Merging and Evolution: Improving Convolutional Neural Networks for Mobile Applications

Hao Yu; Shiqing Zhang; Yuxing Peng; Zhaoning Zhang; Zheng Qin

RECORD

Merging and Evolution: Improving Convolutional Neural Networks for Mobile Applications

Zheng Qin, Zhaoning Zhang, Shiqing Zhang, Hao Yu, Yuxing Peng

Computer Vision and Pattern Recognition · Submitted 24 Mar 2018

abstract compiled · 2026-09-23 · no language-model review

The authors' main claim Extensive experiments on ILSVRC 2012 dataset and PASCAL VOC 2007 dataset demonstrate that MENet consistently outperforms other state-of-the-art compact networks under different computational budgets.

how it is stated · asserted 1.00 read the record →

Comment (0) Read on arXiv ↗

Cite ▾

arXiv ID1803.09127 (version 1)

Pith IDHGRHG2DH, Pith's permanent record of this paper and its review

CitationHao Yu, Shiqing Zhang, Yuxing Peng, Zhaoning Zhang, Zheng Qin. "Merging and Evolution: Improving Convolutional Neural Networks for Mobile Applications." arXiv:1803.09127 (2018). https://pith.science/paper/HGRHG2DH

Claim paper

The compiled record

The paper's own sentences, arranged. Each label is drawn from a closed vocabulary; the number beside it is the probability the classifier gave that label. Numbers in the tables are the authors', checked by code. No sentence in this record was written by a model.

Open the compiled record

Compiled by classifier-1.13.0 via TypeSafe API, question bank abstract-p1-20260923, 2026-09-23: 1 request, 7,017 input tokens.

What the abstract itself states, as probabilities: claims first 0.05 · states limitation 0.08 · human participants 0.02 · controlled comparison 0.73 · randomized assignment 0.01 · states quantitative result 0.99 · compares with prior results 0.98 · mentions code or data availability 0.02.

The abstract, sentence by sentence

arXiv listing abstract v1. Results and the main claim are underlined; show every role.

Compact neural networks are inclined to exploit "sparsely-connected" convolutions such as depthwise convolution and group convolution for employment in mobile applications. Compared with standard "fully-connected" convolutions, these convolutions are more computationally economical. However, "sparsely-connected" convolutions block the inter-group information exchange, which induces severe performance degradation. To address this issue, we present two novel operations named merging and evolution to leverage the inter-group information. Our key idea is encoding the inter-group information with a narrow feature map, then combining the generated features with the original network for better representation. Taking advantage of the proposed operations, we then introduce the Merging-and-Evolution (ME) module, an architectural unit specifically designed for compact networks. Finally, we propose a family of compact neural networks called MENet based on ME modules. Extensive experiments on ILSVRC 2012 dataset and PASCAL VOC 2007 dataset demonstrate that MENet consistently outperforms other state-of-the-art compact networks under different computational budgets. For instance, under the computational budget of 140 MFLOPs, MENet surpasses ShuffleNet by 1% and MobileNet by 1.95% on ILSVRC 2012 top-1 accuracy, while by 2.3% and 4.1% on PASCAL VOC 2007 mAP, respectively.

background contribution method result interpretation main claim

What the classifier reports for each sentence. The number is the probability it gave the label.
#	sentence	role	the authors' own work	states the main conclusion	strength
s0	Compact neural networks are inclined to exploit "sparsely-connected" convolutions such as depthwise convolution and group convolution for employment in mobile applications.	background 1.00	0.07	0.02	asserted 0.52
s1	Compared with standard "fully-connected" convolutions, these convolutions are more computationally economical.	background 1.00	0.10	0.02	asserted 1.00
s2	However, "sparsely-connected" convolutions block the inter-group information exchange, which induces severe performance degradation.	background 0.80	0.13	0.03	asserted 1.00
s3	To address this issue, we present two novel operations named merging and evolution to leverage the inter-group information.	contribution 0.97	0.91	0.06	asserted 0.97
s4	Our key idea is encoding the inter-group information with a narrow feature map, then combining the generated features with the original network for better representation.	method 0.92	0.75	0.06	asserted 0.99
s5	Taking advantage of the proposed operations, we then introduce the Merging-and-Evolution (ME) module, an architectural unit specifically designed for compact networks.	contribution 0.63	0.92	0.06	asserted 0.99
s6	Finally, we propose a family of compact neural networks called MENet based on ME modules.	contribution 0.89	0.94	0.06	asserted 1.00
s7	Extensive experiments on ILSVRC 2012 dataset and PASCAL VOC 2007 dataset demonstrate that MENet consistently outperforms other state-of-the-art compact networks under different computational budgets.	result 0.98	0.96	0.79	asserted 1.00
s8	For instance, under the computational budget of 140 MFLOPs, MENet surpasses ShuffleNet by 1% and MobileNet by 1.95% on ILSVRC 2012 top-1 accuracy, while by 2.3% and 4.1% on PASCAL VOC 2007 mAP, respectively.	result 1.00	0.95	0.31	asserted 1.00

What kind of paper the classifier reports this is: method or tool 1.00 · evidence from observation or experiment 0.99.

What was examined, and what was not

sources read

arXiv listing abstract v1

not examined

full text
figures and tables
references and citing sentences
relations to other papers

levels run

P abstract

not run

S sections
T body sentences
C citations
R cross-paper relations

Every quoted sentence on this record is the authors'. Every label is from Pith's closed vocabulary, with the probability shown beside it. Every number in the measurements table is the authors', checked by code. Record as JSON

The paper

Original abstract

Compact neural networks are inclined to exploit "sparsely-connected" convolutions such as depthwise convolution and group convolution for employment in mobile applications. Compared with standard "fully-connected" convolutions, these convolutions are more computationally economical. However, "sparsely-connected" convolutions block the inter-group information exchange, which induces severe performance degradation. To address this issue, we present two novel operations named merging and evolution to leverage the inter-group information. Our key idea is encoding the inter-group information with a narrow feature map, then combining the generated features with the original network for better representation. Taking advantage of the proposed operations, we then introduce the Merging-and-Evolution (ME) module, an architectural unit specifically designed for compact networks. Finally, we propose a family of compact neural networks called MENet based on ME modules. Extensive experiments on ILSVRC 2012 dataset and PASCAL VOC 2007 dataset demonstrate that MENet consistently outperforms other state-of-the-art compact networks under different computational budgets. For instance, under the computational budget of 140 MFLOPs, MENet surpasses ShuffleNet by 1% and MobileNet by 1.95% on ILSVRC 2012 top-1 accuracy, while by 2.3% and 4.1% on PASCAL VOC 2007 mAP, respectively.

Human review

How Pith checks papers → · Full audit of this paper →

How the literature uses it

What this paper does with the arXiv papers it cites

Each citing sentence is taken from the citing paper's LaTeX source and typed by Pith's classifier. A label appears at probability 0.8 or more (0.95 for support), a contest at 0.7 or more; the quoted words are the citing authors'. How each label was checked · JSON

What it cites · 8 sentences without a confident label

Uses the method or tool of 2 papers

RoI align [ref] is used to encode RoIs instead of RoI pooling [ref] .

Mask R-CNN · Object Detection on PASCAL VOC · 1.00
Following [ref] , we replace the ME modules with other structures in the architecture shown in Table [ref] and adapting the number of channels to the computational budgets.

ShuffleNet: An Extremely Efficient Convolutional Neural Network for Mobile Devices · Comparison with Other State-of-the-art Networks · 0.90

Compares with 4 papers

We also compare the performance of MENet with two state-of-the-art compact networks: ShuffleNet [ref] and MobileNet [ref] .

MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications · Comparison with Other Compact Networks · 1.00
We also compare the performance of MENet with two state-of-the-art compact networks: ShuffleNet [ref] and MobileNet [ref] .

ShuffleNet: An Extremely Efficient Convolutional Neural Network for Mobile Devices · Comparison with Other Compact Networks · 1.00
We further compare the ME module with the building structures of three state-of-the-art networks, ResNet [ref] , Xception [ref] and ResNeXt [ref] .

Xception: Deep Learning with Depthwise Separable Convolutions · Comparison with Other State-of-the-art Networks · 0.98

Cites as background 9 papers

MobileNet [ref] exploited depthwise separable convolutions as its building unit, which decompose a standard convolution into a combination of a depthwise convolution and a pointwise convolution.

MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications · Related Work · 0.95
He et al. [ref] proposed a LASSO regression based method to prune redundant channels, achieving $5 ×$ speedup with comparable accuracy.

Channel Pruning for Accelerating Very Deep Neural Networks · Related Work · 0.89
Convolutional neural networks (CNNs) have achieved significant progress in computer vision tasks such as image classification [ref] , object detection [ref] and semantic segmentation [ref] .

Very Deep Convolutional Networks for Large-Scale Image Recognition · Introduction · 0.99

Discussion

Know something about this paper? Leave a comment, or highlight any passage above to comment on it.