Inverted Residuals and Linear Bottlenecks: Mobile Networks for Classification, Detection and Segmentation

By Mark Sandler, Andrew G. Howard, Menglong Zhu, Andrey Zhmoginov, Liang-Chieh Chen. 2018

1,045 citations recorded by the source corpus.

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

In Efficiency & Compression

Citations in the corpus

Do Better ImageNet Models Transfer Better? (2018, methodology): “These networks encompassed widely used Inception architectures [69, 34, 70, 68]; ResNets [29, 26, 25]; DenseNets [31]; MobileNets [30, 61]; and NASNets [84].”
ShuffleNet V2: Practical Guidelines for Efficient CNN Architecture Design (2018, methodology): “Recent advances in light-weight neural network architectures [2,8,18,21,24, 35,39] are mostly based on the metric of FLOPs and do not consider these properties above.”
Efficient Multi-Objective Neural Architecture Search via Lamarckian Evolution (2018, methodology): “Recently, several architectures have been manually designed that aim at reducing resource-consumption while retaining high predictive performance (Iandola et al., 2016; Howard et al., 2017; Sandler et al., 2018).”
FBNetV3: Joint Architecture-Recipe Search using Neural Acquisition Function (2020, methodology): “This includes convolutional variants like depthwise convolutions in MobileNet, inverted residual blocks in MobileNetV2; activations such as hswish in MobileNetV3 [14, 31, 13]; and memory-based operations like shift [39] and shufﬂe [24].”
BigNAS: Scaling Up Neural Architecture Search with Big Single-Stage Models (2020, methodology): “On the left we show the visualization of FLOPs-Accuracy benchmarks compared with the previous arts including MobileNetV1 [16], NASNet [39], MobileNetV2 [28], AutoSlim-MobileNetV2 [35], MNasNet [31], MobileNetV3 [15], EfficientNet [32] and concurrent work Once-For-All [4].”
AMC: AutoML for Model Compression and Acceleration on Mobile Devices (2018, methodology): “Apart from VGG-16, we also test AMC on modern efficient neural networks MobileNet-V1 [23] and MobileNet-V2 [44].”
MorphNet: Fast & Simple Resource-Constrained Structure Learning of Deep Networks (2017, methodology): “We use the same trainer from MobileNet v2 (2), described below.”
NetAdapt: Platform-Aware Neural Network Adaptation for Mobile Applications (2018, methodology): “The main diﬀer-ences between the MobileNetV1 and MobileNetV2 experiment settings are that each network proposal is short-term ﬁne-tuned with ten thousand iterations, the initial latency reduction is 1ms, the latency reduction decay is 0.995, the batch size is 96, and dropout and label smoothing are ”
A2-Nets: Double Attention Networks (2018, methodology): “In the future, we want to explore integrating the double attention in recent compact network architectures [19, 16, 4], to leverage the expressiveness of the proposed method for smaller, mobile-friendly models.”
ChamNet: Towards Efficient Network Design Through Platform-Aware Model Adaptation (2018, methodology): “It utilizes inverted residual and linear bottleneck to significantly cut down on the number of operations and memory needed per inference [24].”
Slimmable Neural Networks (2018, methodology): “One exception is that for MobileNet v1 and MobileNet v2, we use stochastic gradient descent (SGD) as the optimizer instead of the RMSPropOptimizer (Howard et al., 2017; Sandler et al., 2018).”
Quantizing deep convolutional networks for efficient inference: A whitepaper (2018, methodology): “We note that Mobilenet-v1 [2] and Mobilenet-v2[1] architectures use separable depthwise and pointwise convolutions with Mobilenet-v2 also using skip connections.”
Fast Neural Architecture Search of Compact Semantic Segmentation Models via Auxiliary Cells (2018, methodology): “The encoder part of our network is MobileNet-v2 [32], pretrained on MS COCO [20] for semantic segmentation using the Light-Weight RefineNet decoder [26].”
Non-Discriminative Data or Weak Model? On the Relative Importance of Data and Model Resolution (2019, methodology): “For example, for MobileNets we show the tradeoff on figure 3.”
Fast-SCNN: Fast Semantic Segmentation Network (2019, methodology): “This result indicates that Fast-SCNN has insufﬁcient capacity to reach comparable performance to most standard DCNNs on ImageNet ( > 70% top-1) [10, 28].”