MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications
By Andrew G. Howard, Menglong Zhu, Dmitry Kalenichenko, Weijun Wang, Tobias Weyand, M. Andreetto, Hartwig Adam. 2017
25,507 citations recorded by the source corpus.
We present a class of efficient models called MobileNets for mobile and embedded vision applications. MobileNets are based on a streamlined architecture that uses depth-wise separable convolutions to build light weight deep neural networks. We introduce two simple global hyper-parameters that efficiently trade off between latency and accuracy. These hyper-parameters allow the model builder to choose the right sized model for their application based on the constraints of the problem. We present extensive experiments on resource and accuracy tradeoffs and show strong performance compared to other popular models on ImageNet classification. We then demonstrate the effectiveness of MobileNets across a wide range of applications and use cases including object detection, finegrain classification, face attributes and large scale geo-localization.
Subfields: Computer Vision and Pattern Recognition, Efficiency & Compression
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Computer Vision and Pattern Recognition
- Luc Van Gool
- Dacheng Tao
- Ming-Hsuan Yang
- Xiaogang Wang
- Trevor Darrell
- Yi Yang
- Jitendra Malik
- Alan L. Yuille
In Efficiency & Compression
Citations in the corpus
- Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift (2015, background): “4.2.1, we significantly increase the training speed of the network.”
- A Constructive Prediction of the Generalization Error Across Scales (2019, background): “For instance, image classification networks can be scaled by depth (number of layers; He et al., 2016) or width (number of channels; Zagoruyko & Komodakis, 2016; Howard et al., 2017).”
- EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks (2019, methodology): “Width ( w w w ): Scaling network width is commonly used for small size models (Howard et al., 2017; Sandler et al., 2018; Bigger networks with larger width, depth, or resolution tend to achieve higher accuracy, but the accuracy gain quickly saturate after reaching 80%, demonstrating the limitation…”
- Chasing Carbon: The Elusive Environmental Footprint of Computing (2020, background): “, inference images per second) [61], [62], and the manufacturing carbon footprint.”
- Axial-DeepLab: Stand-Alone Axial-Attention for Panoptic Segmentation (2020, methodology): “Our method bears a similarity to decoupled convolution [40], which factorizes a depthwise convolution [73,34,19] to a column convolution and a row convolution.”
- Local Relation Networks for Image Recognition (2019, background): “regular ω X all/one/no local ω X top-down group [17, 31] ω X group/one/no local ω X top-down depthwise [5, 11] ω X one/one/no local ω X top-down dilated [4, 33] ω X all/one/no atrous ω X top-down active [14] ω, Ω X all/one/no Ω ω X top-down local connected [25] ω ✗ all/one/no local ω X top-down dyna”
- Group Normalization (2018, background): “MobileNet (Howard et al. 2017) andXception (Chollet 2017) exploit channel-wise (also called “depth-wise”) convolutions, which are group convolutions with a group number equal to the channel number.”
- Late Temporal Modeling in 3D CNN Architectures with BERT for Action Recognition (2020, background): “Another important 3D CNN architecture is Channel-Separated Convolutional Networks (CSN) [27] which separates the channel interactions and spatio-temporal interactions which can be thought as the 3D CNN version of depth-wise separable convolution [14].”
- Do ImageNet Classifiers Generalize to ImageNet? (2019, background): “[25] Andrew G Howard, Menglong Zhu, Bo Chen, Dmitry Kalenichenko, Weijun Wang, Tobias Weyand, Marco Andreetto, and Hartwig Adam.”
- Bag of Tricks for Image Classification with Convolutional Neural Networks (2018, methodology): “In addition, we show that our approach can generalize to other networks (Inception V3 [1] and MobileNet [11]) and datasets (Place365 [32]).”
- Speed/Accuracy Trade-Offs for Modern Convolutional Object Detectors (2016, methodology): “• MobileNet [2]: We extract features from the “Conv2d 11” layer whose stride size is 16 pixels.”
- On Buggy Resizing Libraries and Surprising Subtleties in FID Calculation (2021, neutral): “For linear filters, this corresponds to a “depthwise convolution”, using deep learning parlance [31,68].”
- Measuring the Algorithmic Efficiency of Neural Networks (2020, positive): “3 [Howard et al., 2017] Howard, A. G., Zhu, M., Chen, B., Kalenichenko, D., Wang, W., Weyand, T., An-dreetto, M., & Adam, H. (2017).”
- Reinforce Data, Multiply Impact: Improved Model Accuracy and Robustness with Dataset Reinforcement (2023, background): “Knowledge distillation (KD) refers to training a student model using the outputs of a teacher model [9, 2, 35].”
- FocalClick: Towards Practical Interactive Image Segmentation (2022, methodology): “Besides, as FocalClick is a universal pipeline that could be adapted to various Segmentors, the FLOPs could be further reduced by using more light-weighted architectures like [17, 37].”