Rethinking the Inception Architecture for Computer Vision
Christian Szegedy
1
,
Zbigniew Wojna
2
Publication type: Proceedings Article
Publication date: 2016-06-01
Abstract
Convolutional networks are at the core of most state of-the-art computer vision solutions for a wide variety of tasks. Since 2014 very deep convolutional networks started to become mainstream, yielding substantial gains in various benchmarks. Although increased model size and computational cost tend to translate to immediate quality gains for most tasks (as long as enough labeled data is provided for training), computational efficiency and low parameter count are still enabling factors for various use cases such as mobile vision and big-data scenarios. Here we are exploring ways to scale up networks in ways that aim at utilizing the added computation as efficiently as possible by suitably factorized convolutions and aggressive regularization. We benchmark our methods on the ILSVRC 2012 classification challenge validation set demonstrate substantial gains over the state of the art: 21:2% top-1 and 5:6% top-5 error for single frame evaluation using a network with a computational cost of 5 billion multiply-adds per inference and with using less than 25 million parameters. With an ensemble of 4 models and multi-crop evaluation, we report 3:5% top-5 error and 17:3% top-1 error on the validation set and 3:6% top-5 error on the official test set.
Top-30
Journals
200
400
600
800
1000
1200
|
|
Lecture Notes in Computer Science
1155 publications, 6.57%
|
|
IEEE Access
859 publications, 4.89%
|
|
Multimedia Tools and Applications
320 publications, 1.82%
|
|
Communications in Computer and Information Science
233 publications, 1.33%
|
|
Scientific Reports
225 publications, 1.28%
|
|
Lecture Notes in Networks and Systems
188 publications, 1.07%
|
|
Sensors
165 publications, 0.94%
|
|
Applied Sciences (Switzerland)
161 publications, 0.92%
|
|
IEEE Transactions on Pattern Analysis and Machine Intelligence
154 publications, 0.88%
|
|
Neural Computing and Applications
148 publications, 0.84%
|
|
Neurocomputing
138 publications, 0.79%
|
|
Expert Systems with Applications
134 publications, 0.76%
|
|
IEEE Transactions on Image Processing
126 publications, 0.72%
|
|
IEEE Transactions on Geoscience and Remote Sensing
124 publications, 0.71%
|
|
Advances in Intelligent Systems and Computing
112 publications, 0.64%
|
|
PLoS ONE
103 publications, 0.59%
|
|
Applied Intelligence
102 publications, 0.58%
|
|
IEEE Transactions on Multimedia
102 publications, 0.58%
|
|
Remote Sensing
98 publications, 0.56%
|
|
Biomedical Signal Processing and Control
98 publications, 0.56%
|
|
IEEE Transactions on Neural Networks and Learning Systems
94 publications, 0.53%
|
|
IEEE Journal of Biomedical and Health Informatics
91 publications, 0.52%
|
|
IEEE Transactions on Circuits and Systems for Video Technology
91 publications, 0.52%
|
|
Electronics (Switzerland)
84 publications, 0.48%
|
|
Lecture Notes in Electrical Engineering
83 publications, 0.47%
|
|
Neural Networks
80 publications, 0.46%
|
|
Computers and Electronics in Agriculture
79 publications, 0.45%
|
|
Computers in Biology and Medicine
77 publications, 0.44%
|
|
Engineering Applications of Artificial Intelligence
77 publications, 0.44%
|
|
200
400
600
800
1000
1200
|
Publishers
1000
2000
3000
4000
5000
6000
7000
8000
|
|
Institute of Electrical and Electronics Engineers (IEEE)
7202 publications, 40.98%
|
|
Springer Nature
4370 publications, 24.87%
|
|
Elsevier
2016 publications, 11.47%
|
|
MDPI
1001 publications, 5.7%
|
|
Association for Computing Machinery (ACM)
713 publications, 4.06%
|
|
Wiley
308 publications, 1.75%
|
|
Cold Spring Harbor Laboratory
188 publications, 1.07%
|
|
Taylor & Francis
129 publications, 0.73%
|
|
Institution of Engineering and Technology (IET)
121 publications, 0.69%
|
|
Frontiers Media S.A.
118 publications, 0.67%
|
|
Public Library of Science (PLoS)
113 publications, 0.64%
|
|
SPIE-Intl Soc Optical Eng
88 publications, 0.5%
|
|
Hindawi Limited
81 publications, 0.46%
|
|
IOP Publishing
72 publications, 0.41%
|
|
SAGE
58 publications, 0.33%
|
|
IGI Global
50 publications, 0.28%
|
|
Oxford University Press
38 publications, 0.22%
|
|
JMIR Publications
37 publications, 0.21%
|
|
World Scientific
31 publications, 0.18%
|
|
Walter de Gruyter
26 publications, 0.15%
|
|
Cambridge University Press
24 publications, 0.14%
|
|
AIP Publishing
22 publications, 0.13%
|
|
American Institute of Mathematical Sciences (AIMS)
22 publications, 0.13%
|
|
EDP Sciences
21 publications, 0.12%
|
|
Emerald
21 publications, 0.12%
|
|
Institute of Electronics, Information and Communications Engineers (IEICE)
17 publications, 0.1%
|
|
Pleiades Publishing
16 publications, 0.09%
|
|
American Chemical Society (ACS)
15 publications, 0.09%
|
|
BMJ
15 publications, 0.09%
|
|
1000
2000
3000
4000
5000
6000
7000
8000
|
- We do not take into account publications without a DOI.
- Statistics recalculated only for publications connected to researchers, organizations and labs registered on the platform.
- Statistics recalculated weekly.
Are you a researcher?
Create a profile to get free access to personal recommendations for colleagues and new articles.
Found error?
Found error?