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
色恒常性仮説に基づく色補正ライブラリcolorcorrect / 2015-01-31-kan...
Search
shunsukeaihara
January 30, 2015
Technology
3
2.5k
色恒常性仮説に基づく色補正ライブラリcolorcorrect / 2015-01-31-kantocv27
@関東CV勉強会 2015/01/31
shunsukeaihara
January 30, 2015
Tweet
Share
More Decks by shunsukeaihara
See All by shunsukeaihara
BONXを支える技術:発話区間検出(VAD)の話/Akerun & BONX Tech Talk
shunsukeaihara
4
7.7k
Goのnet.TCPConnの話/shibuya.go01
shunsukeaihara
3
820
Norikra in Gunosy Network Ads@Norikra meetup #2
shunsukeaihara
1
6k
LevelDB on S3 As A KVS
shunsukeaihara
1
2.8k
Sparkによる分散処理 / 2015-01-16 PyData.Tokyo#3
shunsukeaihara
11
3.5k
ゼロから始めた Gunosyアドサーバ開発運用記 / 2014-12-16-dots
shunsukeaihara
6
1.2k
Gunosy.Go#5 index/io/log
shunsukeaihara
0
160
Gunosy.go#2 package/compress
shunsukeaihara
0
110
Other Decks in Technology
See All in Technology
AI時代に必要なデータプラットフォームの要件とは by @Kazaneya_PR / 20251107
kazaneya
PRO
4
740
GTC 2025 : 가속되고 있는 미래
inureyes
PRO
0
150
How Fast Is Fast Enough? [PerfNow 2025]
tammyeverts
2
270
datadog-incident-management-intro
tetsuya28
0
120
AIがコードを書いてくれるなら、新米エンジニアは何をする? / komekaigi2025
nkzn
25
17k
LLM APIを2年間本番運用して苦労した話
ivry_presentationmaterials
10
8.6k
GPUをつかってベクトル検索を扱う手法のお話し~NVIDIA cuVSとCAGRA~
fshuhe
0
390
re:Inventに行くまでにやっておきたいこと
nagisa53
0
1.1k
触れるけど壊れないWordPressの作り方
masakawai
0
680
どうなる Remix 3
tanakahisateru
0
250
Copilotの精度を上げる!カスタムプロンプト入門.pdf
ismk
6
1.5k
進化する大規模言語モデル評価: Swallowプロジェクトにおける実践と知見
chokkan
PRO
3
470
Featured
See All Featured
Visualization
eitanlees
150
16k
Practical Orchestrator
shlominoach
190
11k
Designing for humans not robots
tammielis
254
26k
Into the Great Unknown - MozCon
thekraken
40
2.1k
Writing Fast Ruby
sferik
630
62k
KATA
mclloyd
PRO
32
15k
The Illustrated Children's Guide to Kubernetes
chrisshort
51
51k
How to Ace a Technical Interview
jacobian
280
24k
RailsConf & Balkan Ruby 2019: The Past, Present, and Future of Rails at GitHub
eileencodes
140
34k
Speed Design
sergeychernyshev
32
1.2k
The Cult of Friendly URLs
andyhume
79
6.7k
Why You Should Never Use an ORM
jnunemaker
PRO
60
9.6k
Transcript
colorcorrect ৭߃ৗੑԾઆʹجͮ͘৭ิਖ਼ϥΠϒϥϦ Gunosy inc Shunsuke Aihara
ࣗݾհ • ҄൧ݪढ़հ (http://argmax.jp) @shunsukeaihara • GunosyͷϚωʔδϟʔ • ࠂ৴γεςϜͷ։ൃશମͱR&DܥΛ୲ •
ઐ: ܭࢉݴޠֶ • PythonͱඇಉظࢄγεςϜΛΉ • DNN࣭มͱ͔झຯͰͬͯΔ • ը૾ॲཧɾԻ৴߸ॲཧͰ͍Ζ͍ΖϥΠϒϥϦ࡞ͬͯΔ • https://bitbucket.org/aihara
͜Εը૾ೝࣝͷͨΊͷ ຯͳલॲཧͷͰ͢
colorcorrect
৭߃ৗੑ/໌Δ͞߃ৗੑͷԾઆʹج͖ͮ ม৭ɾwhite baranceͷζϨΛࣗಈิਖ਼
ը૾ͷޫݯ৭ͱϗϫΠτόϥϯε ࣮σʔλͰόϥόϥ • ৭Λ༻͍ͨମೝࣝΛߦ͓͏ͱ͢Δͱɺֶशσʔληοτͱ৭͕ͣΕ͍ͯΔͱఆਫ਼ མͪΔ • ৭ใΛ༻͍ͨBag-Of-Keypointܥͷख๏Ͱ݁ߏ৭ิਖ਼͕ޮ͍ͯ͘Δ • ྨࣅը૾ݕࡧ /
ܠ؍ը૾͔Βͷମݕग़ͱ͔࡞ͬͨ࣌ʹ૬ϋϚͬͨ • Deep Learningͱ͔͍Ζ͍Ζظͯ͠Δ͚Ͳɺ࣮ࡍʹࣄͰֶशσʔληοτूΊΔͱ ͳΔͱ͍ΖΜͳࡱӨ݅ͷը૾ूΊΔͷ݁ߏେม • ෳর໌݅ʹର͢ΔData AugmentationΑ͘Θ͔Βͳ͍ • ͋ͱɺը૾ͷݟͨΩϨΠʹ͍ͨ͠ͱ͔͍Ζ͍Ζ͋Δ • ͳΜͱָ͔͔ͨͬͨ͠ͷͰิਖ਼ख๏ͷจͱ͔͍Ζ͍Ζௐͨ
༻్ • ม৭ͨࣸ͠ਅͷิਖ਼ • ઃఆϛεͬͯࡱͬͨࣸਅͷमਖ਼ • ίϯσδͰઃఆؒҧ͑ͯࡱͬͨࣸਅͱ͔ • ମೝࣝͷલॲཧ •
దʹެ։͓͍ͯͨ͠ΒCVPR2013ͷจͰ࣮ ࡍʹΘΕͨʂ
͋ͱ᧙৭ը૾Λ͔ͨͬͨ͠ ͜Ε৭ͷใ͕ࣦΘΕ͍ͯΔͷͰͲ͏ͬͯϜϦ
pip install colorcorrect
৭߃ৗੑ (Color Constancy) ໌Δ͞߃ৗੑ (lightness constancy)
• র໌ޫͷεϖΫτϧ͕มԽͯ͠ɺମ৭ͷ ೝେ͖͘มԽ͠ͳ͍(ݩͷ৭͕ೝࣝͰ͖Δ) ৭߃ৗੑ
νΣοΧʔγϟυʔࡨࢹ ໌Δ͞ͱਓ͕ؒೝͨ͠ͷͱҟͳΔ ྡ෦ҐͷίϯτϥετʹӨڹΛड͚ͯҟͳΔೝ 'SPN8JLJNFEJB$PNNPOT
৭ͷ֮ͷ؆୯ͳϞσϧ • ʹೖࣹ͢ΔޫޫݯޫͱରͷޫࣹͰܾ· Δ • ৭߃ৗੑೖࣹޫ͔ΒޫݯޫΛਪఆͯ͠ɺޫࣹ Λਪఆ͢Δͱఆٛग़དྷΔ S( ) =
I( ) · R( ) • S(λ) ೖࣹޫ • I(λ) ޫݯޫ • R(λ) ମͷޫࣹ
৭߃ৗੑԾઆͷ؆୯ͳ • փ৭Ծઆ • ന৭ޫݯԼͷ߹ɺࢹ֮ͷ৭ͷฏۉΛऔΔͱփ৭ (127,127,127)ʹͳΔͣͳͷͰͦͷζϨΛิਖ਼͢Δ • ً - ৭૬ؔ
• ࢹ֮ͷ৭ʹภΓ͕͋Δ߹փ৭ԾઆͰ߃ৗੑΛ୲อग़ དྷͳ͍ͷͰɺًͱ৭ͷ૬͔ؔΒ৭ͷภΓ͕ޫݯʹΑΔ͔ݩ ͷମ৭ʹΑΔ͔ਪఆͯ͠ิਖ਼ • ͜ΕΒΛϕʔεʹ༷ʑͳΞϧΰϦζϜ͕ఏҊ͞Ε͍ͯΔ
Algorithms
࣮ͨ͠৭߃ৗੑΞϧΰϦζϜ • gray world • max white • stretch •
retinex, retinex with adjust • weighted grey world • standard deviation weighted grey world • luminance weighted gray world • standard deviation and luminance weighted gray world • automatic color equalization • ->ώϡʔϦεςΟΫε͚ͩͲิਖ਼ྗ͕͍͢͝
Implementation
࣮ • PythonʹΑΔ࣮(2ܥ) • PIL • Numpy • ͕ඞཁͳ෦C++ͷϥΠϒϥϦΛ numpy.ctypesͰϥοϓ
• pip install colorcorrectͰར༻Մೳ
DEMO http://colorcorrect.argmax.jp/
·ͱΊ • खܰʹը૾ͷ৭ิਖ਼͕ग़དྷͯศརͳͷͰͬͯΈ ͍ͯͩ͘͞ • Python3ܥ + PillowରԠͦΖͦΖ͠·͢ • ԻͷલॲཧϥΠϒϥϦͱ͔ެ։ͯ͠·͢
• https://bitbucket.org/aihara/pyssp • Gunosyʹମೝࣝɾը૾clipͷࣄ͋Γ·͢