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
Synthesizing Human Images in SIGGRAPH Asia 2021
Search
Sponsored
·
SiteGround - Reliable hosting with speed, security, and support you can count on.
→
Udon
February 27, 2022
Technology
190
0
Share
Synthesizing Human Images in SIGGRAPH Asia 2021
SIGGRAPH Asia 2021の「Synthesizing Human Images」セッションに採択された5本の論文を紹介します.
Udon
February 27, 2022
More Decks by Udon
See All by Udon
MIRU2024_招待講演_RALF_in_CVPR2024
udonda
1
440
[CVPR24 Oral] Retrieval-Augmented Layout Transformer for Content-Aware Layout Generation
udonda
0
380
Survey of Image Editing with GANs in SIGGRAPH'21
udonda
2
940
Network-to-Network Translation with Conditional Invertible Neural Networks
udonda
1
350
DARTS: Differentiable Architecture Search
udonda
0
170
Other Decks in Technology
See All in Technology
関西に縁あるMicrosoft MVPsが語るCopilotの未来
kasada
0
720
Claude code Orchestra
ozakiomumkj
3
780
Javaで学ぶSOLID原則
negima
1
240
Diagnosing performance problems without the guesswork
elenatanasoiu
0
130
long-running-tasks
cipepser
2
450
AIガバナンス実践 - 生成AIコネクタのデータ漏洩リスクと実務対策
knishioka
0
150
AI フレンドリーなエラー監視を TypeScript で実現する
shinyaigeek
2
200
Kiro CLI v2.0.0がやってきた!
kentapapa
0
240
類似画像検索モデルの開発ノウハウ
lycorptech_jp
PRO
4
1.1k
JJUG CCC 2026 Spring AI時代の開発こそ標準化を武器に! ― 方式・プロセス・プラットフォームの標準化
s27watanabe
2
640
開発を止めない CI/CD ~CI Visibilityによる継続的最適化~
pensuke628
0
220
大学生が本気でDatabricksを活用してDiscordサークルをデータ駆動させてみた
phantomjuju
1
300
Featured
See All Featured
Learning to Love Humans: Emotional Interface Design
aarron
275
41k
Noah Learner - AI + Me: how we built a GSC Bulk Export data pipeline
techseoconnect
PRO
0
190
First, design no harm
axbom
PRO
2
1.2k
How to make the Groovebox
asonas
2
2.2k
The AI Search Optimization Roadmap by Aleyda Solis
aleyda
1
5.8k
Why Mistakes Are the Best Teachers: Turning Failure into a Pathway for Growth
auna
0
150
The Illustrated Children's Guide to Kubernetes
chrisshort
51
52k
The untapped power of vector embeddings
frankvandijk
2
1.7k
Prompt Engineering for Job Search
mfonobong
0
320
Building AI with AI
inesmontani
PRO
1
1k
Site-Speed That Sticks
csswizardry
13
1.2k
AI Search: Implications for SEO and How to Move Forward - #ShenzhenSEOConference
aleyda
1
1.3k
Transcript
%BJDIJ)PSJUBൃද 4*((3"1)"TJB Synthesizing Human Images Session
୭ʁ 2
Synthesizing Human Images Session • URL: https://sa2021.siggraph.org/jp/attend/technical-papers/8/session/73 • 5ຊͷจ͕࠾ •
Barbershop • EyelashNet • Neural Actor • Pose with Style • SketchHairSalon 3
4
Barbershop ઃఆ — GAN inversion-based hair editing 5 Face identity
Generated image Hair color condition Hair texture condition Hair shape condition
• ͱإͷؒͷϒϨϯυΞʔςΟϑΝΫτΛੜͤ͡͞ͳ͍ੜ Barbershop ݁Ռ 6
Barbershop ख๏ 7 ͷલΛ” ” W+ F ͷޙΛ” ” W+
S C = (F, S) Face Hair ೖྗ ࠶ߏ Swap S ֤ೖྗ݅ ͷద༻ Blend ݟͨͷ࠷దԽ Face Identity Hair style
Barbershop ੍ݶ 8 • ᶃϚεΫ͕ͳ͍෦͋·Γ៉ྷʹ ͳΒͳ͍ • ᶄᶆϐΞεͷΑ͏ͳαϯϓϧ ແཧ •
ᶅᶇإΛःΔΑ͏ͳੜແཧ
9
EyelashNet ಈػ 10 ·ͭ͛ͷ3D࠶ߏࠔ ʢमਖ਼ʹ5࣌ؒఔඞཁʣ ·ͭ͛ͷMattingΛߦ͍ আڈ͔ͯ͠Βͷ3D࠶ߏ៉ྷʹͰ͖Δ with ·ͭ͛ w/o
·ͭ͛
EyelashNet ·ͭ͛σʔληοτ࡞ͷ 11 • ·ͭ͛άϦʔϯόοΫࡱӨͷΑ͏ͳํ๏ͰࡱӨͰ͖ͳ͍ • എܠʹͰͳ͘ɼલܠʹʢ·ͭ͛ʹܬޫృྉΛృͬͯࡱӨʣ
EyelashNet ·ͭ͛σʔληοτ࡞ͷ 12 • 2ຕͷࡱӨͷؒʹਓಈ͍ͯ͠·͏ͷͰ…ʁʁ • Optical flowͰҐஔ߹Θͤ XPҐஔ߹Θͤ XҐஔ߹Θͤ
EyelashNet ݁Ռ 13
14
Neural Actor ֓ཁ • ҙͷࢹɾ੍ޚՄೳͳϙʔζͰͷߴ࣭ͳਓؒ߹ͷͨΊͷख๏ • ϙʔζ<->ඪ४ۭؒͷมΛֶश 15
16 Neural Actor ܇࿅
NeuralActor ৽نࢹ߹࣮ݧ 17 ఆྔൺֱ ʢFID͕ۃʹྑ͍; γʔέϯεͰݟͨ࣌ʹ༏ΕΔʣ ఆੑൺֱ
18
Pose with Style ֓ཁ 19 • UVϚοϓ͕݅ͳStyleGAN2ʹΑΔશը૾ੜ ೖྗ ग़ྗ ਓੜ
Ծࢼண ೖྗ ग़ྗ
Pose with Style ܇࿅ 20 Ґஔ߹Θͤ 5BSHFUDPPSE HFOFSBUPS
Pose with Style ࣮ݧ — vs. Img-to-imgมϞσϧͱͷൺֱ 21 • ߴप៉ྷʹ
ੜ • ͷ༷ • إͷৄࡉ
Pose with Style ࣮ݧ — vs. StyleGAN-basedϞσϧͱͷൺֱ 22 • StylePoseGAN[Liu+
arXiv21] • Pose with StyleͷUVϚοϓʹର͢ ΔΛ΄ͱΜͲແͨ͘͠Α͏ͳ ख๏ • UVϚοϓΛ࠷େݶར༻͢Εߴ࣭
23
SketchHairSalon σʔληοτ࡞ 24 ೖྗ soft mask ࣗಈੜ खಈ •
soft maskMattingख๏Ͱ ਪ • ࣗಈੜͰฤΈࠐΈͳͲΛ දݱͰ͖ͳ͍ • खಈσʔληοτ͕ඞਢ
SketchHairSalon σʔληοτ࡞ 25 • ਪ࣌ʹ{ฤΈࠐΈ, ςΫενϟ}Λॻ͔ͤΔͷඇৗʹ໘͍͘͞ͷͰ…? • ύϥϝτϦοΫͳදݱؔΛఏҊ ฤΈࠐΈ ඇฤΈࠐΈ
SketchHairSalon ख๏ 26
SketchHairSalon ࣮ݧ 27
SketchHairSalon ੍ݶ 28 • Ԟߦ͖͕ඞཁͳܕݫ͍͠ • ແݶໟଋܕϑϥοτ ͳੜ • רࢴֶशαϯϓϧͱಉ͡
ੜ • Ԟߦ͖ߟྀͨ͠ੜͰղܾ(?)
·ͱΊ • ਓυϝΠϯͷΛѻ͏5ຊͷจΛհ • ฤू • Animatable NeRF • ·ͭ͛
• શੜ • ݻ༗ͷΛͲ͏ղܾ͢Δ͔ʁΛ͔ͳΓ۩ମతͳΞϓϩʔνͰղܾ͍ͯ͠Δ จ͕ଟ͔ͬͨ 29 Synthesizing Human Images Session