What is CNN?
Deep Learning algorithm
One of the Neural networks which have more than one convolutional layers
Takes input image → assigns importance to various aspects using “weight and bias” → allows distinguishing one from the other
Why is CNN used for driverless cars?
CNN is used because it extracts features from images
*features = characteristics of an object, image, etc. For instance, CNN will extract the eyes, nose, and ears as a cat feature.
Different patterns layers at the beginning of the network will capture edges
Deep layers can capture more complex features of the shape
Why is CNN used for automatic vehicles
CNN weights are shareable
This means…same weight parameters can be used to represent two different transformation network → this saves a lot of processing space + produce more diverse feature representations learned by the network
Three important properties
local receptive fields
shared weights
spatial sampling
→ these properties reduce overfitting and store representations features that are vital for image classification, segmentation, localization
When is CNN used
CNN's primary purpose: recognize + classify different parts of the road
The main evidence for making appropriate decisions (ex. accelerate, turn)
Use during "perception" → when the car sees objects
ex. traffic light, pedestrians
Process of CNN: perception → localization → prediction → decision making
Real-life examples
HydraNet by Tesla, ChauffeurNet by Google
Author: Seohee Choy
Editor: Jiho Chang
Citation
Comments