Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Features
Speaker Deck
PRO
Sign in
Sign up for free
Search
Search
Food Image Object Detection and Classification
Search
Leszek Rybicki
February 16, 2017
Research
2
15k
Food Image Object Detection and Classification
Part 1: Detection
Leszek Rybicki
February 16, 2017
Tweet
Share
More Decks by Leszek Rybicki
See All by Leszek Rybicki
Let's talk about Fakes
lunardog
0
110
How to Patch Image Classifiers
lunardog
0
1.8k
Towards Realistic Predictors - EN
lunardog
0
1.7k
Towards Realistic Predictors
lunardog
1
2k
Deep Learning Hot Dog Detector
lunardog
0
240
Finding beans in burgers: paper reading notes
lunardog
0
1.3k
Kelner: Serve Your Models
lunardog
0
100
Image Analysis at Cookpad
lunardog
1
1.6k
Kelner: serve your models
lunardog
1
350
Other Decks in Research
See All in Research
Human-Informed Machine Learning Models and Interactions
hiromu1996
2
550
Global Evidence Summit (GES) 参加報告
daimoriwaki
0
220
ベイズ的方法に基づく統計的因果推論の基礎
holyshun
0
710
アプリケーションから知るモデルマージ
maguro27
0
230
Leveraging LLMs for Unsupervised Dense Retriever Ranking (SIGIR 2024)
kampersanda
2
280
Large Vision Language Model (LVLM) に関する最新知見まとめ (Part 1)
onely7
23
5.4k
機械学習による言語パフォーマンスの評価
langstat
6
860
Tiaccoon: コンテナネットワークにおいて複数トランスポート方式で統一的なアクセス制御
hiroyaonoe
0
260
marukotenant01/tenant-20240916
marketing2024
0
650
データサイエンティストをめぐる環境の違い 2024年版〈一般ビジネスパーソン調査の国際比較〉
datascientistsociety
PRO
0
930
Weekly AI Agents News! 11月号 論文のアーカイブ
masatoto
0
260
医療支援AI開発における臨床と情報学の連携を円滑に進めるために
moda0
0
140
Featured
See All Featured
Product Roadmaps are Hard
iamctodd
PRO
50
11k
Faster Mobile Websites
deanohume
305
30k
Improving Core Web Vitals using Speculation Rules API
sergeychernyshev
3
180
実際に使うSQLの書き方 徹底解説 / pgcon21j-tutorial
soudai
173
51k
Design and Strategy: How to Deal with People Who Don’t "Get" Design
morganepeng
127
18k
Building an army of robots
kneath
302
45k
Raft: Consensus for Rubyists
vanstee
137
6.7k
Code Reviewing Like a Champion
maltzj
521
39k
Statistics for Hackers
jakevdp
797
220k
Intergalactic Javascript Robots from Outer Space
tanoku
270
27k
Dealing with People You Can't Stand - Big Design 2015
cassininazir
365
25k
Become a Pro
speakerdeck
PRO
26
5.1k
Transcript
Food Image Object Detection and Classification Challenges and Solutions
Part 1: Detection
自己紹介 • リビツキ レシェック • ポーランド出身 • 2016~ クックパッド • github:
lunardog
Warning! This presentation contains images that may cause severe drooling
and stomach grumbling. @cookpad
History 歴史
ImageNet KWWSLPDJHQHWRUJ
ImageNet Large Scale Visual Recognition Competition KWWSZZZLPDJHQHWRUJFKDOOHQJHV/695&
ILSVRC 2010 task Classification )RUHDFKLPDJHDOJRULWKPV ZLOOSURGXFHDOLVWRIDWPRVW REMHFWFDWHJRULHVLQWKH GHVFHQGLQJRUGHURI FRQILGHQFH KWWSZZZLPDJHQHWRUJFKDOOHQJHV/695&
ILSVRC 2011 tasks 1. Classification 2. *Classification with localization *tester
task
KWWSFVQVWDQIRUGHGXV\OODEXVKWPO Classification + Localization
ILSVRC 2012 tasks 1. Classification 2. Classification with localization 3.
Fine-grained classification
Fine-grained classification KWWSZZZLPDJHQHWRUJFKDOOHQJHV/695&
AlexNet ,PDJHQHWFODVVLILFDWLRQZLWKGHHSFRQYROXWLRQDOQHXUDOQHWZRUNV $.UL]KHYVN\,6XWVNHYHU*(+LQWRQ$GYDQFHVLQQHXUDOLQIRUPDWLRQ SURFHVVLQJV\VWHPV
ILSVRC 2013 tasks 1. Detection 2. Classification 3. Classification with
localization
ILSVRC 2014 tasks 1. Detection 2. Classification 3. Classification with
localization
Object Detection KWWSFVQVWDQIRUGHGXV\OODEXVKWPO
Deep Learning KWWSVGHYEORJVQYLGLDFRP
ILSVRC 2015 tasks 1. Object detection 2. Object localization 3.
*Object detection from video 4. *Scene classification
ILSVRC 2016 tasks 1. Object localization 2. Object detection 3.
Object detection from video 4. Scene classification 5. Scene parsing
Cookpad 2016
画像データセット 1997年~ レシピ数:国内約260万 + 国外 + つくれぽ + 手順写真 17言語、60カ国
※数字は2017年02月時点のものです
画像解析の研究関心 • これは料理ですか? • どの料理ですか? • 料理はどこですか? • 。。。 Part
2
Where is the food? 料理はどこですか?
ゴール )LQGIRRGLQWKHLPDJHGUDZ DERXQGLQJER[DURXQGWKH IRRGLWHPLQFOXGLQJWKH GLVKLIYLVLEOH
,IWKHUHDUHPXOWLSOHLWHPV GUDZDERXQGLQJER[ DURXQGHDFKRQH ゴール
ground truth bounding box > 0.9 We count it as
a positive detection if Intersection over Union ratio is greater than 0.9. ƴ
QXPEHURIWUXHSRVLWLYHV QXPEHURIJURXQGWUXWKER[HV ƴ ƴ ƴ QXPEHURIWUXHSRVLWLYHV QXPEHURIJHQHUDWHGER[HV 再現率 (precision) (recall)
ƴ ƴ
Methods
1. Build a classifier 2. Pick Regions of Interest 3.
Run classifier on each region 4. Remove duplicate detections IDEA
Fast, Faster R-CNN 5LFKIHDWXUHKLHUDUFKLHVIRUDFFXUDWHREMHFWGHWHFWLRQDQGVHPDQWLFVHJPHQWDWLRQ 5RVV*LUVKLFN-HII'RQDKXH7UHYRU'DUUHOO-LWHQGUD0DOLN )DVWHU5&117RZDUGV5HDO7LPH2EMHFW'HWHFWLRQZLWK5HJLRQ3URSRVDO1HWZRUNV 6KDRTLQJ5HQ.DLPLQJ+H5RVV*LUVKLFN-LDQ6XQ
)DVW5&11 5RVV*LUVKLFN
問題 1. Computational cost 2. Context is important 3. ...but
context can be confusing. KDQG IRRG JUDVV IRRG KWWSSL[DED\FRP
Single Shot Detector 66'6LQJOH6KRW0XOWL%R['HWHFWRU :HL/LX'UDJRPLU$QJXHORY'XPLWUX(UKDQ&KULVWLDQ6]HJHG\ 6FRWW5HHG&KHQJ<DQJ)X$OH[DQGHU&%HUJ
Either The Least Or Most Employable Person Ever 7KH+XIILQJWRQ3RVW JLWKXEFRPSMUHGGLH
SMUHGGLHFRPGDUNQHW ZZZNDJJOHFRPSMUHGGLH Joseph Redmon
You Only Look Once <RX2QO\/RRN2QFH8QLILHG 5HDO7LPH2EMHFW'HWHFWLRQ -RVHSK5HGPRQ6DQWRVK'LYYDOD5RVV *LUVKLFN$OL)DUKDGL 'HF
<2/2%HWWHU)DVWHU 6WURQJHU -RVHSK5HGPRQ$OL)DUKDGL
<RX2QO\/RRN2QFH8QLILHG5HDO7LPH2EMHFW'HWHFWLRQ -RVHSK5HGPRQ6DQWRVK'LYYDOD5RVV*LUVKLFN$OL)DUKDGL YOLO in Context
None