MobileNet
In plain wordsA family of compact image-recognition networks from Google, designed to run quickly on phones and other small devices with little loss of accuracy.
A family of compact convolutional networksconvolutional neural network, n.: A kind of neural network built for images and other grid-like data. It slides small pattern detectors across the input, finding edges first, then shapes, then whole objects. from Google built to fit the memory, power and speed limits of phones and embedded chips. The first version (2017) replaced nearly every ordinary convolution with a depthwise separable convolutiondepthwise separable convolution, n.: A cheaper way to do the main operation of image networks: first scan each channel of the input separately, then mix the channels together with a tiny one-by-one step. and gave developers two dials, a width multiplier that thins every layer and a resolution multiplier that shrinks the input image, for deciding how much accuracy to give up for speed. MobileNetV2 (2018) introduced the inverted residual block: a cheap 1 × 1 expansion to many channels, a 3 × 3 depthwise convolution, and a linear (activation-free) 1 × 1 projection back to a narrow bottleneck, with the residual connectionresidual connection, n.: A shortcut in a neural network that adds a block's input straight to its output, so each block only has to learn a correction. It is what makes networks hundreds of layers deep trainable. joining the narrow ends. MobileNetV3 (2019) was refined by hardware-aware neural architecture searchneural architecture search, n.: Using an automated search, rather than human trial and error, to design the structure of a neural network: which layers to use, how many, and how to connect them. and added squeeze-and-excitationsqueeze-and-excitation network, n.: An image-network design whose add-on blocks let the network decide, for each picture, which of its internal feature channels to turn up and which to turn down. and a 'hard-swish' activation. These blocks became the basis of EfficientNetEfficientNet, prop. n.: A family of image-recognition networks from 2019 that grows a small base design in a balanced way, deeper, wider and sharper together, to get more accuracy for each unit of computing. and of many on-deviceon-device AI, n.: AI that runs directly on your own phone, laptop, car or headset rather than on a company's distant servers, so it can work offline, respond instantly and keep your data on the device. models for classification, detection and segmentation.
The detector runs a MobileNetV2 backbone at 30 frames per second on the phone's CPU.
Origin
Introduced by Andrew Howard and colleagues at Google in 'MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications' (arXiv, April 2017). Mark Sandler and colleagues followed with MobileNetV2 (January 2018) and Howard and colleagues with MobileNetV3 (May 2019).
Related
- depthwise separable convolutiondepthwise separable convolution, n.: A cheaper way to do the main operation of image networks: first scan each channel of the input separately, then mix the channels together with a tiny one-by-one step.
- EfficientNetEfficientNet, prop. n.: A family of image-recognition networks from 2019 that grows a small base design in a balanced way, deeper, wider and sharper together, to get more accuracy for each unit of computing.
- edge AIedge AI, n.: Doing AI's computing on or near the devices that collect the data, such as cameras, cars, factory machines, phones or local servers, instead of sending everything off to distant data centers.
- on-device AIon-device AI, n.: AI that runs directly on your own phone, laptop, car or headset rather than on a company's distant servers, so it can work offline, respond instantly and keep your data on the device.
- neural architecture searchneural architecture search, n.: Using an automated search, rather than human trial and error, to design the structure of a neural network: which layers to use, how many, and how to connect them.
- squeeze-and-excitation networksqueeze-and-excitation network, n.: An image-network design whose add-on blocks let the network decide, for each picture, which of its internal feature channels to turn up and which to turn down.
Sources
- Howard, A. G. et al. (2017). MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications. arXiv:1704.04861. arxiv.org/abs/1704.04861
Mentioned in
squeeze-and-excitation networksqueeze-and-excitation network, n.: An image-network design whose add-on blocks let the network decide, for each picture, which of its internal feature channels to turn up and which to turn down.