MobileNet

MobileNet is a family of efficient Convolutional Neural Networks (CNNs) designed by Google (2017) to enable high-performance, low-latency computer vision on mobile and embedded devices.

This architecture is a game-changer for on-device inference: it achieves efficiency by replacing standard convolutions with depthwise separable convolutions, a technique that drastically reduces computation and model size. The original MobileNetV1, introduced in 2017, features 28 layers and uses two global hyperparameters (width and resolution multipliers) to precisely tune the trade-off between latency and accuracy. This factorization of convolution results in models that approach the accuracy of larger networks like VGG16 on datasets like ImageNet, yet operate with substantially fewer parameters and lower power consumption, making real-time applications like object detection and facial recognition feasible on a Google Pixel or similar resource-constrained hardware.

https://arxiv.org/abs/1704.04861