SlideShare a Scribd company logo
DeepStrip: High Resolution Boundary Refinement
Hwang seung hyun
Yonsei University Severance Hospital CCIDS
University of Maryland & Adobe Research
CVPR 2020
2020.05.17
Introduction Related Work Methods and
Experiments
01 02 03
Conclusion
04
Yonsei Unversity Severance Hospital CCIDS
Contents
DeepStrip
Introduction – Proposal
• Boundary detection is a well-studied problem and fundamental
for human recognition
• Current methods are usually computed on low resolution(LR)
images, but most photos taken these days are much larger and
high resolution(HR) images
• Most studies simply upsample LR prediction to reach HR
prediction.
• Deep Strip targets on refining the boundaries in high resolution
images given low resolution masks
Introduction / Related Work / Methods and Experiments / Conclusion
DeepStrip
Introduction – Contributions
• Propose an approach to predict the boundary in a strip image, which is
computationally and memory wise efficient.
• To improve performance, propose novel losses including boundary distance, matching
and C0 continuity loss.
• Create a high resolution dataset “PixaHR” for evaluation.
Introduction / Related Work / Methods and Experiments / Conclusion
Related Work
1. Boundary Refinement
Introduction / Related Work / Methods and Experiments / Conclusion
• Explore rich convolutional features or fuse both low and high level features to
detect edges
• “Conditional Random Fields(CRF)”, “Graph Cuts”
• These methods mainly explore edge detection in LR images, while DeepStrip
target HR boundary refinement.
2. Active Contours
• “Snakes” (Active contour model)
• “Deep active contour” predict boundary pixels in a patch. But, cannot
guarantee a continuous boundary prediction
• These methods process the entire image or perform patch-based training,
which requires heavy computation and memory overhead
3. High Resolution Up-sampling
• Conventional methods reach HR segmentation masks by applying upsampling
to LR mask.
Methods and Experiments
DeepStrip – Architecture
Introduction / Related Work / Methods and Experiments / Conclusion
• Predict on strip image that captures the potential boundary region rather than the
entire HR image.
• Refines the edges on the strip image using a network
• Reconstruct prediction in the original image from the strip boundary prediction.
Methods and Experiments
DeepStrip – Strip Image Creation
Introduction / Related Work / Methods and Experiments / Conclusion
• Extract pixels near the upsampled boundary to create a strip image
• Use B-spline method to represent contour in the LR mask
• HR region along the normal direction at each point on the curve of the contour is extracted
• For GT label, add labels at the border of strip if no boundary pixel is included in strip image.
• If the strip height is large and multiple boundary pixels are included in each column, filter out
the extraneous boundaries that are not connected to the current one.
Methods and Experiments
DeepStrip – Strip Boundary Prediction
Introduction / Related Work / Methods and Experiments / Conclusion
• Train U-Net to predict the corresponding boundaries within the strip domain.
• Use instance normalization to apply for different resolution of images
• Extract the last upsampling layer and apply sigmoid function to predict all potential boundaries.
• Selection layer pick up the target boundary from potential boundaries
s = final output, x = initial prediction, m = softmax output of the selection layer
Methods and Experiments
DeepStrip – Loss Function
Introduction / Related Work / Methods and Experiments / Conclusion
1. Basic Loss Function (l1, Dice)
2. Boundary Distance Loss
3. Matching Loss (l1)
4. C0 Continuity Regularization (calculate
marginal difference between columns and penalize
the discontinuous position)
5. Total Loss
Methods and Experiments
DeepStrip – Strip Reconstruction at Inference stage
Introduction / Related Work / Methods and Experiments / Conclusion
• Mapping between the predicted strip boundaries and the full HR mask is required at
inference
• For every strip image, coordinates in the HR image are recorded for reconstruction
• Use dynamic programming similar to “seam carving” to find the path.
• Enables different strip sizes (width of strip) for different images
• Fix the height of strip, assuming all target boundaries are involved
w
h
Methods and Experiments
Dataset
Introduction / Related Work / Methods and Experiments / Conclusion
- DAVIS 2016 (benchmark for video segmentation, consists of 50 classes with
precise annotations in both 480P and 1080P)
- Pixa HR (100 manually annotated images with average resolution 7K x 7K)
• Downsample HR mask to LR by 8x, 16x, 32x for evaluation and training.
• Boundary-based F score for evaluation metrics
Methods and Experiments
Main Results
Introduction / Related Work / Methods and Experiments / Conclusion
* Baseline Model: only trained with l1 loss, without selection layer
Methods and Experiments
Main Results
Introduction / Related Work / Methods and Experiments / Conclusion
Methods and Experiments
Main Results
Introduction / Related Work / Methods and Experiments / Conclusion
Methods and Experiments
Main Results
Introduction / Related Work / Methods and Experiments / Conclusion
Methods and Experiments
Main Results
Introduction / Related Work / Methods and Experiments / Conclusion
Methods and Experiments
Ablation Studies
Introduction / Related Work / Methods and Experiments / Conclusion
Methods and Experiments
Ablation Studies
Introduction / Related Work / Methods and Experiments / Conclusion
• Performance increased when dividing the whole contour into 2 segments
which allows variable height for different regions
• Showed effectiveness of having flexible height
Conclusion
Introduction / Related Work / Methods and Experiments / Conclusion
• This paper presented a novel strategy to handle HR boundary refinement
computationally and memory efficiently given LR precise masks.
• Proposed extracting boundary regions along the upsampled boundary
spline to form strip images and make prediction within them.
• Boundary distance, matching loss, and C0 continuity regularization have
been proposed
• Current approach still has difficulty predicting complicated topology and
soft boundary regions
• Smarter adaptive strip height adjustment for every pixel might be a
potential solution

More Related Content

What's hot

SINGLE IMAGE SUPER RESOLUTION IN SPATIAL AND WAVELET DOMAIN
SINGLE IMAGE SUPER RESOLUTION IN SPATIAL AND WAVELET DOMAINSINGLE IMAGE SUPER RESOLUTION IN SPATIAL AND WAVELET DOMAIN
SINGLE IMAGE SUPER RESOLUTION IN SPATIAL AND WAVELET DOMAIN
ijma
 
Image segmentation based on color
Image segmentation based on colorImage segmentation based on color
Image segmentation based on color
eSAT Journals
 
FutureTech 2010
FutureTech 2010FutureTech 2010
FutureTech 2010
Dakshina Kisku
 
Ijetr021211
Ijetr021211Ijetr021211
Automatic left ventricle segmentation
Automatic left ventricle segmentationAutomatic left ventricle segmentation
Automatic left ventricle segmentation
ahmad abdelhafeez
 
Object video tracking using a pan tilt-zoom system
Object video tracking using a pan tilt-zoom systemObject video tracking using a pan tilt-zoom system
Object video tracking using a pan tilt-zoom system
Mohammed Abdalhakam Taha
 
4D AUTOMATIC LIP-READING FOR SPEAKER'S FACE IDENTIFCATION
4D AUTOMATIC LIP-READING FOR SPEAKER'S FACE IDENTIFCATION4D AUTOMATIC LIP-READING FOR SPEAKER'S FACE IDENTIFCATION
4D AUTOMATIC LIP-READING FOR SPEAKER'S FACE IDENTIFCATION
cscpconf
 
How useful is self-supervised pretraining for Visual tasks?
How useful is self-supervised pretraining for Visual tasks?How useful is self-supervised pretraining for Visual tasks?
How useful is self-supervised pretraining for Visual tasks?
Seunghyun Hwang
 
Tracking emerges by colorizing videos
Tracking emerges by colorizing videosTracking emerges by colorizing videos
Tracking emerges by colorizing videos
Oh Yoojin
 
Ibica2014(p15)image fusion based on broveywavelet
Ibica2014(p15)image fusion based on broveywaveletIbica2014(p15)image fusion based on broveywavelet
Ibica2014(p15)image fusion based on broveywavelet
Aboul Ella Hassanien
 
Imagefusfinalppt 140413102757-phpapp02
Imagefusfinalppt 140413102757-phpapp02Imagefusfinalppt 140413102757-phpapp02
Imagefusfinalppt 140413102757-phpapp02
Praveen Kumar
 
Further Improvements of CFA 3.0 by Combining Inpainting and Pansharpening Tec...
Further Improvements of CFA 3.0 by Combining Inpainting and Pansharpening Tec...Further Improvements of CFA 3.0 by Combining Inpainting and Pansharpening Tec...
Further Improvements of CFA 3.0 by Combining Inpainting and Pansharpening Tec...
sipij
 
Multifocus image fusion based on nsct
Multifocus image fusion based on nsctMultifocus image fusion based on nsct
Multifocus image fusion based on nsct
jpstudcorner
 
Multispectral Satellite Color Image Segmentation Using Fuzzy Based Innovative...
Multispectral Satellite Color Image Segmentation Using Fuzzy Based Innovative...Multispectral Satellite Color Image Segmentation Using Fuzzy Based Innovative...
Multispectral Satellite Color Image Segmentation Using Fuzzy Based Innovative...
Dibya Jyoti Bora
 
Object Elimination and Reconstruction Using an Effective Inpainting Method
Object Elimination and Reconstruction Using an Effective Inpainting MethodObject Elimination and Reconstruction Using an Effective Inpainting Method
Object Elimination and Reconstruction Using an Effective Inpainting Method
IOSR Journals
 
FickleNet: Weakly and Semi-supervised Semantic Image Segmentation using Stoch...
FickleNet: Weakly and Semi-supervised Semantic Image Segmentation using Stoch...FickleNet: Weakly and Semi-supervised Semantic Image Segmentation using Stoch...
FickleNet: Weakly and Semi-supervised Semantic Image Segmentation using Stoch...
Seunghyun Hwang
 
Image resolution enhancement via multi surface fitting
Image resolution enhancement via multi surface fittingImage resolution enhancement via multi surface fitting
Image resolution enhancement via multi surface fitting
International Journal of Science and Research (IJSR)
 
Fahad Fazal Elahi Guraya
Fahad Fazal Elahi GurayaFahad Fazal Elahi Guraya
Fahad Fazal Elahi Guraya
kimberleychen
 
conferense
conferense conferense
conferense
Omkar Shewale
 

What's hot (19)

SINGLE IMAGE SUPER RESOLUTION IN SPATIAL AND WAVELET DOMAIN
SINGLE IMAGE SUPER RESOLUTION IN SPATIAL AND WAVELET DOMAINSINGLE IMAGE SUPER RESOLUTION IN SPATIAL AND WAVELET DOMAIN
SINGLE IMAGE SUPER RESOLUTION IN SPATIAL AND WAVELET DOMAIN
 
Image segmentation based on color
Image segmentation based on colorImage segmentation based on color
Image segmentation based on color
 
FutureTech 2010
FutureTech 2010FutureTech 2010
FutureTech 2010
 
Ijetr021211
Ijetr021211Ijetr021211
Ijetr021211
 
Automatic left ventricle segmentation
Automatic left ventricle segmentationAutomatic left ventricle segmentation
Automatic left ventricle segmentation
 
Object video tracking using a pan tilt-zoom system
Object video tracking using a pan tilt-zoom systemObject video tracking using a pan tilt-zoom system
Object video tracking using a pan tilt-zoom system
 
4D AUTOMATIC LIP-READING FOR SPEAKER'S FACE IDENTIFCATION
4D AUTOMATIC LIP-READING FOR SPEAKER'S FACE IDENTIFCATION4D AUTOMATIC LIP-READING FOR SPEAKER'S FACE IDENTIFCATION
4D AUTOMATIC LIP-READING FOR SPEAKER'S FACE IDENTIFCATION
 
How useful is self-supervised pretraining for Visual tasks?
How useful is self-supervised pretraining for Visual tasks?How useful is self-supervised pretraining for Visual tasks?
How useful is self-supervised pretraining for Visual tasks?
 
Tracking emerges by colorizing videos
Tracking emerges by colorizing videosTracking emerges by colorizing videos
Tracking emerges by colorizing videos
 
Ibica2014(p15)image fusion based on broveywavelet
Ibica2014(p15)image fusion based on broveywaveletIbica2014(p15)image fusion based on broveywavelet
Ibica2014(p15)image fusion based on broveywavelet
 
Imagefusfinalppt 140413102757-phpapp02
Imagefusfinalppt 140413102757-phpapp02Imagefusfinalppt 140413102757-phpapp02
Imagefusfinalppt 140413102757-phpapp02
 
Further Improvements of CFA 3.0 by Combining Inpainting and Pansharpening Tec...
Further Improvements of CFA 3.0 by Combining Inpainting and Pansharpening Tec...Further Improvements of CFA 3.0 by Combining Inpainting and Pansharpening Tec...
Further Improvements of CFA 3.0 by Combining Inpainting and Pansharpening Tec...
 
Multifocus image fusion based on nsct
Multifocus image fusion based on nsctMultifocus image fusion based on nsct
Multifocus image fusion based on nsct
 
Multispectral Satellite Color Image Segmentation Using Fuzzy Based Innovative...
Multispectral Satellite Color Image Segmentation Using Fuzzy Based Innovative...Multispectral Satellite Color Image Segmentation Using Fuzzy Based Innovative...
Multispectral Satellite Color Image Segmentation Using Fuzzy Based Innovative...
 
Object Elimination and Reconstruction Using an Effective Inpainting Method
Object Elimination and Reconstruction Using an Effective Inpainting MethodObject Elimination and Reconstruction Using an Effective Inpainting Method
Object Elimination and Reconstruction Using an Effective Inpainting Method
 
FickleNet: Weakly and Semi-supervised Semantic Image Segmentation using Stoch...
FickleNet: Weakly and Semi-supervised Semantic Image Segmentation using Stoch...FickleNet: Weakly and Semi-supervised Semantic Image Segmentation using Stoch...
FickleNet: Weakly and Semi-supervised Semantic Image Segmentation using Stoch...
 
Image resolution enhancement via multi surface fitting
Image resolution enhancement via multi surface fittingImage resolution enhancement via multi surface fitting
Image resolution enhancement via multi surface fitting
 
Fahad Fazal Elahi Guraya
Fahad Fazal Elahi GurayaFahad Fazal Elahi Guraya
Fahad Fazal Elahi Guraya
 
conferense
conferense conferense
conferense
 

Similar to DeepStrip: High Resolution Boundary Refinement

Large Scale GAN Training for High Fidelity Natural Image Synthesis
Large Scale GAN Training for High Fidelity Natural Image SynthesisLarge Scale GAN Training for High Fidelity Natural Image Synthesis
Large Scale GAN Training for High Fidelity Natural Image Synthesis
Seunghyun Hwang
 
Lec16: Medical Image Registration (Advanced): Deformable Registration
Lec16: Medical Image Registration (Advanced): Deformable RegistrationLec16: Medical Image Registration (Advanced): Deformable Registration
Lec16: Medical Image Registration (Advanced): Deformable Registration
Ulaş Bağcı
 
Image enhancement
Image enhancementImage enhancement
Image enhancement
Ayaelshiwi
 
project_PPT_final
project_PPT_finalproject_PPT_final
project_PPT_final
Ranjan Ganguli
 
NEAL-2016 ARL Symposium Poster
NEAL-2016 ARL Symposium PosterNEAL-2016 ARL Symposium Poster
NEAL-2016 ARL Symposium Poster
Barbara Jean Neal
 
Moving object detection in complex scene
Moving object detection in complex sceneMoving object detection in complex scene
Moving object detection in complex scene
Kumar Mayank
 
Detecting Boundaries for Image Segmentation and Object Recognition
Detecting Boundaries for Image Segmentation and Object RecognitionDetecting Boundaries for Image Segmentation and Object Recognition
Detecting Boundaries for Image Segmentation and Object Recognition
IRJET Journal
 
E017443136
E017443136E017443136
E017443136
IOSR Journals
 
Anits dip
Anits dipAnits dip
seminar reprtv hdchjbjfkdbf dgusghdfs gsdgjsbk
seminar reprtv hdchjbjfkdbf dgusghdfs gsdgjsbkseminar reprtv hdchjbjfkdbf dgusghdfs gsdgjsbk
seminar reprtv hdchjbjfkdbf dgusghdfs gsdgjsbk
RajeshKotian11
 
Rethinking Data Augmentation for Image Super-resolution: A Comprehensive Anal...
Rethinking Data Augmentation for Image Super-resolution: A Comprehensive Anal...Rethinking Data Augmentation for Image Super-resolution: A Comprehensive Anal...
Rethinking Data Augmentation for Image Super-resolution: A Comprehensive Anal...
JaeJun Yoo
 
ResNeSt: Split-Attention Networks
ResNeSt: Split-Attention NetworksResNeSt: Split-Attention Networks
ResNeSt: Split-Attention Networks
Seunghyun Hwang
 
Lane detection by use of canny edge
Lane detection by use of canny edgeLane detection by use of canny edge
Lane detection by use of canny edge
banz23
 
Carved visual hulls for image based modeling
Carved visual hulls for image based modelingCarved visual hulls for image based modeling
Carved visual hulls for image based modeling
aftab alam
 
Enhanced characterness for text detection in the wild
Enhanced characterness for text detection in the wildEnhanced characterness for text detection in the wild
Enhanced characterness for text detection in the wild
Prerana Mukherjee
 
OpenCV.pdf
OpenCV.pdfOpenCV.pdf
OpenCV.pdf
sagarawasthi5
 
project final ppt.pptx
project final ppt.pptxproject final ppt.pptx
project final ppt.pptx
HarishKumarHarijan
 
Hands-On Machine Learning with Scikit-Learn and TensorFlow - Chapter8
Hands-On Machine Learning with Scikit-Learn and TensorFlow - Chapter8Hands-On Machine Learning with Scikit-Learn and TensorFlow - Chapter8
Hands-On Machine Learning with Scikit-Learn and TensorFlow - Chapter8
Hakky St
 
I010634450
I010634450I010634450
I010634450
IOSR Journals
 
Performance of Efficient Closed-Form Solution to Comprehensive Frontier Exposure
Performance of Efficient Closed-Form Solution to Comprehensive Frontier ExposurePerformance of Efficient Closed-Form Solution to Comprehensive Frontier Exposure
Performance of Efficient Closed-Form Solution to Comprehensive Frontier Exposure
iosrjce
 

Similar to DeepStrip: High Resolution Boundary Refinement (20)

Large Scale GAN Training for High Fidelity Natural Image Synthesis
Large Scale GAN Training for High Fidelity Natural Image SynthesisLarge Scale GAN Training for High Fidelity Natural Image Synthesis
Large Scale GAN Training for High Fidelity Natural Image Synthesis
 
Lec16: Medical Image Registration (Advanced): Deformable Registration
Lec16: Medical Image Registration (Advanced): Deformable RegistrationLec16: Medical Image Registration (Advanced): Deformable Registration
Lec16: Medical Image Registration (Advanced): Deformable Registration
 
Image enhancement
Image enhancementImage enhancement
Image enhancement
 
project_PPT_final
project_PPT_finalproject_PPT_final
project_PPT_final
 
NEAL-2016 ARL Symposium Poster
NEAL-2016 ARL Symposium PosterNEAL-2016 ARL Symposium Poster
NEAL-2016 ARL Symposium Poster
 
Moving object detection in complex scene
Moving object detection in complex sceneMoving object detection in complex scene
Moving object detection in complex scene
 
Detecting Boundaries for Image Segmentation and Object Recognition
Detecting Boundaries for Image Segmentation and Object RecognitionDetecting Boundaries for Image Segmentation and Object Recognition
Detecting Boundaries for Image Segmentation and Object Recognition
 
E017443136
E017443136E017443136
E017443136
 
Anits dip
Anits dipAnits dip
Anits dip
 
seminar reprtv hdchjbjfkdbf dgusghdfs gsdgjsbk
seminar reprtv hdchjbjfkdbf dgusghdfs gsdgjsbkseminar reprtv hdchjbjfkdbf dgusghdfs gsdgjsbk
seminar reprtv hdchjbjfkdbf dgusghdfs gsdgjsbk
 
Rethinking Data Augmentation for Image Super-resolution: A Comprehensive Anal...
Rethinking Data Augmentation for Image Super-resolution: A Comprehensive Anal...Rethinking Data Augmentation for Image Super-resolution: A Comprehensive Anal...
Rethinking Data Augmentation for Image Super-resolution: A Comprehensive Anal...
 
ResNeSt: Split-Attention Networks
ResNeSt: Split-Attention NetworksResNeSt: Split-Attention Networks
ResNeSt: Split-Attention Networks
 
Lane detection by use of canny edge
Lane detection by use of canny edgeLane detection by use of canny edge
Lane detection by use of canny edge
 
Carved visual hulls for image based modeling
Carved visual hulls for image based modelingCarved visual hulls for image based modeling
Carved visual hulls for image based modeling
 
Enhanced characterness for text detection in the wild
Enhanced characterness for text detection in the wildEnhanced characterness for text detection in the wild
Enhanced characterness for text detection in the wild
 
OpenCV.pdf
OpenCV.pdfOpenCV.pdf
OpenCV.pdf
 
project final ppt.pptx
project final ppt.pptxproject final ppt.pptx
project final ppt.pptx
 
Hands-On Machine Learning with Scikit-Learn and TensorFlow - Chapter8
Hands-On Machine Learning with Scikit-Learn and TensorFlow - Chapter8Hands-On Machine Learning with Scikit-Learn and TensorFlow - Chapter8
Hands-On Machine Learning with Scikit-Learn and TensorFlow - Chapter8
 
I010634450
I010634450I010634450
I010634450
 
Performance of Efficient Closed-Form Solution to Comprehensive Frontier Exposure
Performance of Efficient Closed-Form Solution to Comprehensive Frontier ExposurePerformance of Efficient Closed-Form Solution to Comprehensive Frontier Exposure
Performance of Efficient Closed-Form Solution to Comprehensive Frontier Exposure
 

More from Seunghyun Hwang

An annotation sparsification strategy for 3D medical image segmentation via r...
An annotation sparsification strategy for 3D medical image segmentation via r...An annotation sparsification strategy for 3D medical image segmentation via r...
An annotation sparsification strategy for 3D medical image segmentation via r...
Seunghyun Hwang
 
Do wide and deep networks learn the same things? Uncovering how neural networ...
Do wide and deep networks learn the same things? Uncovering how neural networ...Do wide and deep networks learn the same things? Uncovering how neural networ...
Do wide and deep networks learn the same things? Uncovering how neural networ...
Seunghyun Hwang
 
Deep Learning-based Fully Automated Detection and Quantification of Acute Inf...
Deep Learning-based Fully Automated Detection and Quantification of Acute Inf...Deep Learning-based Fully Automated Detection and Quantification of Acute Inf...
Deep Learning-based Fully Automated Detection and Quantification of Acute Inf...
Seunghyun Hwang
 
Diagnosis of Maxillary Sinusitis in Water’s view based on Deep learning model
Diagnosis of Maxillary Sinusitis in Water’s view based on Deep learning model Diagnosis of Maxillary Sinusitis in Water’s view based on Deep learning model
Diagnosis of Maxillary Sinusitis in Water’s view based on Deep learning model
Seunghyun Hwang
 
Energy-based Model for Out-of-Distribution Detection in Deep Medical Image Se...
Energy-based Model for Out-of-Distribution Detection in Deep Medical Image Se...Energy-based Model for Out-of-Distribution Detection in Deep Medical Image Se...
Energy-based Model for Out-of-Distribution Detection in Deep Medical Image Se...
Seunghyun Hwang
 
End-to-End Object Detection with Transformers
End-to-End Object Detection with TransformersEnd-to-End Object Detection with Transformers
End-to-End Object Detection with Transformers
Seunghyun Hwang
 
Deep Generative model-based quality control for cardiac MRI segmentation
Deep Generative model-based quality control for cardiac MRI segmentation Deep Generative model-based quality control for cardiac MRI segmentation
Deep Generative model-based quality control for cardiac MRI segmentation
Seunghyun Hwang
 
Segmenting Medical MRI via Recurrent Decoding Cell
Segmenting Medical MRI via Recurrent Decoding CellSegmenting Medical MRI via Recurrent Decoding Cell
Segmenting Medical MRI via Recurrent Decoding Cell
Seunghyun Hwang
 
Progressive learning and Disentanglement of hierarchical representations
Progressive learning and Disentanglement of hierarchical representationsProgressive learning and Disentanglement of hierarchical representations
Progressive learning and Disentanglement of hierarchical representations
Seunghyun Hwang
 
Learning Sparse Networks using Targeted Dropout
Learning Sparse Networks using Targeted DropoutLearning Sparse Networks using Targeted Dropout
Learning Sparse Networks using Targeted Dropout
Seunghyun Hwang
 
A Simple Framework for Contrastive Learning of Visual Representations
A Simple Framework for Contrastive Learning of Visual RepresentationsA Simple Framework for Contrastive Learning of Visual Representations
A Simple Framework for Contrastive Learning of Visual Representations
Seunghyun Hwang
 
Your Classifier is Secretly an Energy based model and you should treat it lik...
Your Classifier is Secretly an Energy based model and you should treat it lik...Your Classifier is Secretly an Energy based model and you should treat it lik...
Your Classifier is Secretly an Energy based model and you should treat it lik...
Seunghyun Hwang
 
Mix Conv: Mixed Depthwise Convolutional Kernels
Mix Conv: Mixed Depthwise Convolutional KernelsMix Conv: Mixed Depthwise Convolutional Kernels
Mix Conv: Mixed Depthwise Convolutional Kernels
Seunghyun Hwang
 

More from Seunghyun Hwang (13)

An annotation sparsification strategy for 3D medical image segmentation via r...
An annotation sparsification strategy for 3D medical image segmentation via r...An annotation sparsification strategy for 3D medical image segmentation via r...
An annotation sparsification strategy for 3D medical image segmentation via r...
 
Do wide and deep networks learn the same things? Uncovering how neural networ...
Do wide and deep networks learn the same things? Uncovering how neural networ...Do wide and deep networks learn the same things? Uncovering how neural networ...
Do wide and deep networks learn the same things? Uncovering how neural networ...
 
Deep Learning-based Fully Automated Detection and Quantification of Acute Inf...
Deep Learning-based Fully Automated Detection and Quantification of Acute Inf...Deep Learning-based Fully Automated Detection and Quantification of Acute Inf...
Deep Learning-based Fully Automated Detection and Quantification of Acute Inf...
 
Diagnosis of Maxillary Sinusitis in Water’s view based on Deep learning model
Diagnosis of Maxillary Sinusitis in Water’s view based on Deep learning model Diagnosis of Maxillary Sinusitis in Water’s view based on Deep learning model
Diagnosis of Maxillary Sinusitis in Water’s view based on Deep learning model
 
Energy-based Model for Out-of-Distribution Detection in Deep Medical Image Se...
Energy-based Model for Out-of-Distribution Detection in Deep Medical Image Se...Energy-based Model for Out-of-Distribution Detection in Deep Medical Image Se...
Energy-based Model for Out-of-Distribution Detection in Deep Medical Image Se...
 
End-to-End Object Detection with Transformers
End-to-End Object Detection with TransformersEnd-to-End Object Detection with Transformers
End-to-End Object Detection with Transformers
 
Deep Generative model-based quality control for cardiac MRI segmentation
Deep Generative model-based quality control for cardiac MRI segmentation Deep Generative model-based quality control for cardiac MRI segmentation
Deep Generative model-based quality control for cardiac MRI segmentation
 
Segmenting Medical MRI via Recurrent Decoding Cell
Segmenting Medical MRI via Recurrent Decoding CellSegmenting Medical MRI via Recurrent Decoding Cell
Segmenting Medical MRI via Recurrent Decoding Cell
 
Progressive learning and Disentanglement of hierarchical representations
Progressive learning and Disentanglement of hierarchical representationsProgressive learning and Disentanglement of hierarchical representations
Progressive learning and Disentanglement of hierarchical representations
 
Learning Sparse Networks using Targeted Dropout
Learning Sparse Networks using Targeted DropoutLearning Sparse Networks using Targeted Dropout
Learning Sparse Networks using Targeted Dropout
 
A Simple Framework for Contrastive Learning of Visual Representations
A Simple Framework for Contrastive Learning of Visual RepresentationsA Simple Framework for Contrastive Learning of Visual Representations
A Simple Framework for Contrastive Learning of Visual Representations
 
Your Classifier is Secretly an Energy based model and you should treat it lik...
Your Classifier is Secretly an Energy based model and you should treat it lik...Your Classifier is Secretly an Energy based model and you should treat it lik...
Your Classifier is Secretly an Energy based model and you should treat it lik...
 
Mix Conv: Mixed Depthwise Convolutional Kernels
Mix Conv: Mixed Depthwise Convolutional KernelsMix Conv: Mixed Depthwise Convolutional Kernels
Mix Conv: Mixed Depthwise Convolutional Kernels
 

Recently uploaded

UiPath Test Automation using UiPath Test Suite series, part 5
UiPath Test Automation using UiPath Test Suite series, part 5UiPath Test Automation using UiPath Test Suite series, part 5
UiPath Test Automation using UiPath Test Suite series, part 5
DianaGray10
 
Introducing Milvus Lite: Easy-to-Install, Easy-to-Use vector database for you...
Introducing Milvus Lite: Easy-to-Install, Easy-to-Use vector database for you...Introducing Milvus Lite: Easy-to-Install, Easy-to-Use vector database for you...
Introducing Milvus Lite: Easy-to-Install, Easy-to-Use vector database for you...
Zilliz
 
GraphSummit Singapore | The Art of the Possible with Graph - Q2 2024
GraphSummit Singapore | The Art of the  Possible with Graph - Q2 2024GraphSummit Singapore | The Art of the  Possible with Graph - Q2 2024
GraphSummit Singapore | The Art of the Possible with Graph - Q2 2024
Neo4j
 
Data structures and Algorithms in Python.pdf
Data structures and Algorithms in Python.pdfData structures and Algorithms in Python.pdf
Data structures and Algorithms in Python.pdf
TIPNGVN2
 
GraphSummit Singapore | Enhancing Changi Airport Group's Passenger Experience...
GraphSummit Singapore | Enhancing Changi Airport Group's Passenger Experience...GraphSummit Singapore | Enhancing Changi Airport Group's Passenger Experience...
GraphSummit Singapore | Enhancing Changi Airport Group's Passenger Experience...
Neo4j
 
Full-RAG: A modern architecture for hyper-personalization
Full-RAG: A modern architecture for hyper-personalizationFull-RAG: A modern architecture for hyper-personalization
Full-RAG: A modern architecture for hyper-personalization
Zilliz
 
Essentials of Automations: The Art of Triggers and Actions in FME
Essentials of Automations: The Art of Triggers and Actions in FMEEssentials of Automations: The Art of Triggers and Actions in FME
Essentials of Automations: The Art of Triggers and Actions in FME
Safe Software
 
How to Get CNIC Information System with Paksim Ga.pptx
How to Get CNIC Information System with Paksim Ga.pptxHow to Get CNIC Information System with Paksim Ga.pptx
How to Get CNIC Information System with Paksim Ga.pptx
danishmna97
 
Why You Should Replace Windows 11 with Nitrux Linux 3.5.0 for enhanced perfor...
Why You Should Replace Windows 11 with Nitrux Linux 3.5.0 for enhanced perfor...Why You Should Replace Windows 11 with Nitrux Linux 3.5.0 for enhanced perfor...
Why You Should Replace Windows 11 with Nitrux Linux 3.5.0 for enhanced perfor...
SOFTTECHHUB
 
GraphSummit Singapore | Neo4j Product Vision & Roadmap - Q2 2024
GraphSummit Singapore | Neo4j Product Vision & Roadmap - Q2 2024GraphSummit Singapore | Neo4j Product Vision & Roadmap - Q2 2024
GraphSummit Singapore | Neo4j Product Vision & Roadmap - Q2 2024
Neo4j
 
Goodbye Windows 11: Make Way for Nitrux Linux 3.5.0!
Goodbye Windows 11: Make Way for Nitrux Linux 3.5.0!Goodbye Windows 11: Make Way for Nitrux Linux 3.5.0!
Goodbye Windows 11: Make Way for Nitrux Linux 3.5.0!
SOFTTECHHUB
 
Video Streaming: Then, Now, and in the Future
Video Streaming: Then, Now, and in the FutureVideo Streaming: Then, Now, and in the Future
Video Streaming: Then, Now, and in the Future
Alpen-Adria-Universität
 
How to use Firebase Data Connect For Flutter
How to use Firebase Data Connect For FlutterHow to use Firebase Data Connect For Flutter
How to use Firebase Data Connect For Flutter
Daiki Mogmet Ito
 
Large Language Model (LLM) and it’s Geospatial Applications
Large Language Model (LLM) and it’s Geospatial ApplicationsLarge Language Model (LLM) and it’s Geospatial Applications
Large Language Model (LLM) and it’s Geospatial Applications
Rohit Gautam
 
20240607 QFM018 Elixir Reading List May 2024
20240607 QFM018 Elixir Reading List May 202420240607 QFM018 Elixir Reading List May 2024
20240607 QFM018 Elixir Reading List May 2024
Matthew Sinclair
 
Communications Mining Series - Zero to Hero - Session 1
Communications Mining Series - Zero to Hero - Session 1Communications Mining Series - Zero to Hero - Session 1
Communications Mining Series - Zero to Hero - Session 1
DianaGray10
 
Introduction to CHERI technology - Cybersecurity
Introduction to CHERI technology - CybersecurityIntroduction to CHERI technology - Cybersecurity
Introduction to CHERI technology - Cybersecurity
mikeeftimakis1
 
Let's Integrate MuleSoft RPA, COMPOSER, APM with AWS IDP along with Slack
Let's Integrate MuleSoft RPA, COMPOSER, APM with AWS IDP along with SlackLet's Integrate MuleSoft RPA, COMPOSER, APM with AWS IDP along with Slack
Let's Integrate MuleSoft RPA, COMPOSER, APM with AWS IDP along with Slack
shyamraj55
 
Enchancing adoption of Open Source Libraries. A case study on Albumentations.AI
Enchancing adoption of Open Source Libraries. A case study on Albumentations.AIEnchancing adoption of Open Source Libraries. A case study on Albumentations.AI
Enchancing adoption of Open Source Libraries. A case study on Albumentations.AI
Vladimir Iglovikov, Ph.D.
 
Unlock the Future of Search with MongoDB Atlas_ Vector Search Unleashed.pdf
Unlock the Future of Search with MongoDB Atlas_ Vector Search Unleashed.pdfUnlock the Future of Search with MongoDB Atlas_ Vector Search Unleashed.pdf
Unlock the Future of Search with MongoDB Atlas_ Vector Search Unleashed.pdf
Malak Abu Hammad
 

Recently uploaded (20)

UiPath Test Automation using UiPath Test Suite series, part 5
UiPath Test Automation using UiPath Test Suite series, part 5UiPath Test Automation using UiPath Test Suite series, part 5
UiPath Test Automation using UiPath Test Suite series, part 5
 
Introducing Milvus Lite: Easy-to-Install, Easy-to-Use vector database for you...
Introducing Milvus Lite: Easy-to-Install, Easy-to-Use vector database for you...Introducing Milvus Lite: Easy-to-Install, Easy-to-Use vector database for you...
Introducing Milvus Lite: Easy-to-Install, Easy-to-Use vector database for you...
 
GraphSummit Singapore | The Art of the Possible with Graph - Q2 2024
GraphSummit Singapore | The Art of the  Possible with Graph - Q2 2024GraphSummit Singapore | The Art of the  Possible with Graph - Q2 2024
GraphSummit Singapore | The Art of the Possible with Graph - Q2 2024
 
Data structures and Algorithms in Python.pdf
Data structures and Algorithms in Python.pdfData structures and Algorithms in Python.pdf
Data structures and Algorithms in Python.pdf
 
GraphSummit Singapore | Enhancing Changi Airport Group's Passenger Experience...
GraphSummit Singapore | Enhancing Changi Airport Group's Passenger Experience...GraphSummit Singapore | Enhancing Changi Airport Group's Passenger Experience...
GraphSummit Singapore | Enhancing Changi Airport Group's Passenger Experience...
 
Full-RAG: A modern architecture for hyper-personalization
Full-RAG: A modern architecture for hyper-personalizationFull-RAG: A modern architecture for hyper-personalization
Full-RAG: A modern architecture for hyper-personalization
 
Essentials of Automations: The Art of Triggers and Actions in FME
Essentials of Automations: The Art of Triggers and Actions in FMEEssentials of Automations: The Art of Triggers and Actions in FME
Essentials of Automations: The Art of Triggers and Actions in FME
 
How to Get CNIC Information System with Paksim Ga.pptx
How to Get CNIC Information System with Paksim Ga.pptxHow to Get CNIC Information System with Paksim Ga.pptx
How to Get CNIC Information System with Paksim Ga.pptx
 
Why You Should Replace Windows 11 with Nitrux Linux 3.5.0 for enhanced perfor...
Why You Should Replace Windows 11 with Nitrux Linux 3.5.0 for enhanced perfor...Why You Should Replace Windows 11 with Nitrux Linux 3.5.0 for enhanced perfor...
Why You Should Replace Windows 11 with Nitrux Linux 3.5.0 for enhanced perfor...
 
GraphSummit Singapore | Neo4j Product Vision & Roadmap - Q2 2024
GraphSummit Singapore | Neo4j Product Vision & Roadmap - Q2 2024GraphSummit Singapore | Neo4j Product Vision & Roadmap - Q2 2024
GraphSummit Singapore | Neo4j Product Vision & Roadmap - Q2 2024
 
Goodbye Windows 11: Make Way for Nitrux Linux 3.5.0!
Goodbye Windows 11: Make Way for Nitrux Linux 3.5.0!Goodbye Windows 11: Make Way for Nitrux Linux 3.5.0!
Goodbye Windows 11: Make Way for Nitrux Linux 3.5.0!
 
Video Streaming: Then, Now, and in the Future
Video Streaming: Then, Now, and in the FutureVideo Streaming: Then, Now, and in the Future
Video Streaming: Then, Now, and in the Future
 
How to use Firebase Data Connect For Flutter
How to use Firebase Data Connect For FlutterHow to use Firebase Data Connect For Flutter
How to use Firebase Data Connect For Flutter
 
Large Language Model (LLM) and it’s Geospatial Applications
Large Language Model (LLM) and it’s Geospatial ApplicationsLarge Language Model (LLM) and it’s Geospatial Applications
Large Language Model (LLM) and it’s Geospatial Applications
 
20240607 QFM018 Elixir Reading List May 2024
20240607 QFM018 Elixir Reading List May 202420240607 QFM018 Elixir Reading List May 2024
20240607 QFM018 Elixir Reading List May 2024
 
Communications Mining Series - Zero to Hero - Session 1
Communications Mining Series - Zero to Hero - Session 1Communications Mining Series - Zero to Hero - Session 1
Communications Mining Series - Zero to Hero - Session 1
 
Introduction to CHERI technology - Cybersecurity
Introduction to CHERI technology - CybersecurityIntroduction to CHERI technology - Cybersecurity
Introduction to CHERI technology - Cybersecurity
 
Let's Integrate MuleSoft RPA, COMPOSER, APM with AWS IDP along with Slack
Let's Integrate MuleSoft RPA, COMPOSER, APM with AWS IDP along with SlackLet's Integrate MuleSoft RPA, COMPOSER, APM with AWS IDP along with Slack
Let's Integrate MuleSoft RPA, COMPOSER, APM with AWS IDP along with Slack
 
Enchancing adoption of Open Source Libraries. A case study on Albumentations.AI
Enchancing adoption of Open Source Libraries. A case study on Albumentations.AIEnchancing adoption of Open Source Libraries. A case study on Albumentations.AI
Enchancing adoption of Open Source Libraries. A case study on Albumentations.AI
 
Unlock the Future of Search with MongoDB Atlas_ Vector Search Unleashed.pdf
Unlock the Future of Search with MongoDB Atlas_ Vector Search Unleashed.pdfUnlock the Future of Search with MongoDB Atlas_ Vector Search Unleashed.pdf
Unlock the Future of Search with MongoDB Atlas_ Vector Search Unleashed.pdf
 

DeepStrip: High Resolution Boundary Refinement

  • 1. DeepStrip: High Resolution Boundary Refinement Hwang seung hyun Yonsei University Severance Hospital CCIDS University of Maryland & Adobe Research CVPR 2020 2020.05.17
  • 2. Introduction Related Work Methods and Experiments 01 02 03 Conclusion 04 Yonsei Unversity Severance Hospital CCIDS Contents
  • 3. DeepStrip Introduction – Proposal • Boundary detection is a well-studied problem and fundamental for human recognition • Current methods are usually computed on low resolution(LR) images, but most photos taken these days are much larger and high resolution(HR) images • Most studies simply upsample LR prediction to reach HR prediction. • Deep Strip targets on refining the boundaries in high resolution images given low resolution masks Introduction / Related Work / Methods and Experiments / Conclusion
  • 4. DeepStrip Introduction – Contributions • Propose an approach to predict the boundary in a strip image, which is computationally and memory wise efficient. • To improve performance, propose novel losses including boundary distance, matching and C0 continuity loss. • Create a high resolution dataset “PixaHR” for evaluation. Introduction / Related Work / Methods and Experiments / Conclusion
  • 5. Related Work 1. Boundary Refinement Introduction / Related Work / Methods and Experiments / Conclusion • Explore rich convolutional features or fuse both low and high level features to detect edges • “Conditional Random Fields(CRF)”, “Graph Cuts” • These methods mainly explore edge detection in LR images, while DeepStrip target HR boundary refinement. 2. Active Contours • “Snakes” (Active contour model) • “Deep active contour” predict boundary pixels in a patch. But, cannot guarantee a continuous boundary prediction • These methods process the entire image or perform patch-based training, which requires heavy computation and memory overhead 3. High Resolution Up-sampling • Conventional methods reach HR segmentation masks by applying upsampling to LR mask.
  • 6. Methods and Experiments DeepStrip – Architecture Introduction / Related Work / Methods and Experiments / Conclusion • Predict on strip image that captures the potential boundary region rather than the entire HR image. • Refines the edges on the strip image using a network • Reconstruct prediction in the original image from the strip boundary prediction.
  • 7. Methods and Experiments DeepStrip – Strip Image Creation Introduction / Related Work / Methods and Experiments / Conclusion • Extract pixels near the upsampled boundary to create a strip image • Use B-spline method to represent contour in the LR mask • HR region along the normal direction at each point on the curve of the contour is extracted • For GT label, add labels at the border of strip if no boundary pixel is included in strip image. • If the strip height is large and multiple boundary pixels are included in each column, filter out the extraneous boundaries that are not connected to the current one.
  • 8. Methods and Experiments DeepStrip – Strip Boundary Prediction Introduction / Related Work / Methods and Experiments / Conclusion • Train U-Net to predict the corresponding boundaries within the strip domain. • Use instance normalization to apply for different resolution of images • Extract the last upsampling layer and apply sigmoid function to predict all potential boundaries. • Selection layer pick up the target boundary from potential boundaries s = final output, x = initial prediction, m = softmax output of the selection layer
  • 9. Methods and Experiments DeepStrip – Loss Function Introduction / Related Work / Methods and Experiments / Conclusion 1. Basic Loss Function (l1, Dice) 2. Boundary Distance Loss 3. Matching Loss (l1) 4. C0 Continuity Regularization (calculate marginal difference between columns and penalize the discontinuous position) 5. Total Loss
  • 10. Methods and Experiments DeepStrip – Strip Reconstruction at Inference stage Introduction / Related Work / Methods and Experiments / Conclusion • Mapping between the predicted strip boundaries and the full HR mask is required at inference • For every strip image, coordinates in the HR image are recorded for reconstruction • Use dynamic programming similar to “seam carving” to find the path. • Enables different strip sizes (width of strip) for different images • Fix the height of strip, assuming all target boundaries are involved w h
  • 11. Methods and Experiments Dataset Introduction / Related Work / Methods and Experiments / Conclusion - DAVIS 2016 (benchmark for video segmentation, consists of 50 classes with precise annotations in both 480P and 1080P) - Pixa HR (100 manually annotated images with average resolution 7K x 7K) • Downsample HR mask to LR by 8x, 16x, 32x for evaluation and training. • Boundary-based F score for evaluation metrics
  • 12. Methods and Experiments Main Results Introduction / Related Work / Methods and Experiments / Conclusion * Baseline Model: only trained with l1 loss, without selection layer
  • 13. Methods and Experiments Main Results Introduction / Related Work / Methods and Experiments / Conclusion
  • 14. Methods and Experiments Main Results Introduction / Related Work / Methods and Experiments / Conclusion
  • 15. Methods and Experiments Main Results Introduction / Related Work / Methods and Experiments / Conclusion
  • 16. Methods and Experiments Main Results Introduction / Related Work / Methods and Experiments / Conclusion
  • 17. Methods and Experiments Ablation Studies Introduction / Related Work / Methods and Experiments / Conclusion
  • 18. Methods and Experiments Ablation Studies Introduction / Related Work / Methods and Experiments / Conclusion • Performance increased when dividing the whole contour into 2 segments which allows variable height for different regions • Showed effectiveness of having flexible height
  • 19. Conclusion Introduction / Related Work / Methods and Experiments / Conclusion • This paper presented a novel strategy to handle HR boundary refinement computationally and memory efficiently given LR precise masks. • Proposed extracting boundary regions along the upsampled boundary spline to form strip images and make prediction within them. • Boundary distance, matching loss, and C0 continuity regularization have been proposed • Current approach still has difficulty predicting complicated topology and soft boundary regions • Smarter adaptive strip height adjustment for every pixel might be a potential solution