The second version of MobileNet, referred to as MobileNetV2, is even faster than MobileNet. The second version has fewer parameters as well. Since its launch, MobileNetV2 has been widely used in state-of-the-art object detection and segmentation architectures to make object detection or segmentation possible on devices with limited resources. Let's look at the motivation behind the creation of MobileNetV2.
MobileNetV2
Motivation behind MobileNetV2
The researchers at Google wanted MobileNet to be even lighter. How can we make MobileNet have fewer parameters? All the CNN-based models increase the number of feature maps (depth channel) while reducing the width and height. One easy way to reduce the network size would be...
