SlideShare a Scribd company logo
1 of 22
Solving 3D Inverse Problems using Pre-
trained 2D Diffusion Models
Dohoon Ryu Michael T. Mccann Marc L. Klasky Jong Chul Ye
1 / 63
Hyungjin Chung
CVPR 2023
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 / 63
http://yang-song.github.io/blog/2021/score/
Background: Diffusion models and Stein score 3 / 63
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 / 63
Background: Examples of inverse problems in 2D
𝐲
𝐱
• Inpainting
⨀
• Deblurring (Deconvolution)
𝐱 𝐲
∗
• CS-MRI
𝓕
⨀
5 / 63
Diffusion model-based inverse problem solvers in 2D 6 / 63
(DPS) Chung et al. ICLR 2023
Non-linear, noisy inverse problems
(DDRM) Kawar et al. NeurIPS 2022
Linear, noisy inverse problems
2D medical image inverse problem solving 7 / 63
5
view
2D medical image inverse problem solving: extend to 3D? 8 / 63
Axial
Sagittal
Coronal
𝑥
𝑦
𝑧
𝑥
𝑥
𝑦
FBP 2D diffusion GT
2D medical image inverse problem solving: extend to 3D? 9 / 63
Axial
Sagittal
Coronal
𝑥
𝑦
𝑧
𝑥
𝑥
𝑦
FBP 2D diffusion GT
- Significant artifacts
- No dependency across
slices
3D inverse problems in medical imaging 10 / 63
How do we construct 3D inverse problems with diffusion models?
 3D representations are memory heavy
• Voxels: Hard to deal with > 128^3 data
• Point clouds: Sparse representation, but not suitable for medical imaging
inverse problems
 3D voxel diffusion?
• The whole diffusion process stays in the data dimension
• Computationally too heavy
11 / 63
Augmenting 2D diffusion prior with model-based prior
Model-based prior (TV) 𝑇𝑉 𝒙 ≔ 𝐷𝑥𝒙, 𝐷𝑦𝒙, 𝐷𝑧𝒙
1
Diffusion prior ∇𝒙log 𝑝(𝒙)
12 / 63
Augmenting 2D diffusion prior with model-based prior
Model-based prior (TV: z) 𝑇𝑉
z 𝒙 ≔ 𝐷𝑧𝒙 1
Diffusion prior (xy) ∇𝒙xy
log 𝑝(𝒙xy)
13 / 63
Augmenting 2D diffusion prior with model-based prior
1. Denoising with score function (parallel)
𝒙′𝑖−1 ← 𝜎𝑖
2
− 𝜎𝑖−1
2
𝒔𝜃∗ 𝒙𝑡, 𝑡 + 𝜎𝑖
2
− 𝜎𝑖−1
2
𝝐
2. Data consistency + TV prior augmenting (joint)
𝒙i−1 ← argmin
𝒙′𝑖−1
1
2
‖𝒚 − 𝑨𝒙′𝑖−1‖2
2
+ ‖𝑫𝒛𝒙′𝑖−1‖1
14 / 63
Augmenting 2D diffusion prior with model-based prior
1. Denoising with score function (parallel)
𝒙′𝑖−1 ← 𝜎𝑖
2
− 𝜎𝑖−1
2
𝒔𝜃∗ 𝒙𝑡, 𝑡 + 𝜎𝑖
2
− 𝜎𝑖−1
2
𝝐
2. Data consistency + TV prior augmenting (joint)
𝒙i−1 ← argmin
𝒙′𝑖−1
1
2
‖𝒚 − 𝑨𝒙′𝑖−1‖2
2
+ ‖𝑫𝒛𝒙′𝑖−1‖1
Effectively solved with, e.g. ADMM
15 / 63
Augmenting 2D diffusion prior with model-based prior 16 / 63
Fast DiffusionMBIR
Sharing primal/dual variables
• Warm start
• Much faster convergence
17 / 63
Sharing primal/dual variables
• Warm start
• Much faster convergence
Score update (denoising)
• Prior in 𝑥𝑦 dimension
Fast DiffusionMBIR 18 / 63
Sharing primal/dual variables
• Warm start
• Much faster convergence
Score update (denoising)
• Prior in 𝑥𝑦 dimension
ADMM-TV iteration
• Data consistency
• Prior in 𝑧 dimension
Fast DiffusionMBIR 19 / 63
Results (8-view sparse view reconstruction)
13.36 / 0.404 14.19 / 0.537 15.65 / 0.674
33.34 / 0.938 34.23 / 0.968 34.06 / 0.960
Coherent results across the whole volume
20 / 63
Results: All three problems 21 / 63
Thank you!

More Related Content

What's hot

Image Interpolation Techniques with Optical and Digital Zoom Concepts
Image Interpolation Techniques with Optical and Digital Zoom ConceptsImage Interpolation Techniques with Optical and Digital Zoom Concepts
Image Interpolation Techniques with Optical and Digital Zoom Conceptsmmjalbiaty
 
COVID-19 detection using Chest X-rays.pdf
COVID-19 detection using Chest X-rays.pdfCOVID-19 detection using Chest X-rays.pdf
COVID-19 detection using Chest X-rays.pdfDeepakSahu480384
 
Transformers in Vision: From Zero to Hero
Transformers in Vision: From Zero to HeroTransformers in Vision: From Zero to Hero
Transformers in Vision: From Zero to HeroBill Liu
 
Generative Models for General Audiences
Generative Models for General AudiencesGenerative Models for General Audiences
Generative Models for General AudiencesSangwoo Mo
 
Lec13: Clustering Based Medical Image Segmentation Methods
Lec13: Clustering Based Medical Image Segmentation MethodsLec13: Clustering Based Medical Image Segmentation Methods
Lec13: Clustering Based Medical Image Segmentation MethodsUlaş Bağcı
 
ProxGen: Adaptive Proximal Gradient Methods for Structured Neural Networks (N...
ProxGen: Adaptive Proximal Gradient Methods for Structured Neural Networks (N...ProxGen: Adaptive Proximal Gradient Methods for Structured Neural Networks (N...
ProxGen: Adaptive Proximal Gradient Methods for Structured Neural Networks (N...Jihun Yun
 
Generative Adversarial Networks and Their Medical Imaging Applications
Generative Adversarial Networks and Their Medical Imaging ApplicationsGenerative Adversarial Networks and Their Medical Imaging Applications
Generative Adversarial Networks and Their Medical Imaging ApplicationsKyuhwan Jung
 
Transformer in Computer Vision
Transformer in Computer VisionTransformer in Computer Vision
Transformer in Computer VisionDongmin Choi
 
Image Processing with OpenCV
Image Processing with OpenCVImage Processing with OpenCV
Image Processing with OpenCVdebayanin
 
Naive Bayes Classifier | Naive Bayes Algorithm | Naive Bayes Classifier With ...
Naive Bayes Classifier | Naive Bayes Algorithm | Naive Bayes Classifier With ...Naive Bayes Classifier | Naive Bayes Algorithm | Naive Bayes Classifier With ...
Naive Bayes Classifier | Naive Bayes Algorithm | Naive Bayes Classifier With ...Simplilearn
 
Knn Algorithm presentation
Knn Algorithm presentationKnn Algorithm presentation
Knn Algorithm presentationRishavSharma112
 
Tutorial on Deep Generative Models
 Tutorial on Deep Generative Models Tutorial on Deep Generative Models
Tutorial on Deep Generative ModelsMLReview
 
Lec11: Active Contour and Level Set for Medical Image Segmentation
Lec11: Active Contour and Level Set for Medical Image SegmentationLec11: Active Contour and Level Set for Medical Image Segmentation
Lec11: Active Contour and Level Set for Medical Image SegmentationUlaş Bağcı
 
Unsupervised learning represenation with DCGAN
Unsupervised learning represenation with DCGANUnsupervised learning represenation with DCGAN
Unsupervised learning represenation with DCGANShyam Krishna Khadka
 
Focal loss for dense object detection
Focal loss for dense object detectionFocal loss for dense object detection
Focal loss for dense object detectionDaeHeeKim31
 
Introduction to CNN
Introduction to CNNIntroduction to CNN
Introduction to CNNShuai Zhang
 
Superpixel algorithms (whatershed, mean-shift, SLIC, BSLIC), Foolad
Superpixel algorithms (whatershed, mean-shift, SLIC, BSLIC), FooladSuperpixel algorithms (whatershed, mean-shift, SLIC, BSLIC), Foolad
Superpixel algorithms (whatershed, mean-shift, SLIC, BSLIC), FooladShima Foolad
 
Lec14: Evaluation Framework for Medical Image Segmentation
Lec14: Evaluation Framework for Medical Image SegmentationLec14: Evaluation Framework for Medical Image Segmentation
Lec14: Evaluation Framework for Medical Image SegmentationUlaş Bağcı
 

What's hot (20)

Image Interpolation Techniques with Optical and Digital Zoom Concepts
Image Interpolation Techniques with Optical and Digital Zoom ConceptsImage Interpolation Techniques with Optical and Digital Zoom Concepts
Image Interpolation Techniques with Optical and Digital Zoom Concepts
 
COVID-19 detection using Chest X-rays.pdf
COVID-19 detection using Chest X-rays.pdfCOVID-19 detection using Chest X-rays.pdf
COVID-19 detection using Chest X-rays.pdf
 
Transformers in Vision: From Zero to Hero
Transformers in Vision: From Zero to HeroTransformers in Vision: From Zero to Hero
Transformers in Vision: From Zero to Hero
 
Generative Models for General Audiences
Generative Models for General AudiencesGenerative Models for General Audiences
Generative Models for General Audiences
 
Lec13: Clustering Based Medical Image Segmentation Methods
Lec13: Clustering Based Medical Image Segmentation MethodsLec13: Clustering Based Medical Image Segmentation Methods
Lec13: Clustering Based Medical Image Segmentation Methods
 
ProxGen: Adaptive Proximal Gradient Methods for Structured Neural Networks (N...
ProxGen: Adaptive Proximal Gradient Methods for Structured Neural Networks (N...ProxGen: Adaptive Proximal Gradient Methods for Structured Neural Networks (N...
ProxGen: Adaptive Proximal Gradient Methods for Structured Neural Networks (N...
 
Swin transformer
Swin transformerSwin transformer
Swin transformer
 
Generative Adversarial Networks and Their Medical Imaging Applications
Generative Adversarial Networks and Their Medical Imaging ApplicationsGenerative Adversarial Networks and Their Medical Imaging Applications
Generative Adversarial Networks and Their Medical Imaging Applications
 
Transformer in Computer Vision
Transformer in Computer VisionTransformer in Computer Vision
Transformer in Computer Vision
 
Image Processing with OpenCV
Image Processing with OpenCVImage Processing with OpenCV
Image Processing with OpenCV
 
Depth estimation using deep learning
Depth estimation using deep learningDepth estimation using deep learning
Depth estimation using deep learning
 
Naive Bayes Classifier | Naive Bayes Algorithm | Naive Bayes Classifier With ...
Naive Bayes Classifier | Naive Bayes Algorithm | Naive Bayes Classifier With ...Naive Bayes Classifier | Naive Bayes Algorithm | Naive Bayes Classifier With ...
Naive Bayes Classifier | Naive Bayes Algorithm | Naive Bayes Classifier With ...
 
Knn Algorithm presentation
Knn Algorithm presentationKnn Algorithm presentation
Knn Algorithm presentation
 
Tutorial on Deep Generative Models
 Tutorial on Deep Generative Models Tutorial on Deep Generative Models
Tutorial on Deep Generative Models
 
Lec11: Active Contour and Level Set for Medical Image Segmentation
Lec11: Active Contour and Level Set for Medical Image SegmentationLec11: Active Contour and Level Set for Medical Image Segmentation
Lec11: Active Contour and Level Set for Medical Image Segmentation
 
Unsupervised learning represenation with DCGAN
Unsupervised learning represenation with DCGANUnsupervised learning represenation with DCGAN
Unsupervised learning represenation with DCGAN
 
Focal loss for dense object detection
Focal loss for dense object detectionFocal loss for dense object detection
Focal loss for dense object detection
 
Introduction to CNN
Introduction to CNNIntroduction to CNN
Introduction to CNN
 
Superpixel algorithms (whatershed, mean-shift, SLIC, BSLIC), Foolad
Superpixel algorithms (whatershed, mean-shift, SLIC, BSLIC), FooladSuperpixel algorithms (whatershed, mean-shift, SLIC, BSLIC), Foolad
Superpixel algorithms (whatershed, mean-shift, SLIC, BSLIC), Foolad
 
Lec14: Evaluation Framework for Medical Image Segmentation
Lec14: Evaluation Framework for Medical Image SegmentationLec14: Evaluation Framework for Medical Image Segmentation
Lec14: Evaluation Framework for Medical Image Segmentation
 

Similar to DiffusionMBIR_presentation_slide.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
 
BEGAN Boundary Equilibrium Generative Adversarial Networks
BEGAN Boundary Equilibrium Generative Adversarial NetworksBEGAN Boundary Equilibrium Generative Adversarial Networks
BEGAN Boundary Equilibrium Generative Adversarial NetworksChung-Il Kim
 
Super resolution in deep learning era - Jaejun Yoo
Super resolution in deep learning era - Jaejun YooSuper resolution in deep learning era - Jaejun Yoo
Super resolution in deep learning era - Jaejun YooJaeJun Yoo
 
Image classification using neural network
Image classification using neural networkImage classification using neural network
Image classification using neural networkBhavyateja Potineni
 
20230213_ComputerVision_연구.pptx
20230213_ComputerVision_연구.pptx20230213_ComputerVision_연구.pptx
20230213_ComputerVision_연구.pptxssuser7807522
 
Tutorial on Theory and Application of Generative Adversarial Networks
Tutorial on Theory and Application of Generative Adversarial NetworksTutorial on Theory and Application of Generative Adversarial Networks
Tutorial on Theory and Application of Generative Adversarial NetworksMLReview
 
Training DNN Models - II.pptx
Training DNN Models - II.pptxTraining DNN Models - II.pptx
Training DNN Models - II.pptxPrabhuSelvaraj15
 
Nonlinear dimension reduction
Nonlinear dimension reductionNonlinear dimension reduction
Nonlinear dimension reductionYan Xu
 
Effect of Block Sizes on the Attributes of Watermarking Digital Images
Effect of Block Sizes on the Attributes of Watermarking Digital ImagesEffect of Block Sizes on the Attributes of Watermarking Digital Images
Effect of Block Sizes on the Attributes of Watermarking Digital ImagesDr. Michael Agbaje
 
論文読み会@AIST (Deep Virtual Stereo Odometry [ECCV2018])
論文読み会@AIST (Deep Virtual Stereo Odometry [ECCV2018])論文読み会@AIST (Deep Virtual Stereo Odometry [ECCV2018])
論文読み会@AIST (Deep Virtual Stereo Odometry [ECCV2018])Masaya Kaneko
 
Patch-wise Deep Metric Learning for Unsupervised Low-Dose CT Denoising
Patch-wise Deep Metric Learning for Unsupervised Low-Dose CT DenoisingPatch-wise Deep Metric Learning for Unsupervised Low-Dose CT Denoising
Patch-wise Deep Metric Learning for Unsupervised Low-Dose CT Denoisingssuser886075
 
Theories and Engineering Technics of 2D-to-3D Back-Projection Problem
Theories and Engineering Technics of 2D-to-3D Back-Projection ProblemTheories and Engineering Technics of 2D-to-3D Back-Projection Problem
Theories and Engineering Technics of 2D-to-3D Back-Projection ProblemSeongcheol Baek
 
03 image transformations_i
03 image transformations_i03 image transformations_i
03 image transformations_iankit_ppt
 

Similar to DiffusionMBIR_presentation_slide.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)
 
Neural network
Neural networkNeural network
Neural network
 
Image denoising
Image denoisingImage denoising
Image denoising
 
ICASSP19
ICASSP19ICASSP19
ICASSP19
 
BEGAN Boundary Equilibrium Generative Adversarial Networks
BEGAN Boundary Equilibrium Generative Adversarial NetworksBEGAN Boundary Equilibrium Generative Adversarial Networks
BEGAN Boundary Equilibrium Generative Adversarial Networks
 
Super resolution in deep learning era - Jaejun Yoo
Super resolution in deep learning era - Jaejun YooSuper resolution in deep learning era - Jaejun Yoo
Super resolution in deep learning era - Jaejun Yoo
 
Image classification using neural network
Image classification using neural networkImage classification using neural network
Image classification using neural network
 
Thesis presentation
Thesis presentationThesis presentation
Thesis presentation
 
20230213_ComputerVision_연구.pptx
20230213_ComputerVision_연구.pptx20230213_ComputerVision_연구.pptx
20230213_ComputerVision_연구.pptx
 
Tutorial on Theory and Application of Generative Adversarial Networks
Tutorial on Theory and Application of Generative Adversarial NetworksTutorial on Theory and Application of Generative Adversarial Networks
Tutorial on Theory and Application of Generative Adversarial Networks
 
Training DNN Models - II.pptx
Training DNN Models - II.pptxTraining DNN Models - II.pptx
Training DNN Models - II.pptx
 
V2 v posenet
V2 v posenetV2 v posenet
V2 v posenet
 
Nonlinear dimension reduction
Nonlinear dimension reductionNonlinear dimension reduction
Nonlinear dimension reduction
 
Effect of Block Sizes on the Attributes of Watermarking Digital Images
Effect of Block Sizes on the Attributes of Watermarking Digital ImagesEffect of Block Sizes on the Attributes of Watermarking Digital Images
Effect of Block Sizes on the Attributes of Watermarking Digital Images
 
論文読み会@AIST (Deep Virtual Stereo Odometry [ECCV2018])
論文読み会@AIST (Deep Virtual Stereo Odometry [ECCV2018])論文読み会@AIST (Deep Virtual Stereo Odometry [ECCV2018])
論文読み会@AIST (Deep Virtual Stereo Odometry [ECCV2018])
 
Patch-wise Deep Metric Learning for Unsupervised Low-Dose CT Denoising
Patch-wise Deep Metric Learning for Unsupervised Low-Dose CT DenoisingPatch-wise Deep Metric Learning for Unsupervised Low-Dose CT Denoising
Patch-wise Deep Metric Learning for Unsupervised Low-Dose CT Denoising
 
riken-RBlur-slides.pptx
riken-RBlur-slides.pptxriken-RBlur-slides.pptx
riken-RBlur-slides.pptx
 
Theories and Engineering Technics of 2D-to-3D Back-Projection Problem
Theories and Engineering Technics of 2D-to-3D Back-Projection ProblemTheories and Engineering Technics of 2D-to-3D Back-Projection Problem
Theories and Engineering Technics of 2D-to-3D Back-Projection Problem
 
03 image transformations_i
03 image transformations_i03 image transformations_i
03 image transformations_i
 

More from Chung Hyung Jin

BlindDPS_presentation_silde.pptx
BlindDPS_presentation_silde.pptxBlindDPS_presentation_silde.pptx
BlindDPS_presentation_silde.pptxChung 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
 

More from Chung Hyung Jin (10)

BlindDPS_presentation_silde.pptx
BlindDPS_presentation_silde.pptxBlindDPS_presentation_silde.pptx
BlindDPS_presentation_silde.pptx
 
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...
 

Recently uploaded

HARMONY IN THE NATURE AND EXISTENCE - Unit-IV
HARMONY IN THE NATURE AND EXISTENCE - Unit-IVHARMONY IN THE NATURE AND EXISTENCE - Unit-IV
HARMONY IN THE NATURE AND EXISTENCE - Unit-IVRajaP95
 
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
 
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
 
Introduction to Multiple Access Protocol.pptx
Introduction to Multiple Access Protocol.pptxIntroduction to Multiple Access Protocol.pptx
Introduction to Multiple Access Protocol.pptxupamatechverse
 
VIP Call Girls Service Hitech City Hyderabad Call +91-8250192130
VIP Call Girls Service Hitech City Hyderabad Call +91-8250192130VIP Call Girls Service Hitech City Hyderabad Call +91-8250192130
VIP Call Girls Service Hitech City Hyderabad Call +91-8250192130Suhani Kapoor
 
Software Development Life Cycle By Team Orange (Dept. of Pharmacy)
Software Development Life Cycle By  Team Orange (Dept. of Pharmacy)Software Development Life Cycle By  Team Orange (Dept. of Pharmacy)
Software Development Life Cycle By Team Orange (Dept. of Pharmacy)Suman Mia
 
OSVC_Meta-Data based Simulation Automation to overcome Verification Challenge...
OSVC_Meta-Data based Simulation Automation to overcome Verification Challenge...OSVC_Meta-Data based Simulation Automation to overcome Verification Challenge...
OSVC_Meta-Data based Simulation Automation to overcome Verification Challenge...Soham Mondal
 
Porous Ceramics seminar and technical writing
Porous Ceramics seminar and technical writingPorous Ceramics seminar and technical writing
Porous Ceramics seminar and technical writingrakeshbaidya232001
 
(RIA) Call Girls Bhosari ( 7001035870 ) HI-Fi Pune Escorts Service
(RIA) Call Girls Bhosari ( 7001035870 ) HI-Fi Pune Escorts Service(RIA) Call Girls Bhosari ( 7001035870 ) HI-Fi Pune Escorts Service
(RIA) Call Girls Bhosari ( 7001035870 ) HI-Fi Pune Escorts Serviceranjana rawat
 
Coefficient of Thermal Expansion and their Importance.pptx
Coefficient of Thermal Expansion and their Importance.pptxCoefficient of Thermal Expansion and their Importance.pptx
Coefficient of Thermal Expansion and their Importance.pptxAsutosh Ranjan
 
main PPT.pptx of girls hostel security using rfid
main PPT.pptx of girls hostel security using rfidmain PPT.pptx of girls hostel security using rfid
main PPT.pptx of girls hostel security using rfidNikhilNagaraju
 
Sheet Pile Wall Design and Construction: A Practical Guide for Civil Engineer...
Sheet Pile Wall Design and Construction: A Practical Guide for Civil Engineer...Sheet Pile Wall Design and Construction: A Practical Guide for Civil Engineer...
Sheet Pile Wall Design and Construction: A Practical Guide for Civil Engineer...Dr.Costas Sachpazis
 
High Profile Call Girls Nagpur Meera Call 7001035870 Meet With Nagpur Escorts
High Profile Call Girls Nagpur Meera Call 7001035870 Meet With Nagpur EscortsHigh Profile Call Girls Nagpur Meera Call 7001035870 Meet With Nagpur Escorts
High Profile Call Girls Nagpur Meera Call 7001035870 Meet With Nagpur EscortsCall Girls in Nagpur High Profile
 
Call Girls Delhi {Jodhpur} 9711199012 high profile service
Call Girls Delhi {Jodhpur} 9711199012 high profile serviceCall Girls Delhi {Jodhpur} 9711199012 high profile service
Call Girls Delhi {Jodhpur} 9711199012 high profile servicerehmti665
 
Gfe Mayur Vihar Call Girls Service WhatsApp -> 9999965857 Available 24x7 ^ De...
Gfe Mayur Vihar Call Girls Service WhatsApp -> 9999965857 Available 24x7 ^ De...Gfe Mayur Vihar Call Girls Service WhatsApp -> 9999965857 Available 24x7 ^ De...
Gfe Mayur Vihar Call Girls Service WhatsApp -> 9999965857 Available 24x7 ^ De...srsj9000
 
The Most Attractive Pune Call Girls Budhwar Peth 8250192130 Will You Miss Thi...
The Most Attractive Pune Call Girls Budhwar Peth 8250192130 Will You Miss Thi...The Most Attractive Pune Call Girls Budhwar Peth 8250192130 Will You Miss Thi...
The Most Attractive Pune Call Girls Budhwar Peth 8250192130 Will You Miss Thi...ranjana rawat
 
SPICE PARK APR2024 ( 6,793 SPICE Models )
SPICE PARK APR2024 ( 6,793 SPICE Models )SPICE PARK APR2024 ( 6,793 SPICE Models )
SPICE PARK APR2024 ( 6,793 SPICE Models )Tsuyoshi Horigome
 
Microscopic Analysis of Ceramic Materials.pptx
Microscopic Analysis of Ceramic Materials.pptxMicroscopic Analysis of Ceramic Materials.pptx
Microscopic Analysis of Ceramic Materials.pptxpurnimasatapathy1234
 
Introduction to IEEE STANDARDS and its different types.pptx
Introduction to IEEE STANDARDS and its different types.pptxIntroduction to IEEE STANDARDS and its different types.pptx
Introduction to IEEE STANDARDS and its different types.pptxupamatechverse
 

Recently uploaded (20)

HARMONY IN THE NATURE AND EXISTENCE - Unit-IV
HARMONY IN THE NATURE AND EXISTENCE - Unit-IVHARMONY IN THE NATURE AND EXISTENCE - Unit-IV
HARMONY IN THE NATURE AND EXISTENCE - Unit-IV
 
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, ...
 
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
 
Introduction to Multiple Access Protocol.pptx
Introduction to Multiple Access Protocol.pptxIntroduction to Multiple Access Protocol.pptx
Introduction to Multiple Access Protocol.pptx
 
VIP Call Girls Service Hitech City Hyderabad Call +91-8250192130
VIP Call Girls Service Hitech City Hyderabad Call +91-8250192130VIP Call Girls Service Hitech City Hyderabad Call +91-8250192130
VIP Call Girls Service Hitech City Hyderabad Call +91-8250192130
 
Software Development Life Cycle By Team Orange (Dept. of Pharmacy)
Software Development Life Cycle By  Team Orange (Dept. of Pharmacy)Software Development Life Cycle By  Team Orange (Dept. of Pharmacy)
Software Development Life Cycle By Team Orange (Dept. of Pharmacy)
 
OSVC_Meta-Data based Simulation Automation to overcome Verification Challenge...
OSVC_Meta-Data based Simulation Automation to overcome Verification Challenge...OSVC_Meta-Data based Simulation Automation to overcome Verification Challenge...
OSVC_Meta-Data based Simulation Automation to overcome Verification Challenge...
 
Porous Ceramics seminar and technical writing
Porous Ceramics seminar and technical writingPorous Ceramics seminar and technical writing
Porous Ceramics seminar and technical writing
 
(RIA) Call Girls Bhosari ( 7001035870 ) HI-Fi Pune Escorts Service
(RIA) Call Girls Bhosari ( 7001035870 ) HI-Fi Pune Escorts Service(RIA) Call Girls Bhosari ( 7001035870 ) HI-Fi Pune Escorts Service
(RIA) Call Girls Bhosari ( 7001035870 ) HI-Fi Pune Escorts Service
 
Coefficient of Thermal Expansion and their Importance.pptx
Coefficient of Thermal Expansion and their Importance.pptxCoefficient of Thermal Expansion and their Importance.pptx
Coefficient of Thermal Expansion and their Importance.pptx
 
main PPT.pptx of girls hostel security using rfid
main PPT.pptx of girls hostel security using rfidmain PPT.pptx of girls hostel security using rfid
main PPT.pptx of girls hostel security using rfid
 
Sheet Pile Wall Design and Construction: A Practical Guide for Civil Engineer...
Sheet Pile Wall Design and Construction: A Practical Guide for Civil Engineer...Sheet Pile Wall Design and Construction: A Practical Guide for Civil Engineer...
Sheet Pile Wall Design and Construction: A Practical Guide for Civil Engineer...
 
High Profile Call Girls Nagpur Meera Call 7001035870 Meet With Nagpur Escorts
High Profile Call Girls Nagpur Meera Call 7001035870 Meet With Nagpur EscortsHigh Profile Call Girls Nagpur Meera Call 7001035870 Meet With Nagpur Escorts
High Profile Call Girls Nagpur Meera Call 7001035870 Meet With Nagpur Escorts
 
Call Girls Delhi {Jodhpur} 9711199012 high profile service
Call Girls Delhi {Jodhpur} 9711199012 high profile serviceCall Girls Delhi {Jodhpur} 9711199012 high profile service
Call Girls Delhi {Jodhpur} 9711199012 high profile service
 
Gfe Mayur Vihar Call Girls Service WhatsApp -> 9999965857 Available 24x7 ^ De...
Gfe Mayur Vihar Call Girls Service WhatsApp -> 9999965857 Available 24x7 ^ De...Gfe Mayur Vihar Call Girls Service WhatsApp -> 9999965857 Available 24x7 ^ De...
Gfe Mayur Vihar Call Girls Service WhatsApp -> 9999965857 Available 24x7 ^ De...
 
The Most Attractive Pune Call Girls Budhwar Peth 8250192130 Will You Miss Thi...
The Most Attractive Pune Call Girls Budhwar Peth 8250192130 Will You Miss Thi...The Most Attractive Pune Call Girls Budhwar Peth 8250192130 Will You Miss Thi...
The Most Attractive Pune Call Girls Budhwar Peth 8250192130 Will You Miss Thi...
 
SPICE PARK APR2024 ( 6,793 SPICE Models )
SPICE PARK APR2024 ( 6,793 SPICE Models )SPICE PARK APR2024 ( 6,793 SPICE Models )
SPICE PARK APR2024 ( 6,793 SPICE Models )
 
Microscopic Analysis of Ceramic Materials.pptx
Microscopic Analysis of Ceramic Materials.pptxMicroscopic Analysis of Ceramic Materials.pptx
Microscopic Analysis of Ceramic Materials.pptx
 
DJARUM4D - SLOT GACOR ONLINE | SLOT DEMO ONLINE
DJARUM4D - SLOT GACOR ONLINE | SLOT DEMO ONLINEDJARUM4D - SLOT GACOR ONLINE | SLOT DEMO ONLINE
DJARUM4D - SLOT GACOR ONLINE | SLOT DEMO ONLINE
 
Introduction to IEEE STANDARDS and its different types.pptx
Introduction to IEEE STANDARDS and its different types.pptxIntroduction to IEEE STANDARDS and its different types.pptx
Introduction to IEEE STANDARDS and its different types.pptx
 

DiffusionMBIR_presentation_slide.pptx

  • 1. Solving 3D Inverse Problems using Pre- trained 2D Diffusion Models Dohoon Ryu Michael T. Mccann Marc L. Klasky Jong Chul Ye 1 / 63 Hyungjin Chung CVPR 2023
  • 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 / 63
  • 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 / 63
  • 5. Background: Examples of inverse problems in 2D 𝐲 𝐱 • Inpainting ⨀ • Deblurring (Deconvolution) 𝐱 𝐲 ∗ • CS-MRI 𝓕 ⨀ 5 / 63
  • 6. Diffusion model-based inverse problem solvers in 2D 6 / 63 (DPS) Chung et al. ICLR 2023 Non-linear, noisy inverse problems (DDRM) Kawar et al. NeurIPS 2022 Linear, noisy inverse problems
  • 7. 2D medical image inverse problem solving 7 / 63 5 view
  • 8. 2D medical image inverse problem solving: extend to 3D? 8 / 63 Axial Sagittal Coronal 𝑥 𝑦 𝑧 𝑥 𝑥 𝑦 FBP 2D diffusion GT
  • 9. 2D medical image inverse problem solving: extend to 3D? 9 / 63 Axial Sagittal Coronal 𝑥 𝑦 𝑧 𝑥 𝑥 𝑦 FBP 2D diffusion GT - Significant artifacts - No dependency across slices
  • 10. 3D inverse problems in medical imaging 10 / 63
  • 11. How do we construct 3D inverse problems with diffusion models?  3D representations are memory heavy • Voxels: Hard to deal with > 128^3 data • Point clouds: Sparse representation, but not suitable for medical imaging inverse problems  3D voxel diffusion? • The whole diffusion process stays in the data dimension • Computationally too heavy 11 / 63
  • 12. Augmenting 2D diffusion prior with model-based prior Model-based prior (TV) 𝑇𝑉 𝒙 ≔ 𝐷𝑥𝒙, 𝐷𝑦𝒙, 𝐷𝑧𝒙 1 Diffusion prior ∇𝒙log 𝑝(𝒙) 12 / 63
  • 13. Augmenting 2D diffusion prior with model-based prior Model-based prior (TV: z) 𝑇𝑉 z 𝒙 ≔ 𝐷𝑧𝒙 1 Diffusion prior (xy) ∇𝒙xy log 𝑝(𝒙xy) 13 / 63
  • 14. Augmenting 2D diffusion prior with model-based prior 1. Denoising with score function (parallel) 𝒙′𝑖−1 ← 𝜎𝑖 2 − 𝜎𝑖−1 2 𝒔𝜃∗ 𝒙𝑡, 𝑡 + 𝜎𝑖 2 − 𝜎𝑖−1 2 𝝐 2. Data consistency + TV prior augmenting (joint) 𝒙i−1 ← argmin 𝒙′𝑖−1 1 2 ‖𝒚 − 𝑨𝒙′𝑖−1‖2 2 + ‖𝑫𝒛𝒙′𝑖−1‖1 14 / 63
  • 15. Augmenting 2D diffusion prior with model-based prior 1. Denoising with score function (parallel) 𝒙′𝑖−1 ← 𝜎𝑖 2 − 𝜎𝑖−1 2 𝒔𝜃∗ 𝒙𝑡, 𝑡 + 𝜎𝑖 2 − 𝜎𝑖−1 2 𝝐 2. Data consistency + TV prior augmenting (joint) 𝒙i−1 ← argmin 𝒙′𝑖−1 1 2 ‖𝒚 − 𝑨𝒙′𝑖−1‖2 2 + ‖𝑫𝒛𝒙′𝑖−1‖1 Effectively solved with, e.g. ADMM 15 / 63
  • 16. Augmenting 2D diffusion prior with model-based prior 16 / 63
  • 17. Fast DiffusionMBIR Sharing primal/dual variables • Warm start • Much faster convergence 17 / 63
  • 18. Sharing primal/dual variables • Warm start • Much faster convergence Score update (denoising) • Prior in 𝑥𝑦 dimension Fast DiffusionMBIR 18 / 63
  • 19. Sharing primal/dual variables • Warm start • Much faster convergence Score update (denoising) • Prior in 𝑥𝑦 dimension ADMM-TV iteration • Data consistency • Prior in 𝑧 dimension Fast DiffusionMBIR 19 / 63
  • 20. Results (8-view sparse view reconstruction) 13.36 / 0.404 14.19 / 0.537 15.65 / 0.674 33.34 / 0.938 34.23 / 0.968 34.06 / 0.960 Coherent results across the whole volume 20 / 63
  • 21. Results: All three problems 21 / 63

Editor's Notes

  1. So far we’ve considered noisy, non-linear, and blind. Our final piece will be to apply 2D diffusion to 3D inverse problems, especially in the context of medical imaging.
  2. In order to circumvent the intractable explicit score matching, you train your network with denoising score matching, which boils down to essentially training a residual denoiser.
  3. Another interesting view is 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.
  4. On the other hand, we are interested in solving inverse problems. In the inverse problem setting, our aim is to recover the grond truth x from the noisy measurement y, obtained through some integral imaging system A, polluted with the measurement noise n. 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.
  5. Examples of such inverse problems include inpainting, deconvolution, and compressed sensing MRI.
  6. Examples of such inverse problems include inpainting, deconvolution, and compressed sensing MRI.
  7. By 3D inverse problem, we can think of the following three representative examples. In the left, we have limited angle CT, where we only have measurements in the limited green angles. Sparse view CT considers the case where we have very few projections at hand, for example 4-view CT. Finally, compressed sensing MRI of 3D volumes is also the case of 3D inverse problem.
  8. To first look at why this problem is even interesting in the first place, one should note that there is no representation that everybody uses for 3D data. For 2D images, we do have a consensus, use discretized matrices to represent image. For 3D, using voxels isn’t so desirable in many use cases. It’s just too memory inefficient, and tends to be very heavy if we try to scale over 128 cubed resolution. These are the main reasons that the computer vision folks like to use meshes, point clouds, or more recently neural fields. However, using representations other than voxels are not suitable for medical imaging inverse problems, when seeing the interior of the 3D volume is of absolute necessity. Well, then can we use 3D voxel diffusion? The answer is absolutely not, since the diffusion model is already computationally too heavy with 2D diffusion, and our tiny GPUs would not be able to handle such massive memory.
  9. So our proposal is to leave 2D diffusion as is, and try to incorporate lessons from the more classic literature. Specifically, we bring the total variation prior, which essentially states that natural images tend to have sparse gradients.
  10. In particular, we choose to impose TV only to the redundant z-direction, as for xy dimension, the diffusion prior pretty much covers it all. Apologies for the sloppy notation, but I think you can get the point.
  11. We can implement our idea as follows. First, we denoise each slice with 2D score functions in parallel. If we only use step 1 iteratively, we get slices that are independent of each other, such that when seen from the coronal or the sagittal slices, they are incoherent. Hence, in the second step, we aggregate the slices to form the volume, and minimize the following sub-problem, which imposes both the data consistency condition, and the TV prior to the z-direction.
  12. The optimization can be performed effectively, with for example, ADMM.
  13. We can visualize our method as follows. However, if we were to naively implement this approach, it would mean that we would have to run the ADMM steps per every iteration, which would be infeasible.
  14. Hence, we propose another strategy called variable sharing, in which we treat all the primal and the dual variables of ADMM as global variables that are shared through the whole reverse diffusion process. This would mean that we are warm-starting all the ADMM and the CG steps in-between, which will lead to much faster convergence with limited number of steps.
  15. So like I explained with the figure, we first denoise with the score function,
  16. And then we’ll apply ADMM-TV, but we will run a single sweep of ADMM, with a single sweep of CG. You might think that there’s no way that this is going to converge, but it does, since we can slowly update the estimates during the reverse diffusion process.
  17. With the proposed method, we are able to achieve reconstructions from extremely limited number of views.
  18. In the paper, we show that we can apply this same methodology to all the three problems, including sparse-view tomography, limited-angle tomography, and compressed sensing MRI.
  19. So far we’ve considered noisy, non-linear, and blind. Our final piece will be to apply 2D diffusion to 3D inverse problems, especially in the context of medical imaging.