SlideShare une entreprise Scribd logo
1  sur  25
Hyungjin Chung
Parallel Diffusion Model of Operator and Image
for Blind Inverse Problems
CVPR 2023
Jeongsol Kim Jong Chul Ye
Sehui Kim
1 / 25
http://yang-song.github.io/blog/2021/score/
Background: Diffusion models and Stein score
Explicit score matching
Denoising score matching
Equivalent (Vincent et al. 2010)
𝜽∗
= argmin
𝜽
𝔼[‖∇𝐱log 𝑝 𝐱 − 𝐬𝜽(𝐱)‖2
2
]
𝜽∗ = argmin
𝜽
𝔼[‖∇𝐱log 𝑝 𝐱 | 𝐱 − 𝐬𝜽(𝐱)‖2
2
]
2 / 25
Background: Diffusion models and Stein score 3 / 25
Background: Inverse problems
Imaging system
𝒜
𝐱
Ground truth image
𝜼
noise
𝐲
Measurement
• Problem: recover 𝐱 from noisy measurement 𝐲
• Ill-posed: Infinitely many solutions may exist
• We need to know the prior of the data distribution: how should the image look like?
4 / 25
Background: Examples of inverse problems
𝐲
𝐱
• Inpainting
⨀
• Deblurring (Deconvolution)
𝐱 𝐲
∗
• CS-MRI
𝓕
⨀
5 / 25
Posterior sampling for inverse problems
𝐱 ∼ 𝑝(𝐱|𝐲)
𝐲 = 𝒜(𝐱) + 𝜼
6 / 25
∇𝐱𝑡
log 𝑝(𝐱𝑡|𝐲) = ∇𝐱𝑡
log 𝑝(𝐱𝑡) + ∇𝐱𝑡
log 𝑝 𝐲 𝐱𝑡
Diffusion Posterior Sampling (DPS)
≃ 𝑠𝜃∗ 𝐱𝑡, 𝑡 + ∇𝐱𝑡
log 𝑝 𝐲 𝐱0
Chung et al. ICLR 2023
7 / 25
What if we do not know what the forward model is?
𝐲 = 𝒜(𝐱) + 𝜼
unknown
8 / 25
What if we do not know what the forward model is?
𝐲 = 𝒜𝜙(𝐱) + 𝜼
unknown
9 / 25
What if we do not know what the forward model is?
𝐲 = 𝒜𝜙(𝐱) + 𝜼
unknown
𝒜𝜙
Blind deconvolution (deblurring)
10 / 25
Diffusion model for blind deconvolution
𝑝 𝒚 𝒙0, 𝒌𝟎 = 𝒩(𝒚|𝒌𝟎 ∗ 𝒙𝟎, 𝜎2
𝑰)
𝑝 𝒙0, 𝒌𝟎 𝒚 ∝ 𝑝 𝒚 𝒙𝟎, 𝒌𝟎 𝑝 𝒙𝟎 𝑝(𝒌𝟎)
11 / 25
Diffusion model for blind deconvolution
𝑝 𝒚 𝒙0, 𝒌𝟎 = 𝒩(𝒚|𝒌𝟎 ∗ 𝒙𝟎, 𝜎2
𝑰)
𝑝 𝒙0, 𝒌𝟎 𝒚 ∝ 𝑝 𝒚 𝒙𝟎, 𝒌𝟎 𝑝 𝒙𝟎 𝑝(𝒌𝟎)
Image prior
12 / 25
Diffusion model for blind deconvolution
𝑝 𝒚 𝒙0, 𝒌𝟎 = 𝒩(𝒚|𝒌𝟎 ∗ 𝒙𝟎, 𝜎2
𝑰)
𝑝 𝒚 𝒙0, 𝒌𝟎 ∝ 𝑝 𝒚 𝒙𝟎, 𝒌𝟎 𝑝 𝒙𝟎 𝑝(𝒌𝟎)
kernel prior
13 / 25
Diffusion model for blind deconvolution
𝑝 𝒚 𝒙0, 𝒌𝟎 = 𝒩(𝒚|𝒌𝟎 ∗ 𝒙𝟎, 𝜎2
𝑰)
𝑝 𝒚 𝒙0, 𝒌𝟎 ∝ 𝑝 𝒚 𝒙𝟎, 𝒌𝟎 𝑝 𝒙𝟎 𝑝(𝒌𝟎)
Choices? Diffusion prior!
14 / 25
Diffusion model for blind deconvolution
𝑝 𝒚 𝒙0, 𝒌𝟎 = 𝒩(𝒚|𝒌𝟎 ∗ 𝒙𝟎, 𝜎2
𝑰)
𝑝 𝒚 𝒙0, 𝒌𝟎 ∝ 𝑝 𝒚 𝒙𝟎, 𝒌𝟎 𝑝 𝒙𝟎 𝑝(𝒌𝟎)
∇𝒙 log 𝑝(𝒙) ≃ 𝑠𝜃
𝐼
(𝒙)
∇𝒌 log 𝑝(𝒌) ≃ 𝑠𝜃
𝑘
(𝒌)
Trained independently with
DSM
15 / 25
Diffusion model for blind deconvolution
𝑝 𝒚 𝒙0, 𝒌𝟎 ∝ 𝑝 𝒚 𝒙𝟎, 𝒌𝟎 𝑝 𝒙𝟎 𝑝(𝒌𝟎)
𝑑𝒙 = −
𝛽 𝑡
2
𝒙 − 𝛽 𝑡 ∇𝒙𝑡
log 𝑝 𝒙𝑡 𝑑𝑡 + 𝛽(𝑡)𝑑 𝒘
𝑑𝒌 = −
𝛽 𝑡
2
𝒌 − 𝛽 𝑡 ∇𝒌𝑡
log 𝑝 𝒌𝑡 𝑑𝑡 + 𝛽(𝑡)𝑑 𝒘
16 / 25
Diffusion model for blind deconvolution
𝑝 𝒚 𝒙0, 𝒌𝟎 ∝ 𝑝 𝒚 𝒙𝟎, 𝒌𝟎 𝑝 𝒙𝟎 𝑝(𝒌𝟎)
𝑑𝒙 = −
𝛽 𝑡
2
𝒙 − 𝛽 𝑡 [∇𝒙𝑡
log 𝑝 𝒙𝑡 + ∇𝒙𝑡
log 𝑝 𝒚 𝒙𝑡, 𝒌𝑡 ] 𝑑𝑡 + 𝛽(𝑡)𝑑 𝒘
𝑑𝒌 = −
𝛽 𝑡
2
𝒌 − 𝛽 𝑡 [∇𝒌𝑡
log 𝑝 𝒌𝑡 + ∇𝒌𝑡
log 𝑝 𝒚 𝒙𝑡, 𝒌𝑡 ] 𝑑𝑡 + 𝛽(𝑡)𝑑 𝒘
17 / 25
Diffusion model for blind deconvolution
𝑝 𝒚 𝒙0, 𝒌𝟎 ∝ 𝑝 𝒚 𝒙𝟎, 𝒌𝟎 𝑝 𝒙𝟎 𝑝(𝒌𝟎)
𝑑𝒙 = −
𝛽 𝑡
2
𝒙 − 𝛽 𝑡 [∇𝒙𝑡
log 𝑝 𝒙𝑡 + ∇𝒙𝑡
log 𝑝 𝒚 𝒙0, 𝒌0 ] 𝑑𝑡 + 𝛽(𝑡)𝑑 𝒘
𝑑𝒌 = −
𝛽 𝑡
2
𝒌 − 𝛽 𝑡 [∇𝒌𝑡
log 𝑝 𝒌𝑡 + ∇𝒌𝑡
log 𝑝 𝒚 𝒙0, 𝒌0 ] 𝑑𝑡 + 𝛽(𝑡)𝑑 𝒘
Theorem. Under similar conditions as in DPS,
∇𝒙𝑡
log 𝑝 𝒚 𝒙𝑡, 𝒌𝑡 ≃ ∇𝐱𝑡
log 𝑝 𝒚 𝒙𝟎 𝒙𝑡 , 𝒌𝟎(𝒌𝑡)
∇𝒌𝑡
log 𝑝 𝒚 𝒙𝑡, 𝒌𝑡 ≃ ∇𝒌𝑡
log 𝑝 𝒚 𝒙𝟎 𝒙𝑡 , 𝒌𝟎(𝒌𝑡)
18 / 25
Single diffusion model for non-blind inverse problem 19 / 25
Parallel diffusion models for blind inverse problem 20 / 25
Parallel diffusion models for blind inverse problem 21 / 25
Applicability to diverse blind inverse problems 22 / 25
Results: Blind Motion deblurring
Measurement Ours Ground truth
23 / 25
Results: Imaging through turbulence
Measurement Ours Ground truth
24 / 25
Thank you!
25 / 25

Contenu connexe

Tendances

Isolation forest
Isolation forestIsolation forest
Isolation forestkataware
 
近年のHierarchical Vision Transformer
近年のHierarchical Vision Transformer近年のHierarchical Vision Transformer
近年のHierarchical Vision TransformerYusuke Uchida
 
Batch normalization effectiveness_20190206
Batch normalization effectiveness_20190206Batch normalization effectiveness_20190206
Batch normalization effectiveness_20190206Masakazu Shinoda
 
[DL輪読会]Swin Transformer: Hierarchical Vision Transformer using Shifted Windows
[DL輪読会]Swin Transformer: Hierarchical Vision Transformer using Shifted Windows[DL輪読会]Swin Transformer: Hierarchical Vision Transformer using Shifted Windows
[DL輪読会]Swin Transformer: Hierarchical Vision Transformer using Shifted WindowsDeep Learning JP
 
Transformerを多層にする際の勾配消失問題と解決法について
Transformerを多層にする際の勾配消失問題と解決法についてTransformerを多層にする際の勾配消失問題と解決法について
Transformerを多層にする際の勾配消失問題と解決法についてSho Takase
 
Swin Transformer (ICCV'21 Best Paper) を完璧に理解する資料
Swin Transformer (ICCV'21 Best Paper) を完璧に理解する資料Swin Transformer (ICCV'21 Best Paper) を完璧に理解する資料
Swin Transformer (ICCV'21 Best Paper) を完璧に理解する資料Yusuke Uchida
 
[DeepLearning論文読み会] Dataset Distillation
[DeepLearning論文読み会] Dataset Distillation[DeepLearning論文読み会] Dataset Distillation
[DeepLearning論文読み会] Dataset DistillationRyutaro Yamauchi
 
[サーベイ論文] Deep Learningを用いた歩行者検出の研究動向
[サーベイ論文] Deep Learningを用いた歩行者検出の研究動向[サーベイ論文] Deep Learningを用いた歩行者検出の研究動向
[サーベイ論文] Deep Learningを用いた歩行者検出の研究動向Hiroshi Fukui
 
FineDiving: A Fine-grained Dataset for Procedure-aware Action Quality Assessm...
FineDiving: A Fine-grained Dataset for Procedure-aware Action Quality Assessm...FineDiving: A Fine-grained Dataset for Procedure-aware Action Quality Assessm...
FineDiving: A Fine-grained Dataset for Procedure-aware Action Quality Assessm...Shoki Miyagawa
 
backbone としての timm 入門
backbone としての timm 入門backbone としての timm 入門
backbone としての timm 入門Takuji Tahara
 
【DL輪読会】Parameter is Not All You Need:Starting from Non-Parametric Networks fo...
【DL輪読会】Parameter is Not All You Need:Starting from Non-Parametric Networks fo...【DL輪読会】Parameter is Not All You Need:Starting from Non-Parametric Networks fo...
【DL輪読会】Parameter is Not All You Need:Starting from Non-Parametric Networks fo...Deep Learning JP
 
変分推論と Normalizing Flow
変分推論と Normalizing Flow変分推論と Normalizing Flow
変分推論と Normalizing FlowAkihiro Nitta
 
研究室内PRML勉強会 8章1節
研究室内PRML勉強会 8章1節研究室内PRML勉強会 8章1節
研究室内PRML勉強会 8章1節Koji Matsuda
 
Active Convolution, Deformable Convolution ―形状・スケールを学習可能なConvolution―
Active Convolution, Deformable Convolution ―形状・スケールを学習可能なConvolution―Active Convolution, Deformable Convolution ―形状・スケールを学習可能なConvolution―
Active Convolution, Deformable Convolution ―形状・スケールを学習可能なConvolution―Yosuke Shinya
 
【メタサーベイ】Vision and Language のトップ研究室/研究者
【メタサーベイ】Vision and Language のトップ研究室/研究者【メタサーベイ】Vision and Language のトップ研究室/研究者
【メタサーベイ】Vision and Language のトップ研究室/研究者cvpaper. challenge
 
[DL輪読会]Deep Learning 第15章 表現学習
[DL輪読会]Deep Learning 第15章 表現学習[DL輪読会]Deep Learning 第15章 表現学習
[DL輪読会]Deep Learning 第15章 表現学習Deep Learning JP
 
PRML輪読#14
PRML輪読#14PRML輪読#14
PRML輪読#14matsuolab
 
[DL輪読会]3D Human Pose Estimation @ CVPR’19 / ICCV’19
[DL輪読会]3D Human Pose Estimation @ CVPR’19 / ICCV’19[DL輪読会]3D Human Pose Estimation @ CVPR’19 / ICCV’19
[DL輪読会]3D Human Pose Estimation @ CVPR’19 / ICCV’19Deep Learning JP
 
[DL輪読会]Set Transformer: A Framework for Attention-based Permutation-Invariant...
[DL輪読会]Set Transformer: A Framework for Attention-based Permutation-Invariant...[DL輪読会]Set Transformer: A Framework for Attention-based Permutation-Invariant...
[DL輪読会]Set Transformer: A Framework for Attention-based Permutation-Invariant...Deep Learning JP
 

Tendances (20)

Isolation forest
Isolation forestIsolation forest
Isolation forest
 
近年のHierarchical Vision Transformer
近年のHierarchical Vision Transformer近年のHierarchical Vision Transformer
近年のHierarchical Vision Transformer
 
Batch normalization effectiveness_20190206
Batch normalization effectiveness_20190206Batch normalization effectiveness_20190206
Batch normalization effectiveness_20190206
 
[DL輪読会]Swin Transformer: Hierarchical Vision Transformer using Shifted Windows
[DL輪読会]Swin Transformer: Hierarchical Vision Transformer using Shifted Windows[DL輪読会]Swin Transformer: Hierarchical Vision Transformer using Shifted Windows
[DL輪読会]Swin Transformer: Hierarchical Vision Transformer using Shifted Windows
 
Transformerを多層にする際の勾配消失問題と解決法について
Transformerを多層にする際の勾配消失問題と解決法についてTransformerを多層にする際の勾配消失問題と解決法について
Transformerを多層にする際の勾配消失問題と解決法について
 
Swin Transformer (ICCV'21 Best Paper) を完璧に理解する資料
Swin Transformer (ICCV'21 Best Paper) を完璧に理解する資料Swin Transformer (ICCV'21 Best Paper) を完璧に理解する資料
Swin Transformer (ICCV'21 Best Paper) を完璧に理解する資料
 
[DeepLearning論文読み会] Dataset Distillation
[DeepLearning論文読み会] Dataset Distillation[DeepLearning論文読み会] Dataset Distillation
[DeepLearning論文読み会] Dataset Distillation
 
[サーベイ論文] Deep Learningを用いた歩行者検出の研究動向
[サーベイ論文] Deep Learningを用いた歩行者検出の研究動向[サーベイ論文] Deep Learningを用いた歩行者検出の研究動向
[サーベイ論文] Deep Learningを用いた歩行者検出の研究動向
 
FineDiving: A Fine-grained Dataset for Procedure-aware Action Quality Assessm...
FineDiving: A Fine-grained Dataset for Procedure-aware Action Quality Assessm...FineDiving: A Fine-grained Dataset for Procedure-aware Action Quality Assessm...
FineDiving: A Fine-grained Dataset for Procedure-aware Action Quality Assessm...
 
backbone としての timm 入門
backbone としての timm 入門backbone としての timm 入門
backbone としての timm 入門
 
【DL輪読会】Parameter is Not All You Need:Starting from Non-Parametric Networks fo...
【DL輪読会】Parameter is Not All You Need:Starting from Non-Parametric Networks fo...【DL輪読会】Parameter is Not All You Need:Starting from Non-Parametric Networks fo...
【DL輪読会】Parameter is Not All You Need:Starting from Non-Parametric Networks fo...
 
変分推論と Normalizing Flow
変分推論と Normalizing Flow変分推論と Normalizing Flow
変分推論と Normalizing Flow
 
実装レベルで学ぶVQVAE
実装レベルで学ぶVQVAE実装レベルで学ぶVQVAE
実装レベルで学ぶVQVAE
 
研究室内PRML勉強会 8章1節
研究室内PRML勉強会 8章1節研究室内PRML勉強会 8章1節
研究室内PRML勉強会 8章1節
 
Active Convolution, Deformable Convolution ―形状・スケールを学習可能なConvolution―
Active Convolution, Deformable Convolution ―形状・スケールを学習可能なConvolution―Active Convolution, Deformable Convolution ―形状・スケールを学習可能なConvolution―
Active Convolution, Deformable Convolution ―形状・スケールを学習可能なConvolution―
 
【メタサーベイ】Vision and Language のトップ研究室/研究者
【メタサーベイ】Vision and Language のトップ研究室/研究者【メタサーベイ】Vision and Language のトップ研究室/研究者
【メタサーベイ】Vision and Language のトップ研究室/研究者
 
[DL輪読会]Deep Learning 第15章 表現学習
[DL輪読会]Deep Learning 第15章 表現学習[DL輪読会]Deep Learning 第15章 表現学習
[DL輪読会]Deep Learning 第15章 表現学習
 
PRML輪読#14
PRML輪読#14PRML輪読#14
PRML輪読#14
 
[DL輪読会]3D Human Pose Estimation @ CVPR’19 / ICCV’19
[DL輪読会]3D Human Pose Estimation @ CVPR’19 / ICCV’19[DL輪読会]3D Human Pose Estimation @ CVPR’19 / ICCV’19
[DL輪読会]3D Human Pose Estimation @ CVPR’19 / ICCV’19
 
[DL輪読会]Set Transformer: A Framework for Attention-based Permutation-Invariant...
[DL輪読会]Set Transformer: A Framework for Attention-based Permutation-Invariant...[DL輪読会]Set Transformer: A Framework for Attention-based Permutation-Invariant...
[DL輪読会]Set Transformer: A Framework for Attention-based Permutation-Invariant...
 

Similaire à BlindDPS_presentation_silde.pptx

diffusion_posterior_sampling_for_general_noisy_inverse_problems_slideshare.pdf
diffusion_posterior_sampling_for_general_noisy_inverse_problems_slideshare.pdfdiffusion_posterior_sampling_for_general_noisy_inverse_problems_slideshare.pdf
diffusion_posterior_sampling_for_general_noisy_inverse_problems_slideshare.pdfChung Hyung Jin
 
TPDM Presentation Slide (ICCV23)
TPDM Presentation Slide (ICCV23)TPDM Presentation Slide (ICCV23)
TPDM Presentation Slide (ICCV23)Suhyeon Lee
 
Sampling method : MCMC
Sampling method : MCMCSampling method : MCMC
Sampling method : MCMCSEMINARGROOT
 
DiffusionMBIR_presentation_slide.pptx
DiffusionMBIR_presentation_slide.pptxDiffusionMBIR_presentation_slide.pptx
DiffusionMBIR_presentation_slide.pptxChung Hyung Jin
 
Direct solution of sparse network equations by optimally ordered triangular f...
Direct solution of sparse network equations by optimally ordered triangular f...Direct solution of sparse network equations by optimally ordered triangular f...
Direct solution of sparse network equations by optimally ordered triangular f...Dimas Ruliandi
 
ALEXANDER FRACTIONAL INTEGRAL FILTERING OF WAVELET COEFFICIENTS FOR IMAGE DEN...
ALEXANDER FRACTIONAL INTEGRAL FILTERING OF WAVELET COEFFICIENTS FOR IMAGE DEN...ALEXANDER FRACTIONAL INTEGRAL FILTERING OF WAVELET COEFFICIENTS FOR IMAGE DEN...
ALEXANDER FRACTIONAL INTEGRAL FILTERING OF WAVELET COEFFICIENTS FOR IMAGE DEN...sipij
 
6. 2 ray propagation model part 2
6. 2 ray propagation model   part 26. 2 ray propagation model   part 2
6. 2 ray propagation model part 2JAIGANESH SEKAR
 
Folding Cheat Sheet #2 - second in a series
Folding Cheat Sheet #2 - second in a seriesFolding Cheat Sheet #2 - second in a series
Folding Cheat Sheet #2 - second in a seriesPhilip Schwarz
 
Folding Cheat Sheet #2 - second in a series
Folding Cheat Sheet #2 - second in a seriesFolding Cheat Sheet #2 - second in a series
Folding Cheat Sheet #2 - second in a seriesPhilip Schwarz
 
New classes of Adomian polynomials for the Adomian decomposition method
New classes of Adomian polynomials for the Adomian decomposition method New classes of Adomian polynomials for the Adomian decomposition method
New classes of Adomian polynomials for the Adomian decomposition method inventionjournals
 
20180831 riemannian representation learning
20180831 riemannian representation learning20180831 riemannian representation learning
20180831 riemannian representation learningsegwangkim
 
BU_FCAI_SCC430_Modeling&Simulation_Ch05-P2.pptx
BU_FCAI_SCC430_Modeling&Simulation_Ch05-P2.pptxBU_FCAI_SCC430_Modeling&Simulation_Ch05-P2.pptx
BU_FCAI_SCC430_Modeling&Simulation_Ch05-P2.pptxMaiGaafar
 
Visual Impression Localization of Autonomous Robots_#CASE2015
Visual Impression Localization of Autonomous Robots_#CASE2015Visual Impression Localization of Autonomous Robots_#CASE2015
Visual Impression Localization of Autonomous Robots_#CASE2015Soma Boubou
 
Analysis of large scale spiking networks dynamics with spatio-temporal constr...
Analysis of large scale spiking networks dynamics with spatio-temporal constr...Analysis of large scale spiking networks dynamics with spatio-temporal constr...
Analysis of large scale spiking networks dynamics with spatio-temporal constr...Hassan Nasser
 
Blow up in a degenerate keller--segel system(Eng.)
Blow up in a degenerate keller--segel system(Eng.)Blow up in a degenerate keller--segel system(Eng.)
Blow up in a degenerate keller--segel system(Eng.)Takahiro Hashira
 
Variational autoencoders for speech processing d.bielievtsov dataconf 21 04 18
Variational autoencoders for speech processing d.bielievtsov dataconf 21 04 18Variational autoencoders for speech processing d.bielievtsov dataconf 21 04 18
Variational autoencoders for speech processing d.bielievtsov dataconf 21 04 18Olga Zinkevych
 
Basic Concepts of Standard Experimental Designs ( Statistics )
Basic Concepts of Standard Experimental Designs ( Statistics )Basic Concepts of Standard Experimental Designs ( Statistics )
Basic Concepts of Standard Experimental Designs ( Statistics )Hasnat Israq
 

Similaire à BlindDPS_presentation_silde.pptx (20)

diffusion_posterior_sampling_for_general_noisy_inverse_problems_slideshare.pdf
diffusion_posterior_sampling_for_general_noisy_inverse_problems_slideshare.pdfdiffusion_posterior_sampling_for_general_noisy_inverse_problems_slideshare.pdf
diffusion_posterior_sampling_for_general_noisy_inverse_problems_slideshare.pdf
 
TPDM Presentation Slide (ICCV23)
TPDM Presentation Slide (ICCV23)TPDM Presentation Slide (ICCV23)
TPDM Presentation Slide (ICCV23)
 
Sampling method : MCMC
Sampling method : MCMCSampling method : MCMC
Sampling method : MCMC
 
DiffusionMBIR_presentation_slide.pptx
DiffusionMBIR_presentation_slide.pptxDiffusionMBIR_presentation_slide.pptx
DiffusionMBIR_presentation_slide.pptx
 
Direct solution of sparse network equations by optimally ordered triangular f...
Direct solution of sparse network equations by optimally ordered triangular f...Direct solution of sparse network equations by optimally ordered triangular f...
Direct solution of sparse network equations by optimally ordered triangular f...
 
ALEXANDER FRACTIONAL INTEGRAL FILTERING OF WAVELET COEFFICIENTS FOR IMAGE DEN...
ALEXANDER FRACTIONAL INTEGRAL FILTERING OF WAVELET COEFFICIENTS FOR IMAGE DEN...ALEXANDER FRACTIONAL INTEGRAL FILTERING OF WAVELET COEFFICIENTS FOR IMAGE DEN...
ALEXANDER FRACTIONAL INTEGRAL FILTERING OF WAVELET COEFFICIENTS FOR IMAGE DEN...
 
6. 2 ray propagation model part 2
6. 2 ray propagation model   part 26. 2 ray propagation model   part 2
6. 2 ray propagation model part 2
 
AJMS_477_23.pdf
AJMS_477_23.pdfAJMS_477_23.pdf
AJMS_477_23.pdf
 
Folding Cheat Sheet #2 - second in a series
Folding Cheat Sheet #2 - second in a seriesFolding Cheat Sheet #2 - second in a series
Folding Cheat Sheet #2 - second in a series
 
Folding Cheat Sheet #2 - second in a series
Folding Cheat Sheet #2 - second in a seriesFolding Cheat Sheet #2 - second in a series
Folding Cheat Sheet #2 - second in a series
 
New classes of Adomian polynomials for the Adomian decomposition method
New classes of Adomian polynomials for the Adomian decomposition method New classes of Adomian polynomials for the Adomian decomposition method
New classes of Adomian polynomials for the Adomian decomposition method
 
Image denoising
Image denoisingImage denoising
Image denoising
 
20180831 riemannian representation learning
20180831 riemannian representation learning20180831 riemannian representation learning
20180831 riemannian representation learning
 
AJMS_480_23.pdf
AJMS_480_23.pdfAJMS_480_23.pdf
AJMS_480_23.pdf
 
BU_FCAI_SCC430_Modeling&Simulation_Ch05-P2.pptx
BU_FCAI_SCC430_Modeling&Simulation_Ch05-P2.pptxBU_FCAI_SCC430_Modeling&Simulation_Ch05-P2.pptx
BU_FCAI_SCC430_Modeling&Simulation_Ch05-P2.pptx
 
Visual Impression Localization of Autonomous Robots_#CASE2015
Visual Impression Localization of Autonomous Robots_#CASE2015Visual Impression Localization of Autonomous Robots_#CASE2015
Visual Impression Localization of Autonomous Robots_#CASE2015
 
Analysis of large scale spiking networks dynamics with spatio-temporal constr...
Analysis of large scale spiking networks dynamics with spatio-temporal constr...Analysis of large scale spiking networks dynamics with spatio-temporal constr...
Analysis of large scale spiking networks dynamics with spatio-temporal constr...
 
Blow up in a degenerate keller--segel system(Eng.)
Blow up in a degenerate keller--segel system(Eng.)Blow up in a degenerate keller--segel system(Eng.)
Blow up in a degenerate keller--segel system(Eng.)
 
Variational autoencoders for speech processing d.bielievtsov dataconf 21 04 18
Variational autoencoders for speech processing d.bielievtsov dataconf 21 04 18Variational autoencoders for speech processing d.bielievtsov dataconf 21 04 18
Variational autoencoders for speech processing d.bielievtsov dataconf 21 04 18
 
Basic Concepts of Standard Experimental Designs ( Statistics )
Basic Concepts of Standard Experimental Designs ( Statistics )Basic Concepts of Standard Experimental Designs ( Statistics )
Basic Concepts of Standard Experimental Designs ( Statistics )
 

Plus de Chung Hyung Jin

mr-denoising-and-super-resolution-using-regularized-reverse-diffusion.pptx
mr-denoising-and-super-resolution-using-regularized-reverse-diffusion.pptxmr-denoising-and-super-resolution-using-regularized-reverse-diffusion.pptx
mr-denoising-and-super-resolution-using-regularized-reverse-diffusion.pptxChung Hyung Jin
 
Low-dose sparse-view HAADF-STEM-EDX tomography of nanocrystals using unsuperv...
Low-dose sparse-view HAADF-STEM-EDX tomography of nanocrystals using unsuperv...Low-dose sparse-view HAADF-STEM-EDX tomography of nanocrystals using unsuperv...
Low-dose sparse-view HAADF-STEM-EDX tomography of nanocrystals using unsuperv...Chung Hyung Jin
 
Score-based diffusion models for accelerated MRI.pptx
Score-based diffusion models for accelerated MRI.pptxScore-based diffusion models for accelerated MRI.pptx
Score-based diffusion models for accelerated MRI.pptxChung Hyung Jin
 
Come-Closer-Diffuse-Faster Accelerating Conditional Diffusion Models for Inve...
Come-Closer-Diffuse-Faster Accelerating Conditional Diffusion Models for Inve...Come-Closer-Diffuse-Faster Accelerating Conditional Diffusion Models for Inve...
Come-Closer-Diffuse-Faster Accelerating Conditional Diffusion Models for Inve...Chung Hyung Jin
 
Two stage deep learning for accelerated 3 d time-of-flight mra without matche...
Two stage deep learning for accelerated 3 d time-of-flight mra without matche...Two stage deep learning for accelerated 3 d time-of-flight mra without matche...
Two stage deep learning for accelerated 3 d time-of-flight mra without matche...Chung Hyung Jin
 
Missing cone artifact removal in odt using unsupervised deep learning in the ...
Missing cone artifact removal in odt using unsupervised deep learning in the ...Missing cone artifact removal in odt using unsupervised deep learning in the ...
Missing cone artifact removal in odt using unsupervised deep learning in the ...Chung Hyung Jin
 
A deep learning model for diagnosing gastric mucosal lesions using endoscopic...
A deep learning model for diagnosing gastric mucosal lesions using endoscopic...A deep learning model for diagnosing gastric mucosal lesions using endoscopic...
A deep learning model for diagnosing gastric mucosal lesions using endoscopic...Chung Hyung Jin
 
Feature disentanglement in generating a three dimensional structure from a tw...
Feature disentanglement in generating a three dimensional structure from a tw...Feature disentanglement in generating a three dimensional structure from a tw...
Feature disentanglement in generating a three dimensional structure from a tw...Chung Hyung Jin
 
Unsupervised deep learning methods for biological image reconstruction and en...
Unsupervised deep learning methods for biological image reconstruction and en...Unsupervised deep learning methods for biological image reconstruction and en...
Unsupervised deep learning methods for biological image reconstruction and en...Chung Hyung Jin
 

Plus de Chung Hyung Jin (9)

mr-denoising-and-super-resolution-using-regularized-reverse-diffusion.pptx
mr-denoising-and-super-resolution-using-regularized-reverse-diffusion.pptxmr-denoising-and-super-resolution-using-regularized-reverse-diffusion.pptx
mr-denoising-and-super-resolution-using-regularized-reverse-diffusion.pptx
 
Low-dose sparse-view HAADF-STEM-EDX tomography of nanocrystals using unsuperv...
Low-dose sparse-view HAADF-STEM-EDX tomography of nanocrystals using unsuperv...Low-dose sparse-view HAADF-STEM-EDX tomography of nanocrystals using unsuperv...
Low-dose sparse-view HAADF-STEM-EDX tomography of nanocrystals using unsuperv...
 
Score-based diffusion models for accelerated MRI.pptx
Score-based diffusion models for accelerated MRI.pptxScore-based diffusion models for accelerated MRI.pptx
Score-based diffusion models for accelerated MRI.pptx
 
Come-Closer-Diffuse-Faster Accelerating Conditional Diffusion Models for Inve...
Come-Closer-Diffuse-Faster Accelerating Conditional Diffusion Models for Inve...Come-Closer-Diffuse-Faster Accelerating Conditional Diffusion Models for Inve...
Come-Closer-Diffuse-Faster Accelerating Conditional Diffusion Models for Inve...
 
Two stage deep learning for accelerated 3 d time-of-flight mra without matche...
Two stage deep learning for accelerated 3 d time-of-flight mra without matche...Two stage deep learning for accelerated 3 d time-of-flight mra without matche...
Two stage deep learning for accelerated 3 d time-of-flight mra without matche...
 
Missing cone artifact removal in odt using unsupervised deep learning in the ...
Missing cone artifact removal in odt using unsupervised deep learning in the ...Missing cone artifact removal in odt using unsupervised deep learning in the ...
Missing cone artifact removal in odt using unsupervised deep learning in the ...
 
A deep learning model for diagnosing gastric mucosal lesions using endoscopic...
A deep learning model for diagnosing gastric mucosal lesions using endoscopic...A deep learning model for diagnosing gastric mucosal lesions using endoscopic...
A deep learning model for diagnosing gastric mucosal lesions using endoscopic...
 
Feature disentanglement in generating a three dimensional structure from a tw...
Feature disentanglement in generating a three dimensional structure from a tw...Feature disentanglement in generating a three dimensional structure from a tw...
Feature disentanglement in generating a three dimensional structure from a tw...
 
Unsupervised deep learning methods for biological image reconstruction and en...
Unsupervised deep learning methods for biological image reconstruction and en...Unsupervised deep learning methods for biological image reconstruction and en...
Unsupervised deep learning methods for biological image reconstruction and en...
 

Dernier

College Call Girls Nashik Nehal 7001305949 Independent Escort Service Nashik
College Call Girls Nashik Nehal 7001305949 Independent Escort Service NashikCollege Call Girls Nashik Nehal 7001305949 Independent Escort Service Nashik
College Call Girls Nashik Nehal 7001305949 Independent Escort Service NashikCall Girls in Nagpur High Profile
 
UNIT-V FMM.HYDRAULIC TURBINE - Construction and working
UNIT-V FMM.HYDRAULIC TURBINE - Construction and workingUNIT-V FMM.HYDRAULIC TURBINE - Construction and working
UNIT-V FMM.HYDRAULIC TURBINE - Construction and workingrknatarajan
 
Online banking management system project.pdf
Online banking management system project.pdfOnline banking management system project.pdf
Online banking management system project.pdfKamal Acharya
 
Java Programming :Event Handling(Types of Events)
Java Programming :Event Handling(Types of Events)Java Programming :Event Handling(Types of Events)
Java Programming :Event Handling(Types of Events)simmis5
 
Russian Call Girls in Nagpur Grishma Call 7001035870 Meet With Nagpur Escorts
Russian Call Girls in Nagpur Grishma Call 7001035870 Meet With Nagpur EscortsRussian Call Girls in Nagpur Grishma Call 7001035870 Meet With Nagpur Escorts
Russian Call Girls in Nagpur Grishma Call 7001035870 Meet With Nagpur EscortsCall Girls in Nagpur High Profile
 
Porous Ceramics seminar and technical writing
Porous Ceramics seminar and technical writingPorous Ceramics seminar and technical writing
Porous Ceramics seminar and technical writingrakeshbaidya232001
 
VIP Call Girls Ankleshwar 7001035870 Whatsapp Number, 24/07 Booking
VIP Call Girls Ankleshwar 7001035870 Whatsapp Number, 24/07 BookingVIP Call Girls Ankleshwar 7001035870 Whatsapp Number, 24/07 Booking
VIP Call Girls Ankleshwar 7001035870 Whatsapp Number, 24/07 Bookingdharasingh5698
 
Call Girls Service Nagpur Tanvi Call 7001035870 Meet With Nagpur Escorts
Call Girls Service Nagpur Tanvi Call 7001035870 Meet With Nagpur EscortsCall Girls Service Nagpur Tanvi Call 7001035870 Meet With Nagpur Escorts
Call Girls Service Nagpur Tanvi Call 7001035870 Meet With Nagpur EscortsCall Girls in Nagpur High Profile
 
(PRIYA) Rajgurunagar Call Girls Just Call 7001035870 [ Cash on Delivery ] Pun...
(PRIYA) Rajgurunagar Call Girls Just Call 7001035870 [ Cash on Delivery ] Pun...(PRIYA) Rajgurunagar Call Girls Just Call 7001035870 [ Cash on Delivery ] Pun...
(PRIYA) Rajgurunagar Call Girls Just Call 7001035870 [ Cash on Delivery ] Pun...ranjana rawat
 
High Profile Call Girls Nagpur Isha Call 7001035870 Meet With Nagpur Escorts
High Profile Call Girls Nagpur Isha Call 7001035870 Meet With Nagpur EscortsHigh Profile Call Girls Nagpur Isha Call 7001035870 Meet With Nagpur Escorts
High Profile Call Girls Nagpur Isha Call 7001035870 Meet With Nagpur Escortsranjana rawat
 
UNIT - IV - Air Compressors and its Performance
UNIT - IV - Air Compressors and its PerformanceUNIT - IV - Air Compressors and its Performance
UNIT - IV - Air Compressors and its Performancesivaprakash250
 
MANUFACTURING PROCESS-II UNIT-5 NC MACHINE TOOLS
MANUFACTURING PROCESS-II UNIT-5 NC MACHINE TOOLSMANUFACTURING PROCESS-II UNIT-5 NC MACHINE TOOLS
MANUFACTURING PROCESS-II UNIT-5 NC MACHINE TOOLSSIVASHANKAR N
 
Structural Analysis and Design of Foundations: A Comprehensive Handbook for S...
Structural Analysis and Design of Foundations: A Comprehensive Handbook for S...Structural Analysis and Design of Foundations: A Comprehensive Handbook for S...
Structural Analysis and Design of Foundations: A Comprehensive Handbook for S...Dr.Costas Sachpazis
 
Booking open Available Pune Call Girls Koregaon Park 6297143586 Call Hot Ind...
Booking open Available Pune Call Girls Koregaon Park  6297143586 Call Hot Ind...Booking open Available Pune Call Girls Koregaon Park  6297143586 Call Hot Ind...
Booking open Available Pune Call Girls Koregaon Park 6297143586 Call Hot Ind...Call Girls in Nagpur High Profile
 
Extrusion Processes and Their Limitations
Extrusion Processes and Their LimitationsExtrusion Processes and Their Limitations
Extrusion Processes and Their Limitations120cr0395
 
University management System project report..pdf
University management System project report..pdfUniversity management System project report..pdf
University management System project report..pdfKamal Acharya
 
Call for Papers - African Journal of Biological Sciences, E-ISSN: 2663-2187, ...
Call for Papers - African Journal of Biological Sciences, E-ISSN: 2663-2187, ...Call for Papers - African Journal of Biological Sciences, E-ISSN: 2663-2187, ...
Call for Papers - African Journal of Biological Sciences, E-ISSN: 2663-2187, ...Christo Ananth
 
Processing & Properties of Floor and Wall Tiles.pptx
Processing & Properties of Floor and Wall Tiles.pptxProcessing & Properties of Floor and Wall Tiles.pptx
Processing & Properties of Floor and Wall Tiles.pptxpranjaldaimarysona
 

Dernier (20)

College Call Girls Nashik Nehal 7001305949 Independent Escort Service Nashik
College Call Girls Nashik Nehal 7001305949 Independent Escort Service NashikCollege Call Girls Nashik Nehal 7001305949 Independent Escort Service Nashik
College Call Girls Nashik Nehal 7001305949 Independent Escort Service Nashik
 
UNIT-V FMM.HYDRAULIC TURBINE - Construction and working
UNIT-V FMM.HYDRAULIC TURBINE - Construction and workingUNIT-V FMM.HYDRAULIC TURBINE - Construction and working
UNIT-V FMM.HYDRAULIC TURBINE - Construction and working
 
Online banking management system project.pdf
Online banking management system project.pdfOnline banking management system project.pdf
Online banking management system project.pdf
 
Java Programming :Event Handling(Types of Events)
Java Programming :Event Handling(Types of Events)Java Programming :Event Handling(Types of Events)
Java Programming :Event Handling(Types of Events)
 
Russian Call Girls in Nagpur Grishma Call 7001035870 Meet With Nagpur Escorts
Russian Call Girls in Nagpur Grishma Call 7001035870 Meet With Nagpur EscortsRussian Call Girls in Nagpur Grishma Call 7001035870 Meet With Nagpur Escorts
Russian Call Girls in Nagpur Grishma Call 7001035870 Meet With Nagpur Escorts
 
Porous Ceramics seminar and technical writing
Porous Ceramics seminar and technical writingPorous Ceramics seminar and technical writing
Porous Ceramics seminar and technical writing
 
VIP Call Girls Ankleshwar 7001035870 Whatsapp Number, 24/07 Booking
VIP Call Girls Ankleshwar 7001035870 Whatsapp Number, 24/07 BookingVIP Call Girls Ankleshwar 7001035870 Whatsapp Number, 24/07 Booking
VIP Call Girls Ankleshwar 7001035870 Whatsapp Number, 24/07 Booking
 
Call Girls Service Nagpur Tanvi Call 7001035870 Meet With Nagpur Escorts
Call Girls Service Nagpur Tanvi Call 7001035870 Meet With Nagpur EscortsCall Girls Service Nagpur Tanvi Call 7001035870 Meet With Nagpur Escorts
Call Girls Service Nagpur Tanvi Call 7001035870 Meet With Nagpur Escorts
 
(PRIYA) Rajgurunagar Call Girls Just Call 7001035870 [ Cash on Delivery ] Pun...
(PRIYA) Rajgurunagar Call Girls Just Call 7001035870 [ Cash on Delivery ] Pun...(PRIYA) Rajgurunagar Call Girls Just Call 7001035870 [ Cash on Delivery ] Pun...
(PRIYA) Rajgurunagar Call Girls Just Call 7001035870 [ Cash on Delivery ] Pun...
 
High Profile Call Girls Nagpur Isha Call 7001035870 Meet With Nagpur Escorts
High Profile Call Girls Nagpur Isha Call 7001035870 Meet With Nagpur EscortsHigh Profile Call Girls Nagpur Isha Call 7001035870 Meet With Nagpur Escorts
High Profile Call Girls Nagpur Isha Call 7001035870 Meet With Nagpur Escorts
 
UNIT - IV - Air Compressors and its Performance
UNIT - IV - Air Compressors and its PerformanceUNIT - IV - Air Compressors and its Performance
UNIT - IV - Air Compressors and its Performance
 
MANUFACTURING PROCESS-II UNIT-5 NC MACHINE TOOLS
MANUFACTURING PROCESS-II UNIT-5 NC MACHINE TOOLSMANUFACTURING PROCESS-II UNIT-5 NC MACHINE TOOLS
MANUFACTURING PROCESS-II UNIT-5 NC MACHINE TOOLS
 
Structural Analysis and Design of Foundations: A Comprehensive Handbook for S...
Structural Analysis and Design of Foundations: A Comprehensive Handbook for S...Structural Analysis and Design of Foundations: A Comprehensive Handbook for S...
Structural Analysis and Design of Foundations: A Comprehensive Handbook for S...
 
Booking open Available Pune Call Girls Koregaon Park 6297143586 Call Hot Ind...
Booking open Available Pune Call Girls Koregaon Park  6297143586 Call Hot Ind...Booking open Available Pune Call Girls Koregaon Park  6297143586 Call Hot Ind...
Booking open Available Pune Call Girls Koregaon Park 6297143586 Call Hot Ind...
 
Roadmap to Membership of RICS - Pathways and Routes
Roadmap to Membership of RICS - Pathways and RoutesRoadmap to Membership of RICS - Pathways and Routes
Roadmap to Membership of RICS - Pathways and Routes
 
Extrusion Processes and Their Limitations
Extrusion Processes and Their LimitationsExtrusion Processes and Their Limitations
Extrusion Processes and Their Limitations
 
University management System project report..pdf
University management System project report..pdfUniversity management System project report..pdf
University management System project report..pdf
 
Call for Papers - African Journal of Biological Sciences, E-ISSN: 2663-2187, ...
Call for Papers - African Journal of Biological Sciences, E-ISSN: 2663-2187, ...Call for Papers - African Journal of Biological Sciences, E-ISSN: 2663-2187, ...
Call for Papers - African Journal of Biological Sciences, E-ISSN: 2663-2187, ...
 
Water Industry Process Automation & Control Monthly - April 2024
Water Industry Process Automation & Control Monthly - April 2024Water Industry Process Automation & Control Monthly - April 2024
Water Industry Process Automation & Control Monthly - April 2024
 
Processing & Properties of Floor and Wall Tiles.pptx
Processing & Properties of Floor and Wall Tiles.pptxProcessing & Properties of Floor and Wall Tiles.pptx
Processing & Properties of Floor and Wall Tiles.pptx
 

BlindDPS_presentation_silde.pptx

  • 1. Hyungjin Chung Parallel Diffusion Model of Operator and Image for Blind Inverse Problems CVPR 2023 Jeongsol Kim Jong Chul Ye Sehui Kim 1 / 25
  • 2. http://yang-song.github.io/blog/2021/score/ Background: Diffusion models and Stein score Explicit score matching Denoising score matching Equivalent (Vincent et al. 2010) 𝜽∗ = argmin 𝜽 𝔼[‖∇𝐱log 𝑝 𝐱 − 𝐬𝜽(𝐱)‖2 2 ] 𝜽∗ = argmin 𝜽 𝔼[‖∇𝐱log 𝑝 𝐱 | 𝐱 − 𝐬𝜽(𝐱)‖2 2 ] 2 / 25
  • 3. Background: Diffusion models and Stein score 3 / 25
  • 4. Background: Inverse problems Imaging system 𝒜 𝐱 Ground truth image 𝜼 noise 𝐲 Measurement • Problem: recover 𝐱 from noisy measurement 𝐲 • Ill-posed: Infinitely many solutions may exist • We need to know the prior of the data distribution: how should the image look like? 4 / 25
  • 5. Background: Examples of inverse problems 𝐲 𝐱 • Inpainting ⨀ • Deblurring (Deconvolution) 𝐱 𝐲 ∗ • CS-MRI 𝓕 ⨀ 5 / 25
  • 6. Posterior sampling for inverse problems 𝐱 ∼ 𝑝(𝐱|𝐲) 𝐲 = 𝒜(𝐱) + 𝜼 6 / 25
  • 7. ∇𝐱𝑡 log 𝑝(𝐱𝑡|𝐲) = ∇𝐱𝑡 log 𝑝(𝐱𝑡) + ∇𝐱𝑡 log 𝑝 𝐲 𝐱𝑡 Diffusion Posterior Sampling (DPS) ≃ 𝑠𝜃∗ 𝐱𝑡, 𝑡 + ∇𝐱𝑡 log 𝑝 𝐲 𝐱0 Chung et al. ICLR 2023 7 / 25
  • 8. What if we do not know what the forward model is? 𝐲 = 𝒜(𝐱) + 𝜼 unknown 8 / 25
  • 9. What if we do not know what the forward model is? 𝐲 = 𝒜𝜙(𝐱) + 𝜼 unknown 9 / 25
  • 10. What if we do not know what the forward model is? 𝐲 = 𝒜𝜙(𝐱) + 𝜼 unknown 𝒜𝜙 Blind deconvolution (deblurring) 10 / 25
  • 11. Diffusion model for blind deconvolution 𝑝 𝒚 𝒙0, 𝒌𝟎 = 𝒩(𝒚|𝒌𝟎 ∗ 𝒙𝟎, 𝜎2 𝑰) 𝑝 𝒙0, 𝒌𝟎 𝒚 ∝ 𝑝 𝒚 𝒙𝟎, 𝒌𝟎 𝑝 𝒙𝟎 𝑝(𝒌𝟎) 11 / 25
  • 12. Diffusion model for blind deconvolution 𝑝 𝒚 𝒙0, 𝒌𝟎 = 𝒩(𝒚|𝒌𝟎 ∗ 𝒙𝟎, 𝜎2 𝑰) 𝑝 𝒙0, 𝒌𝟎 𝒚 ∝ 𝑝 𝒚 𝒙𝟎, 𝒌𝟎 𝑝 𝒙𝟎 𝑝(𝒌𝟎) Image prior 12 / 25
  • 13. Diffusion model for blind deconvolution 𝑝 𝒚 𝒙0, 𝒌𝟎 = 𝒩(𝒚|𝒌𝟎 ∗ 𝒙𝟎, 𝜎2 𝑰) 𝑝 𝒚 𝒙0, 𝒌𝟎 ∝ 𝑝 𝒚 𝒙𝟎, 𝒌𝟎 𝑝 𝒙𝟎 𝑝(𝒌𝟎) kernel prior 13 / 25
  • 14. Diffusion model for blind deconvolution 𝑝 𝒚 𝒙0, 𝒌𝟎 = 𝒩(𝒚|𝒌𝟎 ∗ 𝒙𝟎, 𝜎2 𝑰) 𝑝 𝒚 𝒙0, 𝒌𝟎 ∝ 𝑝 𝒚 𝒙𝟎, 𝒌𝟎 𝑝 𝒙𝟎 𝑝(𝒌𝟎) Choices? Diffusion prior! 14 / 25
  • 15. Diffusion model for blind deconvolution 𝑝 𝒚 𝒙0, 𝒌𝟎 = 𝒩(𝒚|𝒌𝟎 ∗ 𝒙𝟎, 𝜎2 𝑰) 𝑝 𝒚 𝒙0, 𝒌𝟎 ∝ 𝑝 𝒚 𝒙𝟎, 𝒌𝟎 𝑝 𝒙𝟎 𝑝(𝒌𝟎) ∇𝒙 log 𝑝(𝒙) ≃ 𝑠𝜃 𝐼 (𝒙) ∇𝒌 log 𝑝(𝒌) ≃ 𝑠𝜃 𝑘 (𝒌) Trained independently with DSM 15 / 25
  • 16. Diffusion model for blind deconvolution 𝑝 𝒚 𝒙0, 𝒌𝟎 ∝ 𝑝 𝒚 𝒙𝟎, 𝒌𝟎 𝑝 𝒙𝟎 𝑝(𝒌𝟎) 𝑑𝒙 = − 𝛽 𝑡 2 𝒙 − 𝛽 𝑡 ∇𝒙𝑡 log 𝑝 𝒙𝑡 𝑑𝑡 + 𝛽(𝑡)𝑑 𝒘 𝑑𝒌 = − 𝛽 𝑡 2 𝒌 − 𝛽 𝑡 ∇𝒌𝑡 log 𝑝 𝒌𝑡 𝑑𝑡 + 𝛽(𝑡)𝑑 𝒘 16 / 25
  • 17. Diffusion model for blind deconvolution 𝑝 𝒚 𝒙0, 𝒌𝟎 ∝ 𝑝 𝒚 𝒙𝟎, 𝒌𝟎 𝑝 𝒙𝟎 𝑝(𝒌𝟎) 𝑑𝒙 = − 𝛽 𝑡 2 𝒙 − 𝛽 𝑡 [∇𝒙𝑡 log 𝑝 𝒙𝑡 + ∇𝒙𝑡 log 𝑝 𝒚 𝒙𝑡, 𝒌𝑡 ] 𝑑𝑡 + 𝛽(𝑡)𝑑 𝒘 𝑑𝒌 = − 𝛽 𝑡 2 𝒌 − 𝛽 𝑡 [∇𝒌𝑡 log 𝑝 𝒌𝑡 + ∇𝒌𝑡 log 𝑝 𝒚 𝒙𝑡, 𝒌𝑡 ] 𝑑𝑡 + 𝛽(𝑡)𝑑 𝒘 17 / 25
  • 18. Diffusion model for blind deconvolution 𝑝 𝒚 𝒙0, 𝒌𝟎 ∝ 𝑝 𝒚 𝒙𝟎, 𝒌𝟎 𝑝 𝒙𝟎 𝑝(𝒌𝟎) 𝑑𝒙 = − 𝛽 𝑡 2 𝒙 − 𝛽 𝑡 [∇𝒙𝑡 log 𝑝 𝒙𝑡 + ∇𝒙𝑡 log 𝑝 𝒚 𝒙0, 𝒌0 ] 𝑑𝑡 + 𝛽(𝑡)𝑑 𝒘 𝑑𝒌 = − 𝛽 𝑡 2 𝒌 − 𝛽 𝑡 [∇𝒌𝑡 log 𝑝 𝒌𝑡 + ∇𝒌𝑡 log 𝑝 𝒚 𝒙0, 𝒌0 ] 𝑑𝑡 + 𝛽(𝑡)𝑑 𝒘 Theorem. Under similar conditions as in DPS, ∇𝒙𝑡 log 𝑝 𝒚 𝒙𝑡, 𝒌𝑡 ≃ ∇𝐱𝑡 log 𝑝 𝒚 𝒙𝟎 𝒙𝑡 , 𝒌𝟎(𝒌𝑡) ∇𝒌𝑡 log 𝑝 𝒚 𝒙𝑡, 𝒌𝑡 ≃ ∇𝒌𝑡 log 𝑝 𝒚 𝒙𝟎 𝒙𝑡 , 𝒌𝟎(𝒌𝑡) 18 / 25
  • 19. Single diffusion model for non-blind inverse problem 19 / 25
  • 20. Parallel diffusion models for blind inverse problem 20 / 25
  • 21. Parallel diffusion models for blind inverse problem 21 / 25
  • 22. Applicability to diverse blind inverse problems 22 / 25
  • 23. Results: Blind Motion deblurring Measurement Ours Ground truth 23 / 25
  • 24. Results: Imaging through turbulence Measurement Ours Ground truth 24 / 25

Notes de l'éditeur

  1. Due to the equivalence between explicit score matching and denoising score matching, one can train the score function in the same manner as training a residual denoiser.
  2. When we train with multiple noise levels, we arrive at an interesting view of diffusion models, that one can view the data noising process as some linear forward SDE, and the data generating process as the corresponding reverse SDE, where the drift function is governed by the score function. Hence, when you want to sample data, you can discretize the reverse SDE and numerically solve it using the pre-trained score function.
  3. On the other hand, we are interested in solving inverse problems. In the inverse problem setting, our aim is to recover the ground truth x from the noisy measurement y, obtained through some integral imaging system A, polluted with the measurement noise eta. The problem is naturally ill-posed, which means that there exists infinitely many solutions to this problem. Hence, in order to correctly specify which one of the solutions is the one that we want, we need to specify the prior of the data distribution, in other words, how the images looks like.
  4. Examples of such inverse problems include inpainting, deconvolution, and compressed sensing MRI.
  5. Let’s consider the following measurement model. Now, given y, what we want is to sample from the posterior distribution p(x|y).
  6. Then, we can approximate the former term with the pre-trained score function, and the latter simply depends on the measurement model, for example, gaussian or poisson. Note that these gradients can be analytically computed as their functional forms are known. We see that for Gaussian, we are essentially performing gradient descent that minimizes the squared l2 norm of the residual, whereas for Poisson measurements, we are minimizing the squared weighted norm of the residual.
  7. Blind inverse problems consider the case where we do not know the forward operator.
  8. Here, we specifically constrain ourselves to cases where we know the functional form of the forward operator, but do not know the parameters.
  9. By far the most widely acknowledged application of such case is blind deconvolution, or blind deblurring problem, where we do not know which blur kernel generated the measurement.
  10. In order to solve this problem with diffusion, let us write the posterior distribution for blind deconvolution. We see that the posterior distribution is proportional to the product of likelihood,
  11. The image prior,
  12. And the kernel prior, and these two factor out as they are independent.
  13. Now in order to perform posterior sampling, we need to specify the prior model for the image distribution, and the kernel distribution. Note that the development in the field of blind deblurring can be attributed to building more efficient priors that estimate the true priors of the distribution. And our choice here is to use the diffusion prior for both the image, and the kernel.
  14. And in order to use the diffusion prior, we can simply train two score functions independently with denoising score matching. For the image score function, we can easily take some pre-trained model. For the kernel score function, the training is relatively easy, as the data distribution is much less complicated, and the size of the vector is relatively small.
  15. And specifically because those two are independent, our proposal is to construct two parallel diffusion processes for both the image, and the kernel.
  16. Now, in order to do posterior sampling, as in the non-blind inverse problem case, we need to incorporate the gradient of the likelihood, which is again intractable.
  17. However, we can again establish a similar theorem as in DPS, where we can approximate the intractable likelihood term by plugging in the denoised estimates.
  18. Now, visualizing what the equation says, let us first visualize what DPS was doing for the non-blind deblurring problem. As you see in the figure, we use the denoised estimate hat x0, and convolve with k_0 to obtain the measurement. We compute the residual with y, and backpropagate the difference to the next iteration.
  19. In this work of solving blind inverse problem, in which we call BlindDDPM, we simply add another diffusion model, and run the two in parallel. In the intermediate stage, we also use the score function for the kernel to estimate the denoised estimate, and convolve the two denoised estimates to compute the residual. Now the gradient update steps are applied to two streams separately.
  20. Now in practice we found that it is more robust to augment the diffusion prior with sparsity promoting ell_0 regularization.
  21. Until now, for the sake of brevity, I focused on my argument on blind deblurring, but the methodology can be extended to any cases where we know the functional form, and the parameter should be estimated. Another such case would be imaging through turbulence, where the forward model can be roughly approximated with tilt, followed by blur.
  22. When we visualize the results, we see that we can achieve state-of-the-art blind reconstruction results even from blind measurements. Note that the reconstruction of the image and the kernel are jointly performed
  23. We also achieve state-of-the-art results when we apply our method to imaging through turbulence.