Bag of Tricks for Image Classification with Convolutional Neural Networks
Much of the recent progress made in image classification research can be credited to training procedure refinements, such as changes in data augmentations and optimization methods. In the literature, however, most refinements are either briefly mentioned as implementation details or only visible in source code. In this paper, we will examine a collection of such refinements and empirically evaluate their impact on the final model accuracy through ablation study. We will show that, by combining these refinements together, we are able to improve various CNN models significantly. For example, we raise ResNet-50's top-1 validation accuracy from 75.3% to 79.29% on ImageNet. We will also demonstrate that improvement on image classification accuracy leads to better transfer learning performance in other application domains such as object detection and semantic segmentation.
NurtureToken New!

Token crowdsale for this paper ends in

Buy Nurture Tokens

Authors

Are you an author of this paper? Check the Twitter handle we have for you is correct.

Junyuan Xie (add twitter)
Tong He (edit)
Zhi Zhang (edit)
Hang Zhang (edit)
Zhongyue Zhang (add twitter)
Mu Li (edit)
Ask The Authors

Ask the authors of this paper a question or leave a comment.

Read it. Rate it.
#1. Which part of the paper did you read?

#2. The paper contains new data or analyses that is openly accessible?
#3. The conclusion is supported by the data and analyses?
#4. The conclusion is of scientific interest?
#5. The result is likely to lead to future research?

Github
User:
None (add)
Repo:
None (add)
Stargazers:
0
Forks:
0
Open Issues:
0
Network:
0
Subscribers:
0
Language:
None
Youtube
Link:
None (add)
Views:
0
Likes:
0
Dislikes:
0
Favorites:
0
Comments:
0
Other
Sample Sizes (N=):
Inserted:
Words Total:
Words Unique:
Source:
Abstract:
None
12/04/18 06:05PM
6,347
2,056
Tweets
tawatawara: @currypurin こういう論文(https://t.co/s2R8OL6SyL)もありますけど、悩ましいですね。結局やってみるしかないという... あと最近話題に出てた bag of tricks (https://t.co/RfZXIaQGy3, ※CNNの話)でも 3.1. Large-batch training に batch size と learning rate をきめるヒューリスティクスがありました
banenberg: Bag of Tricks for Image Classification with CNNs - This thorough empirical study highlights the most effective network refinements and optimization methods to eke out a higher validation accuracy. ResNet-50-D, mixup training + more https://t.co/jFeldtiwrT @CVPR #CVPR2019 #CVML https://t.co/PqzMm3FKE8
KokiMadono: RT @tawatawara: 🏺の discussion に貼られてた Bag of Tricks、LT会用の論文探してるときに CVPR2019 (poster) にaccept されてることを知った https://t.co/RfZXIaQGy3
tawatawara: 🏺の discussion に貼られてた Bag of Tricks、LT会用の論文探してるときに CVPR2019 (poster) にaccept されてることを知った https://t.co/RfZXIaQGy3
heghbalz: If you are wondering, this is the paper: Bag of Tricks for Image Classification with Convolutional Neural Networks Tong He, Zhi Zhang, Hang Zhang, Zhongyue Zhang, Junyuan Xie, Mu Li https://t.co/xQVfPtibGV https://t.co/BxyQk9bBEM
crcrpar: https://t.co/jzOnDcNJIx
ikiYosky: あとでよむhttps://t.co/B94aEhw2xd
shunk031: 画像分類において提案されている先行研究の手法を詳細に検討し、ベースラインのResNet50にそれら提案されている手法を適用すると4%程度精度向上 / Bag of Tricks for Image Classification with Convolutional Neural Networks https://t.co/bjCuNgiygF
sermakarevich: Bag of Tricks for Image Classification with Convolutional Neural Networks https://t.co/nUItBppCKR
Swaroopkml: This paper summarizes a few tricks like label smoothing and mixup for image classification. Along with #fastai's progressive resizing and Leslie Smith's 1-cycle learning, these are my new favorites so far. https://t.co/ZvireXJgAP #DeepLearning #AI #100DaysOfMLCode
jp_axs4ll: RT @jeremyphoward: Here's a link to the paper: https://t.co/MqnfIqmUHC
SamiSakly: Looking for yet another "trick" to improve your image classification model? Here you are #ML #computervision #AI @fastdotai link to the paper: https://t.co/nxgGNO4t0l https://t.co/llhZ3zYNiQ
ThingsReallyR: RT @jeremyphoward: Here's a link to the paper: https://t.co/MqnfIqmUHC
ErmiaBivatan: RT @jeremyphoward: Here's a link to the paper: https://t.co/MqnfIqmUHC
muktabh: RT @jeremyphoward: Here's a link to the paper: https://t.co/MqnfIqmUHC
sheikmohdimran: RT @jeremyphoward: Here's a link to the paper: https://t.co/MqnfIqmUHC
MingSeow: https://t.co/w8rfoRVxXA
disigandalf: RT @jeremyphoward: Here's a link to the paper: https://t.co/MqnfIqmUHC
sigitpurnomo: RT @jeremyphoward: Here's a link to the paper: https://t.co/MqnfIqmUHC
AndrewWedemier: RT @jeremyphoward: Here's a link to the paper: https://t.co/MqnfIqmUHC
alexhock: RT @jeremyphoward: Here's a link to the paper: https://t.co/MqnfIqmUHC
Remi_Coulom: @jeremyphoward Direct link for the lazy: https://t.co/uOshX8bubC
jeremyphoward: Here's a link to the paper: https://t.co/MqnfIqmUHC
RexDouglass: mixup: BEYOND EMPIRICAL RISK MINIMIZATION https://t.co/GGwxqEPmej
unsorsodicorda: @RexDouglass 2/ data augmentation tricks such as mixup. You can find a list of useful tricks for ResNet training here: https://t.co/D4sIB6cKmf let me know if it helped!
JoshuaKhan_007: [1812.01187] Bag of Tricks for Image Classification with Convolutional Neural Networks - https://t.co/d38tFmhlOi 🖥 📣
arxiv_pop: 2018/12/03 投稿 1位 CV(Computer Vision and Pattern Recognition) Bag of Tricks for Image Classification with Convolutional Neural Networks https://t.co/0staHEoNBy 9 Tweets 34 Retweets 93 Favorites
ak_eapen: Bag of Tricks for Image Classification with Convolutional Neural Networks https://t.co/zbROBvI3ES
alxcnwy: I really enjoyed this “Bag of Tricks for Image Classification with Convolutional Neural Networks” - surprising that mixup works at all, nevermind so well! https://t.co/1askPWF7dR
Mapio: "Bag of Tricks" https://t.co/TqskeD8ia2 quite a statement on the discipline…
nskm_m: RT @yuyu2172: Cosine Learning Rate Schedulingが有効なのはOpen Imagesのコンペで確認していたけど、他のデータセットでも有効なんですね https://t.co/JWQOVRv2Tv
yuyu2172: Cosine Learning Rate Schedulingが有効なのはOpen Imagesのコンペで確認していたけど、他のデータセットでも有効なんですね https://t.co/JWQOVRv2Tv
GamerGeekNews: Cornell University Library: Bag of Tricks for Image Classification with Convolutional Neural Networks https://t.co/uEjy6DstZF #machinelearning #NeuralNetworks
reddit_ml: [R] Bag of Tricks for Image Classification with Convolutional Neural Networks https://t.co/2FMh75in4i
PanpsySchism: RT @Miles_Brundage: "Bag of Tricks for Image Classification with Convolutional Neural Networks," Xie et al.: https://t.co/sIz0tx8gUp
yasserglez: RT @Miles_Brundage: "Bag of Tricks for Image Classification with Convolutional Neural Networks," Xie et al.: https://t.co/sIz0tx8gUp
J_P_Raymond: RT @Miles_Brundage: "Bag of Tricks for Image Classification with Convolutional Neural Networks," Xie et al.: https://t.co/sIz0tx8gUp
robotic_hands: RT @Miles_Brundage: "Bag of Tricks for Image Classification with Convolutional Neural Networks," Xie et al.: https://t.co/sIz0tx8gUp
EricSchles: RT @Miles_Brundage: "Bag of Tricks for Image Classification with Convolutional Neural Networks," Xie et al.: https://t.co/sIz0tx8gUp
imenurok: Bag of Tricks for Image Classification with Convolutional Neural Networks https://t.co/Z6oZgOVnuh
quidpr: RT @Miles_Brundage: "Bag of Tricks for Image Classification with Convolutional Neural Networks," Xie et al.: https://t.co/sIz0tx8gUp
Quebec_AI: RT @Miles_Brundage: "Bag of Tricks for Image Classification with Convolutional Neural Networks," Xie et al.: https://t.co/sIz0tx8gUp
PerthMLGroup: RT @Miles_Brundage: "Bag of Tricks for Image Classification with Convolutional Neural Networks," Xie et al.: https://t.co/sIz0tx8gUp
truba_news: RT @Miles_Brundage: "Bag of Tricks for Image Classification with Convolutional Neural Networks," Xie et al.: https://t.co/sIz0tx8gUp
AssistedEvolve: RT @Miles_Brundage: "Bag of Tricks for Image Classification with Convolutional Neural Networks," Xie et al.: https://t.co/sIz0tx8gUp
ComputerPapers: Bag of Tricks for Image Classification with Convolutional Neural Networks. https://t.co/XDRw80DwDH
sandmangil: RT @Miles_Brundage: "Bag of Tricks for Image Classification with Convolutional Neural Networks," Xie et al.: https://t.co/sIz0tx8gUp
cghosh_: RT @Miles_Brundage: "Bag of Tricks for Image Classification with Convolutional Neural Networks," Xie et al.: https://t.co/sIz0tx8gUp
desertnaut: RT @Miles_Brundage: "Bag of Tricks for Image Classification with Convolutional Neural Networks," Xie et al.: https://t.co/sIz0tx8gUp
EldarSilver: RT @Miles_Brundage: "Bag of Tricks for Image Classification with Convolutional Neural Networks," Xie et al.: https://t.co/sIz0tx8gUp
awaldock: RT @Miles_Brundage: "Bag of Tricks for Image Classification with Convolutional Neural Networks," Xie et al.: https://t.co/sIz0tx8gUp
MatRazor: RT @Miles_Brundage: "Bag of Tricks for Image Classification with Convolutional Neural Networks," Xie et al.: https://t.co/sIz0tx8gUp
russell_lliu: RT @Miles_Brundage: "Bag of Tricks for Image Classification with Convolutional Neural Networks," Xie et al.: https://t.co/sIz0tx8gUp
goloskokovic: RT @Miles_Brundage: "Bag of Tricks for Image Classification with Convolutional Neural Networks," Xie et al.: https://t.co/sIz0tx8gUp
Montreal_AI: RT @Miles_Brundage: "Bag of Tricks for Image Classification with Convolutional Neural Networks," Xie et al.: https://t.co/sIz0tx8gUp
nova77t: RT @Miles_Brundage: "Bag of Tricks for Image Classification with Convolutional Neural Networks," Xie et al.: https://t.co/sIz0tx8gUp
ceobillionaire: RT @Miles_Brundage: "Bag of Tricks for Image Classification with Convolutional Neural Networks," Xie et al.: https://t.co/sIz0tx8gUp
treasured_write: RT @Miles_Brundage: "Bag of Tricks for Image Classification with Convolutional Neural Networks," Xie et al.: https://t.co/sIz0tx8gUp
BlyNotes: RT @Miles_Brundage: "Bag of Tricks for Image Classification with Convolutional Neural Networks," Xie et al.: https://t.co/sIz0tx8gUp
BrundageBot: Bag of Tricks for Image Classification with Convolutional Neural Networks. Junyuan Xie, Tong He, Zhi Zhang, Hang Zhang, Zhongyue Zhang, and Mu Li https://t.co/Okg3XWuhlS
linkoffate: RT @Miles_Brundage: "Bag of Tricks for Image Classification with Convolutional Neural Networks," Xie et al.: https://t.co/sIz0tx8gUp
Miles_Brundage: "Bag of Tricks for Image Classification with Convolutional Neural Networks," Xie et al.: https://t.co/sIz0tx8gUp
Images
Related