffStds" in the "lane_detection_coordinates.m" file which is mentioned in the documentation attached below: https://www.mathworks.com/help/gpucoder/ug/lane-vehicle-detection-simulink-gpucoder.html

After carefully reviewing the provided documentation, I would like to share my insights on the code:

• The “laneCoeffMeans” and “laneCoeffStds” variables are used to normalize the output of the lane detection network.
• These values are computed during training of deep learning network and are used to scale and shift the network to obtain meaningful parameters for lane boundaries.
• In the example, the values of “laneCoeffMeans” and “laneCoeffStds” variables are set to some default values if they are empty.
• The “laneNetOut” is the normalized output from the deep learning network.
• To revert the normalization process and obtain the original coefficients, the following line of code is implemented:

     params = laneNetOut .* laneCoeffStds + laneCoeffMeans; 

Hope this helps!

Regards,

Aishwarya