In this chapter, we’ll look at how we can modify our MobileNet v1 approach to produce MobileNet v2, which is slightly more accurate and computationally cheaper. This network came out in 2018 and delivered an improved version of the v1 architecture.
> MobileNetV2: Inverted Residuals and Linear Bottlenecks
> https://arxiv.org/abs/1801.04381
The key concepts the Google team introduced in this paper were inverted residual blocks and linear bottleneck layers, so let’s look at how they work.
Inverted residual blocks
In our ResNet 50 bottleneck blocks from before, we pass our input ...
