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
Vision Proで広告フリーな世界を実現したい
Search
Shuhei Shitamori
December 12, 2024
Programming
0
82
Vision Proで広告フリーな世界を実現したい
MIERUNE BBQ #14発表資料
Shuhei Shitamori
December 12, 2024
Tweet
Share
More Decks by Shuhei Shitamori
See All by Shuhei Shitamori
Apple SharePlayで 非対称クロスプレイ チャレンジ (SharePlay使ってみた編)
shitamori1272
0
29
Foundation Models触ってみた - iPhone Dev Sapporo — WWDC25 Recap
shitamori1272
0
81
Wallet API, Verifier APIで実現するIDカード on iPhoneの世界
shitamori1272
1
2.9k
Other Decks in Programming
See All in Programming
ネイティブ製ガントチャートUIを作って学ぶUICollectionViewLayoutの威力
jrsaruo
0
130
Go Conference 2025: Goで体感するMultipath TCP ― Go 1.24 時代の MPTCP Listener を理解する
takehaya
7
1.6k
XP, Testing and ninja testing ZOZ5
m_seki
2
280
デミカツ切り抜きで面倒くさいことはPythonにやらせよう
aokswork3
0
190
プログラマのための作曲入門
cheebow
0
540
株式会社 Sun terras カンパニーデック
sunterras
0
220
複雑化したリポジトリをなんとかした話 pipenvからuvによるモノレポ構成への移行
satoshi256kbyte
1
770
非同期jobをtransaction内で 呼ぶなよ!絶対に呼ぶなよ!
alstrocrack
0
520
GitHub Actions × AWS OIDC連携の仕組みと経緯を理解する
ota1022
0
240
Le côté obscur des IA génératives
pascallemerrer
0
120
なぜGoのジェネリクスはこの形なのか? Featherweight Goが明かす設計の核心
ryotaros
7
1k
Let's Write a Train Tracking Algorithm
twocentstudios
0
220
Featured
See All Featured
Principles of Awesome APIs and How to Build Them.
keavy
127
17k
KATA
mclloyd
32
15k
I Don’t Have Time: Getting Over the Fear to Launch Your Podcast
jcasabona
33
2.5k
Product Roadmaps are Hard
iamctodd
PRO
54
11k
Dealing with People You Can't Stand - Big Design 2015
cassininazir
367
27k
Cheating the UX When There Is Nothing More to Optimize - PixelPioneers
stephaniewalter
285
14k
Reflections from 52 weeks, 52 projects
jeffersonlam
352
21k
Documentation Writing (for coders)
carmenintech
75
5k
[Rails World 2023 - Day 1 Closing Keynote] - The Magic of Rails
eileencodes
36
2.5k
Statistics for Hackers
jakevdp
799
220k
The World Runs on Bad Software
bkeepers
PRO
71
11k
How to train your dragon (web standard)
notwaldorf
96
6.3k
Transcript
Լɹपฏ 2024/11/20 Vision ProͰࠂϑϦʔͳੈքΛ ࣮ݱ͍ͨ͠ MIERUNE BBQ
Լ पฏ w ࠓͷ݄͔Βࡳຈࡏॅ ग़Γ w Χφμͷσδλϧ*%ελʔτΞοϓͰJ04ΤϯδχΞ w 7JTJPO1SPങͬͨ
2 #MVFTLZ -JOLFEJO
XRͱ • Ծੈքͱݱ࣮ੈքͷΈ߹ΘͤʹΑͬͯ ৽ͨͳମݧΛੜΈग़ٕ͢ज़ͷ૯শ https://www.canon-its.co.jp/solution/mr/vr-ar-mr/
AR(֦ுݱ࣮)ͰͰ͖Δମݧ https://k-tai.watch.impress.co.jp/docs/news/1203694.html
ใྔ͕ଟ͍ͱετϨε • ใྔ͕ଟ͍ͱετϨεΛײ͍͢͡ • λεΫύϑΥʔϚϯεʹӨڹ • ใΛݮΒͨ͢ΊͷऔΓΈॏཁ
Diminished Reality(ݮଛݱ࣮) • ARͱରʹ࣮ࡍʹଘࡏ͢ΔͷΛϦΞϧλΠϜͰݟ͑ͳ͘͢Δٕज़ • ΠϠϗϯͷϊΠζΩϟϯηϦϯάͷࢹ֮όʔδϣϯ https://solution.itage.jp/2021/12/16/16738/
Vision ProͰDRΛͬͯΈΔ • ֗த͔ΒࠂுΓࢴΛফͯ͠ೝෛՙͷ͍ੈքΛ࣮ݱ͍ͨ͠
Vision Proͱ • Apple͕2023ʹൢച։࢝ͨ͠MRϔουηοτ • MacBookͱಉͷM2 νοϓࡌ • ߴղ૾ͷө૾ •
ϓϥΠόγʔΛྀͨ͠ମݧઃܭ
ࠂDRʹඞཁͳٕज़ Vision Proͷಛٕज़ཁ݅ʹϚον͍ͯ͠Δʂ • 1. ࢹ֮ใ(Χϝϥ)͔Βࠂͷݕग़ • Vision ProʹߴੑೳͳΧϝϥ͕ࡌ͞ΕͯΔ •
2. ࠂΛফͨ͢Ίͷഎܠը૾Λੜ • Vision ProʹAIΛಈ͔ͨ͢Ίͷߴੑೳͳνοϓ͕ࡌ͞ΕͯΔ • 3. ੜͨ͠ը૾ΛࠂʹॏͶͯදࣔ • Vision ProʹԾମΛۭؒʹஔͰ͖Δ
ࠂDRʹඞཁͳٕज़ Vision Proͷಛٕज़ཁ݅ʹϚον͍ͯ͠Δʂ͠ͳ͍… • 1. ࢹ֮ใ(Χϝϥ)͔Βࠂͷݕग़ • Vision ProʹߴੑೳͳΧϝϥ͕ࡌ͞ΕͯΔ͕ɺΧϝϥө૾ʹΞΫηεͰ͖ͳ͍ •
2. ࠂΛফͨ͢Ίͷഎܠը૾Λੜ • Vision ProʹAIΛಈ͔ͨ͢Ίͷߴੑೳͳνοϓ͕ࡌ͞ΕͯΔ͕ɺAI༻ΤϯδϯʹΞΫηεͰ͖ͳ͍ • 3. ੜͨ͠ը૾ΛࠂʹॏͶͯදࣔ • Vision ProʹԾମΛۭؒʹஔͰ͖Δ
ࠓճͷॲཧϑϩʔ • PCͰࣄલʹʮݕ͍ͨ͠ࠂʯͱʮॏͶ͍ͨഎܠը૾ͷੜʯΛΔ • Vision Proʮੜͨ͠ը૾ΛࠂʹॏͶͯදࣔʯ͚ͩ • ࢹ֮ใ(Χϝϥ)͔Β ࠂͷݕग़ 1.
ࠂΛࡱӨͯ͠σʔληοτ࡞ 2. ը૾͔ΒࠂྖҬΛݕग़ 3. ը૾ੜ༻ʹྖҬΛܗ ࠂΛফͨ͢Ίͷ എܠը૾ͷੜ 1. ࠂʹସΘΔഎܠը૾Λੜ 2. Vision Pro༻ʹը૾Λܗ ੜͨ͠ը૾Λ ࠂʹॏͶͯදࣔ 1. Vision Pro༻ͷΞϓϦΛ࡞ PCͰࣄલʹ AVPͰϦΞϧλΠϜʹ
ࢹ֮ใ(Χϝϥ)͔Βࠂͷݕग़ ࠂΛࡱӨͯ͠σʔληοτ࡞ • ԼమӺʹܝࡌ͞ΕͨࠂுΓࢴΛΧϝϥͰࡱӨ
ࢹ֮ใ(Χϝϥ)͔Βࠂͷݕग़ ը૾͔ΒࠂྖҬΛݕग़ • ը૾͔ΒҙͷΦϒδΣΫτΛݕग़Ͱ͖Δ Segmented Anything Model(SAM)ϕʔεͷϞσϧΛར༻ • prompt=“Advertisement”Ͱ֘͢ΔྖҬΛࣗಈͰݕग़ ”Advertisement”
https://github.com/hustvl/EVF-SAM
ࢹ֮ใ(Χϝϥ)͔Βࠂͷݕग़ ྖҬݕग़݁Ռͷܗ • ΪβΪβ݀ݕग़ྖҬͱͯ͠ෆਖ਼֬ͳͷͰ࢛֯ܗͱͯ͠ܗ • ܗޙͷը૾Λॏը૾ੜͷϚεΫͱͯ͠ར༻
ࠂΛফͨ͢Ίͷഎܠը૾ͷੜ ݕग़ྖҬΛੜAIͰ࠶ඳը • ը૾෮ݩʹಛԽͨ͠stable-di ff usionϞσϧͰϚεΫྖҬΛ࠶ੜ ޭ ύλʔϯ ࣦഊ ύλʔϯ
https://huggingface.co/stabilityai/stable-di ff usion-2-inpainting
ࠂΛফͨ͢Ίͷഎܠը૾ͷੜ Vision ProͰ͏σʔληοτΛ࡞ • Vision ProͰར༻͢Δը૾σʔληοτΛ࡞ • ݕग़༻ͷReference, ॏ༻ͷGeneratedΛ࡞ Reference
Generated
ੜͨ͠ը૾ΛࠂʹॏͶͯදࣔ Vision ProΞϓϦͷ࡞ • Reference ImageΛݕग़ͨ͠Β Generated ImageΛಉ͡Ґஔʹදࣔͤ͞ΔγϯϓϧͳΞϓϦ Reference ൃݟʂ
Generated ࠂ্ʹ ॏͶͯදࣔʂ
͍͟ɺӺߏͰ࣮ݧ ͦͷ1
͍͟ɺӺߏͰ࣮ݧ ͦͷ2
݁Ռ • Reference ImageΛ͖ͪΜͱݕग़ͯ͠Generated ImageΛॏදࣔͰ͖ͨ • ҰํͰ • ࠂʹेʹ͔ۙͮͳ͍ͱVision Pro͕ReferenceΛݕ͠ͳ͍
• Generated Imageʹมͳͷ͕ೖΓɺʮফ͢ʯ͜ͱ͕Ͱ͖ͳ͍࣌͋ͬͨ
ࠓޙվળ͢ΔͳΒ • ेʹࠂʹ͔ۙͮͳ͍ͱVision Pro͕ReferenceΛݕ͠ͳ͍ -> ͋Β͔͡Ί্ۭؒʹGenerated ImageΛஔ͓͚ͯ͠ ReferenceͷݕΛඞཁͱ͠ͳ͍͔ʁ • Generated
Imageʹมͳͷ͕ೖΓɺʮফ͢ʯ͜ͱ͕Ͱ͖ͳ͍࣌͋ͬͨ -> SAMʹΑΔࠂྖҬݕग़ͷਫ਼͕ෆेͩͬͨͷͰɺଞͷख๏ݕ౼ -> Generated ImageΛੜ͢Δࡍʹ”น”ͱ໌ࣔͯ͠ྑ͔͔ͬͨ