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
GNeRF: GAN-based Neural Radiance Field without ...
Search
Sponsored
·
Ship Features Fearlessly
Turn features on and off without deploys. Used by thousands of Ruby developers.
→
peisuke
December 12, 2021
Technology
1
830
GNeRF: GAN-based Neural Radiance Field without Posed Camera
第9回全日本コンピュータビジョン勉強会
ICCV2021論文読み回発表資料
GNeRF: GAN-based Neural Radiance Field without Posed Camera
peisuke
December 12, 2021
Tweet
Share
More Decks by peisuke
See All by peisuke
Moto: Latent Motion Token as the Bridging Language for Learning Robot Manipulation from Videos
peisuke
0
220
VGGT: Visual Geometry Grounded Transformer
peisuke
1
1.6k
AI for Kids:小学生に画像認識を教えてみた話
peisuke
1
86
LangGraphで始めるマルチエージェントシステム
peisuke
14
4.9k
Self-RAG: Learning to Retrieve, Generate and Critique through Self-Reflections
peisuke
9
1.6k
Mip-NeRF 360: Unbounded Anti-Aliased Neural Radiance Fields
peisuke
0
13k
LangChain Toolsの運用と改善
peisuke
5
2.9k
TTS Skins: Speaker Conversion via ASR
peisuke
0
430
A Quantum Computational Approach to Correspondence Problems on Point Sets
peisuke
0
780
Other Decks in Technology
See All in Technology
Oracle Cloud Infrastructure:2026年2月度サービス・アップデート
oracle4engineer
PRO
0
200
トラブルの大半は「言ってない」x「言ってない」じゃねーか!!
ichimichi
0
300
社内ワークショップで終わらせない 業務改善AIエージェント開発
lycorptech_jp
PRO
1
450
【5分でわかる】セーフィー エンジニア向け会社紹介
safie_recruit
0
44k
What's new in Go 1.26?
ciarana
2
280
AWS Bedrock Guardrails / 機密情報の入力・出力をブロックする — Blocking Sensitive Information Input/Output
kazuhitonakayama
2
190
Eight Engineering Unit 紹介資料
sansan33
PRO
1
6.9k
Exadata Fleet Update
oracle4engineer
PRO
0
1.3k
クラウド時代における一時権限取得
krrrr38
1
150
作るべきものと向き合う - ecspresso 8年間の開発史から学ぶ技術選定 / 技術選定con findy 2026
fujiwara3
6
1.8k
Serverless Agent Architecture on Azure / serverless-agent-on-azure
miyake
1
130
AI が Approve する開発フロー / How AI Reviewers Accelerate Our Development
zaimy
1
260
Featured
See All Featured
[Rails World 2023 - Day 1 Closing Keynote] - The Magic of Rails
eileencodes
38
2.8k
svc-hook: hooking system calls on ARM64 by binary rewriting
retrage
1
140
Avoiding the “Bad Training, Faster” Trap in the Age of AI
tmiket
0
96
Gemini Prompt Engineering: Practical Techniques for Tangible AI Outcomes
mfonobong
2
300
Navigating the Design Leadership Dip - Product Design Week Design Leaders+ Conference 2024
apolaine
0
210
Navigating Weather and Climate Data
rabernat
0
130
Git: the NoSQL Database
bkeepers
PRO
432
66k
Typedesign – Prime Four
hannesfritz
42
3k
Marketing Yourself as an Engineer | Alaka | Gurzu
gurzu
0
140
The Cult of Friendly URLs
andyhume
79
6.8k
Evolving SEO for Evolving Search Engines
ryanjones
0
150
Making Projects Easy
brettharned
120
6.6k
Transcript
GNeRF: GAN-based Neural Radiance Field without Posed Camera @peisuke 9
ABEJA Twitter @peisuke Github https://github.com/peisuke Qiita https://qiita.com/peisuke SlideShare https://www.slideshare.net/FujimotoKeisuke
• • GNeRF: GAN-based Neural Radiance Field without Posed Camera
• • NeRF • • • GAN Radiance Field • RF
NeRF • / 5 •
• • (x,y,z)T (φ1,θ1)T (φ2,θ2)T (x,y,z) φ2,θ2 (x,y,z) φ1,θ1
NeRF • • •
NeRF • NeRF in the Wild • NeRF • iNeRF
• iNeRF NeRF in the Wild BP
条件 iNeRF NeRF-- IDR GNeRF(提案⼿法) 事前学習 必要 不要 不要 不要
粗いカメラ座標 必要 必要 必要 不要 前⽅のみの制約 不要 必要 不要 不要 マスク 不要 不要 必要 不要 • •
• • • • • GAN Radiance Field • RF
• Phase A Phase B
RF • D E 3
GAN Real • • Discriminator Real • Dynamic Patch Sampling
/ 16x16 • Dynamic Patch Sampling
GAN Fake • • NeRF 16x16 • G
• 6 𝑹 = 𝑟!! 𝑟!" 𝑟!# 𝑟"! 𝑟"" 𝑟"#
𝑟#! 𝑟#" 𝑟## 𝒓 = 𝑟!! 𝑟!" 𝑟!# 𝑟"! 𝑟"" 𝑟"# $ • SO(3) NN • On the Continuity of Rotation Representations in Neural Networks
• Inversion network • Static Patch Sampling • E •
Static Patch Sampling
G G D G
• • Inversion network • refinement
• Refinement
• • • • • Photometric loss
• • Generator • NeRF • Discriminator • Instance Normalization
Spectral Normalization • Inversion network • Vision Transformer • • A • A B 1 iteration • B
• • COLMAP NeRF C+n • Refine C+r • Ground
Truth NeRF G+n • • Synthetic-NeRF image size: 400x400 DTU 500x400 • 100 • • Peak Signal-to-Noise Ratio SSIM LPIPS • / • Single TAITAN GPU 30
COLMAP •
• COLMAP Drums
難しいシーン 特徴の少ない シーン
IDR • 3D IDR • IDR • PNSR •
• COLMAP • Drums Lego COLMAP
Ablation Study • • GAN InvNet. PhaseB • • A
B A ABABABAB B
• • • COLMAP • •
• NeRF • • GAN RF • RF • •
• COLMAP