SlideShare a Scribd company logo
1 of 23
Download to read offline
2020/08/28
Centernet: Object as Points
Arithmer Dynamics Div. Christian Saravia
Self-Introduction
Christian Martin Saravia Hernandez
● Graduate School
○ Universidad Peruana de Ciencias Aplicadas
○ Bs. Software Engineering
● Former Job
○ Ficha Inc
■ Algorithm Development Engineer
● Research & application of deep learning in computer vision problems
● Current Job
○ Application of machine learning / deep learning to computer vision problems
■ Object detection
■ Object classification
Purpose
Purpose of this material
● Understand an anchor free approach object detection algorithm
Agenda
Agenda
● Current object detection approaches
● Centernet approach
● Object as Points
● Training
○ Keypoint heatmap
○ Local offset
○ Size prediction
○ Loss function
● Network Architecture
○ DLA
○ Modified DLA
● Inference
● Results
Background
Current approaches
● Object detections model (such as Yolo, SSD, etc.) rely on the usage of anchor boxes
● Anchor boxes are not completely optimal:
○ Wasteful: SSD300 does 8732 detections per class, and yolo448 does 98 detections per class,
which means that most of the box are discarded
○ Inefficient: We have to process all the boxes (even we will discard them later), which comes with
more processing time
○ Require post processing: like non-max suppression algorithm
○ Fixed: SSD requires fixed scale and steps of boxes, while yolov3 fixes the size of the anchors per
detection level
Centernet
Centernet approach
● End-to-end differentiable solution
● Relies on keypoint estimation to find the center points and regress all other object properties(such as
size)
● As a result, the model is simpler, faster and more accurate than bounding-box based detectors
Speed-accuracy trade-off on COCO dataset
Object as Points
Training
Keypoint Heatmap
Keypoint Heatmap
* If two gaussians of the same class overlap, take the element-wise maximum
Keypoint Heatmap
Loss calculation
Local Offset
Motivation
Local Offset
Loss calculation
Size Prediction
Motivation
Size Prediction
Loss calculation
Loss Function
Network Architecture
● Authors experiment with different backbone architectures, obtaining different results:
● The backbone that produces best speed/accuracy tradeoff is DLA-34 (modified by authors)
Results without test augmentation(N.A.), flip testing(F), and multi-scale augmentation(MS). HW: Intel
Core i7-8086k CPU, Titan Xp GPU
Network Architecture
● Deep Layer Aggregation:
Iterative deep aggregation
Upsample 2x
Aggregation node:
Conv2d(1x1)
Batch Normalization
Relu
Stage node:
Conv2d(3x3)
Batch Normalization
Relu
Conv2d(3x3)
Batch Normalization
Relu
Network Architecture
● Modified Deep Layer Aggregation:
Iterative deep aggregation
Upsample 2x
Deformable convolution
Aggregation node:
Conv2d(1x1)
Batch Normalization
Relu
Stage node:
Conv2d(3x3)
Batch Normalization
Relu
Conv2d(3x3)
Batch Normalization
Relu
Inference
Points to boxes
Results
● Object detection
● Human pose estimation
Results
Experimental results on Pascal VOC 2007 test. The
results are shown in mAP@0.5. Flip test is used for
Centernet. The FPSs for other methods are
copied from the original publications
COCO test-dev. Frame-per-second (FPS) were measured
on the same machine whenever possible. Italic FPS
highlight the cases, where the performance measure was
copied from the original publication
23

More Related Content

What's hot

Content-Based Image Retrieval (D2L6 Insight@DCU Machine Learning Workshop 2017)
Content-Based Image Retrieval (D2L6 Insight@DCU Machine Learning Workshop 2017)Content-Based Image Retrieval (D2L6 Insight@DCU Machine Learning Workshop 2017)
Content-Based Image Retrieval (D2L6 Insight@DCU Machine Learning Workshop 2017)Universitat Politècnica de Catalunya
 
Faster R-CNN - PR012
Faster R-CNN - PR012Faster R-CNN - PR012
Faster R-CNN - PR012Jinwon Lee
 
GAN - Theory and Applications
GAN - Theory and ApplicationsGAN - Theory and Applications
GAN - Theory and ApplicationsEmanuele Ghelfi
 
Introduction to object detection
Introduction to object detectionIntroduction to object detection
Introduction to object detectionBrodmann17
 
Faster R-CNN: Towards real-time object detection with region proposal network...
Faster R-CNN: Towards real-time object detection with region proposal network...Faster R-CNN: Towards real-time object detection with region proposal network...
Faster R-CNN: Towards real-time object detection with region proposal network...Universitat Politècnica de Catalunya
 
PR-146: CornerNet detecting objects as paired keypoints
PR-146: CornerNet detecting objects as paired keypointsPR-146: CornerNet detecting objects as paired keypoints
PR-146: CornerNet detecting objects as paired keypointsjaewon lee
 
Object detection presentation
Object detection presentationObject detection presentation
Object detection presentationAshwinBicholiya
 
[Mmlab seminar 2016] deep learning for human pose estimation
[Mmlab seminar 2016] deep learning for human pose estimation[Mmlab seminar 2016] deep learning for human pose estimation
[Mmlab seminar 2016] deep learning for human pose estimationWei Yang
 
Image Classification using deep learning
Image Classification using deep learning Image Classification using deep learning
Image Classification using deep learning Asma-AH
 
` Traffic Classification based on Machine Learning
` Traffic Classification based on Machine Learning ` Traffic Classification based on Machine Learning
` Traffic Classification based on Machine Learning butest
 
Tutorial on Object Detection (Faster R-CNN)
Tutorial on Object Detection (Faster R-CNN)Tutorial on Object Detection (Faster R-CNN)
Tutorial on Object Detection (Faster R-CNN)Hwa Pyung Kim
 
Introduction to Grad-CAM (complete version)
Introduction to Grad-CAM (complete version)Introduction to Grad-CAM (complete version)
Introduction to Grad-CAM (complete version)Hsing-chuan Hsieh
 
You only look once (YOLO) : unified real time object detection
You only look once (YOLO) : unified real time object detectionYou only look once (YOLO) : unified real time object detection
You only look once (YOLO) : unified real time object detectionEntrepreneur / Startup
 
[251] implementing deep learning using cu dnn
[251] implementing deep learning using cu dnn[251] implementing deep learning using cu dnn
[251] implementing deep learning using cu dnnNAVER D2
 
Survey of Attention mechanism & Use in Computer Vision
Survey of Attention mechanism & Use in Computer VisionSurvey of Attention mechanism & Use in Computer Vision
Survey of Attention mechanism & Use in Computer VisionSwatiNarkhede1
 

What's hot (20)

Content-Based Image Retrieval (D2L6 Insight@DCU Machine Learning Workshop 2017)
Content-Based Image Retrieval (D2L6 Insight@DCU Machine Learning Workshop 2017)Content-Based Image Retrieval (D2L6 Insight@DCU Machine Learning Workshop 2017)
Content-Based Image Retrieval (D2L6 Insight@DCU Machine Learning Workshop 2017)
 
Faster R-CNN - PR012
Faster R-CNN - PR012Faster R-CNN - PR012
Faster R-CNN - PR012
 
GAN - Theory and Applications
GAN - Theory and ApplicationsGAN - Theory and Applications
GAN - Theory and Applications
 
Introduction to object detection
Introduction to object detectionIntroduction to object detection
Introduction to object detection
 
Faster R-CNN: Towards real-time object detection with region proposal network...
Faster R-CNN: Towards real-time object detection with region proposal network...Faster R-CNN: Towards real-time object detection with region proposal network...
Faster R-CNN: Towards real-time object detection with region proposal network...
 
Faster rcnn
Faster rcnnFaster rcnn
Faster rcnn
 
Mask R-CNN
Mask R-CNNMask R-CNN
Mask R-CNN
 
PR-146: CornerNet detecting objects as paired keypoints
PR-146: CornerNet detecting objects as paired keypointsPR-146: CornerNet detecting objects as paired keypoints
PR-146: CornerNet detecting objects as paired keypoints
 
Object detection presentation
Object detection presentationObject detection presentation
Object detection presentation
 
[Mmlab seminar 2016] deep learning for human pose estimation
[Mmlab seminar 2016] deep learning for human pose estimation[Mmlab seminar 2016] deep learning for human pose estimation
[Mmlab seminar 2016] deep learning for human pose estimation
 
YOLO
YOLOYOLO
YOLO
 
Image Classification using deep learning
Image Classification using deep learning Image Classification using deep learning
Image Classification using deep learning
 
` Traffic Classification based on Machine Learning
` Traffic Classification based on Machine Learning ` Traffic Classification based on Machine Learning
` Traffic Classification based on Machine Learning
 
Tutorial on Object Detection (Faster R-CNN)
Tutorial on Object Detection (Faster R-CNN)Tutorial on Object Detection (Faster R-CNN)
Tutorial on Object Detection (Faster R-CNN)
 
Introduction to Grad-CAM (complete version)
Introduction to Grad-CAM (complete version)Introduction to Grad-CAM (complete version)
Introduction to Grad-CAM (complete version)
 
You only look once (YOLO) : unified real time object detection
You only look once (YOLO) : unified real time object detectionYou only look once (YOLO) : unified real time object detection
You only look once (YOLO) : unified real time object detection
 
[251] implementing deep learning using cu dnn
[251] implementing deep learning using cu dnn[251] implementing deep learning using cu dnn
[251] implementing deep learning using cu dnn
 
CIFAR-10
CIFAR-10CIFAR-10
CIFAR-10
 
Survey of Attention mechanism & Use in Computer Vision
Survey of Attention mechanism & Use in Computer VisionSurvey of Attention mechanism & Use in Computer Vision
Survey of Attention mechanism & Use in Computer Vision
 
Deep learning
Deep learningDeep learning
Deep learning
 

Similar to Centernet

Volodymyr Lyubinets. One startup's journey of building ML pipelines for text ...
Volodymyr Lyubinets. One startup's journey of building ML pipelines for text ...Volodymyr Lyubinets. One startup's journey of building ML pipelines for text ...
Volodymyr Lyubinets. One startup's journey of building ML pipelines for text ...Lviv Startup Club
 
Software Design Practices for Large-Scale Automation
Software Design Practices for Large-Scale AutomationSoftware Design Practices for Large-Scale Automation
Software Design Practices for Large-Scale AutomationHao Xu
 
Lessons learned from designing a QA Automation for analytics databases (big d...
Lessons learned from designing a QA Automation for analytics databases (big d...Lessons learned from designing a QA Automation for analytics databases (big d...
Lessons learned from designing a QA Automation for analytics databases (big d...Omid Vahdaty
 
HP - Jerome Rolia - Hadoop World 2010
HP - Jerome Rolia - Hadoop World 2010HP - Jerome Rolia - Hadoop World 2010
HP - Jerome Rolia - Hadoop World 2010Cloudera, Inc.
 
Data pipelines from zero to solid
Data pipelines from zero to solidData pipelines from zero to solid
Data pipelines from zero to solidLars Albertsson
 
Semantic Segmentation on Satellite Imagery
Semantic Segmentation on Satellite ImagerySemantic Segmentation on Satellite Imagery
Semantic Segmentation on Satellite ImageryRAHUL BHOJWANI
 
Caching in (DevoxxUK 2013)
Caching in (DevoxxUK 2013)Caching in (DevoxxUK 2013)
Caching in (DevoxxUK 2013)RichardWarburton
 
Set Transfomer: A Framework for Attention-based Permutaion-Invariant Neural N...
Set Transfomer: A Framework for Attention-based Permutaion-Invariant Neural N...Set Transfomer: A Framework for Attention-based Permutaion-Invariant Neural N...
Set Transfomer: A Framework for Attention-based Permutaion-Invariant Neural N...Thien Q. Tran
 
Netflix machine learning
Netflix machine learningNetflix machine learning
Netflix machine learningAmer Ather
 
Exploiting GPU's for Columnar DataFrrames by Kiran Lonikar
Exploiting GPU's for Columnar DataFrrames by Kiran LonikarExploiting GPU's for Columnar DataFrrames by Kiran Lonikar
Exploiting GPU's for Columnar DataFrrames by Kiran LonikarSpark Summit
 
Big data 2.0, deep learning and financial Usecases
Big data 2.0, deep learning and financial UsecasesBig data 2.0, deep learning and financial Usecases
Big data 2.0, deep learning and financial UsecasesArvind Rapaka
 
대용량 데이터 분석을 위한 병렬 Clustering 알고리즘 최적화
대용량 데이터 분석을 위한 병렬 Clustering 알고리즘 최적화대용량 데이터 분석을 위한 병렬 Clustering 알고리즘 최적화
대용량 데이터 분석을 위한 병렬 Clustering 알고리즘 최적화NAVER Engineering
 
Automatic Machine Learning, AutoML
Automatic Machine Learning, AutoMLAutomatic Machine Learning, AutoML
Automatic Machine Learning, AutoMLHimadri Mishra
 
Architectural Optimizations for High Performance and Energy Efficient Smith-W...
Architectural Optimizations for High Performance and Energy Efficient Smith-W...Architectural Optimizations for High Performance and Energy Efficient Smith-W...
Architectural Optimizations for High Performance and Energy Efficient Smith-W...NECST Lab @ Politecnico di Milano
 
From Zero to Streaming Healthcare in Production (Alexander Kouznetsov, Invita...
From Zero to Streaming Healthcare in Production (Alexander Kouznetsov, Invita...From Zero to Streaming Healthcare in Production (Alexander Kouznetsov, Invita...
From Zero to Streaming Healthcare in Production (Alexander Kouznetsov, Invita...confluent
 

Similar to Centernet (20)

A Kaggle Talk
A Kaggle TalkA Kaggle Talk
A Kaggle Talk
 
Volodymyr Lyubinets. One startup's journey of building ML pipelines for text ...
Volodymyr Lyubinets. One startup's journey of building ML pipelines for text ...Volodymyr Lyubinets. One startup's journey of building ML pipelines for text ...
Volodymyr Lyubinets. One startup's journey of building ML pipelines for text ...
 
Software Design Practices for Large-Scale Automation
Software Design Practices for Large-Scale AutomationSoftware Design Practices for Large-Scale Automation
Software Design Practices for Large-Scale Automation
 
Lessons learned from designing a QA Automation for analytics databases (big d...
Lessons learned from designing a QA Automation for analytics databases (big d...Lessons learned from designing a QA Automation for analytics databases (big d...
Lessons learned from designing a QA Automation for analytics databases (big d...
 
HP - Jerome Rolia - Hadoop World 2010
HP - Jerome Rolia - Hadoop World 2010HP - Jerome Rolia - Hadoop World 2010
HP - Jerome Rolia - Hadoop World 2010
 
Data pipelines from zero to solid
Data pipelines from zero to solidData pipelines from zero to solid
Data pipelines from zero to solid
 
Semantic Segmentation on Satellite Imagery
Semantic Segmentation on Satellite ImagerySemantic Segmentation on Satellite Imagery
Semantic Segmentation on Satellite Imagery
 
Caching in (DevoxxUK 2013)
Caching in (DevoxxUK 2013)Caching in (DevoxxUK 2013)
Caching in (DevoxxUK 2013)
 
Set Transfomer: A Framework for Attention-based Permutaion-Invariant Neural N...
Set Transfomer: A Framework for Attention-based Permutaion-Invariant Neural N...Set Transfomer: A Framework for Attention-based Permutaion-Invariant Neural N...
Set Transfomer: A Framework for Attention-based Permutaion-Invariant Neural N...
 
Multicore architectures
Multicore architecturesMulticore architectures
Multicore architectures
 
[ppt]
[ppt][ppt]
[ppt]
 
Netflix machine learning
Netflix machine learningNetflix machine learning
Netflix machine learning
 
Exploiting GPU's for Columnar DataFrrames by Kiran Lonikar
Exploiting GPU's for Columnar DataFrrames by Kiran LonikarExploiting GPU's for Columnar DataFrrames by Kiran Lonikar
Exploiting GPU's for Columnar DataFrrames by Kiran Lonikar
 
Druid
DruidDruid
Druid
 
Kubernetes Workload Rebalancing
Kubernetes Workload RebalancingKubernetes Workload Rebalancing
Kubernetes Workload Rebalancing
 
Big data 2.0, deep learning and financial Usecases
Big data 2.0, deep learning and financial UsecasesBig data 2.0, deep learning and financial Usecases
Big data 2.0, deep learning and financial Usecases
 
대용량 데이터 분석을 위한 병렬 Clustering 알고리즘 최적화
대용량 데이터 분석을 위한 병렬 Clustering 알고리즘 최적화대용량 데이터 분석을 위한 병렬 Clustering 알고리즘 최적화
대용량 데이터 분석을 위한 병렬 Clustering 알고리즘 최적화
 
Automatic Machine Learning, AutoML
Automatic Machine Learning, AutoMLAutomatic Machine Learning, AutoML
Automatic Machine Learning, AutoML
 
Architectural Optimizations for High Performance and Energy Efficient Smith-W...
Architectural Optimizations for High Performance and Energy Efficient Smith-W...Architectural Optimizations for High Performance and Energy Efficient Smith-W...
Architectural Optimizations for High Performance and Energy Efficient Smith-W...
 
From Zero to Streaming Healthcare in Production (Alexander Kouznetsov, Invita...
From Zero to Streaming Healthcare in Production (Alexander Kouznetsov, Invita...From Zero to Streaming Healthcare in Production (Alexander Kouznetsov, Invita...
From Zero to Streaming Healthcare in Production (Alexander Kouznetsov, Invita...
 

More from Arithmer Inc.

コーディネートレコメンド
コーディネートレコメンドコーディネートレコメンド
コーディネートレコメンドArithmer Inc.
 
Arithmerソリューション紹介 流体予測システム
Arithmerソリューション紹介 流体予測システムArithmerソリューション紹介 流体予測システム
Arithmerソリューション紹介 流体予測システムArithmer Inc.
 
Weakly supervised semantic segmentation of 3D point cloud
Weakly supervised semantic segmentation of 3D point cloudWeakly supervised semantic segmentation of 3D point cloud
Weakly supervised semantic segmentation of 3D point cloudArithmer Inc.
 
Arithmer NLP 自然言語処理 ソリューション紹介
Arithmer NLP 自然言語処理 ソリューション紹介Arithmer NLP 自然言語処理 ソリューション紹介
Arithmer NLP 自然言語処理 ソリューション紹介Arithmer Inc.
 
Arithmer Robo Introduction
Arithmer Robo IntroductionArithmer Robo Introduction
Arithmer Robo IntroductionArithmer Inc.
 
Arithmer AIチャットボット
Arithmer AIチャットボットArithmer AIチャットボット
Arithmer AIチャットボットArithmer Inc.
 
Arithmer R3 Introduction
Arithmer R3 Introduction Arithmer R3 Introduction
Arithmer R3 Introduction Arithmer Inc.
 
VIBE: Video Inference for Human Body Pose and Shape Estimation
VIBE: Video Inference for Human Body Pose and Shape EstimationVIBE: Video Inference for Human Body Pose and Shape Estimation
VIBE: Video Inference for Human Body Pose and Shape EstimationArithmer Inc.
 
Arithmer Inspection Introduction
Arithmer Inspection IntroductionArithmer Inspection Introduction
Arithmer Inspection IntroductionArithmer Inc.
 
全力解説!Transformer
全力解説!Transformer全力解説!Transformer
全力解説!TransformerArithmer Inc.
 
Arithmer NLP Introduction
Arithmer NLP IntroductionArithmer NLP Introduction
Arithmer NLP IntroductionArithmer Inc.
 
Introduction of Quantum Annealing and D-Wave Machines
Introduction of Quantum Annealing and D-Wave MachinesIntroduction of Quantum Annealing and D-Wave Machines
Introduction of Quantum Annealing and D-Wave MachinesArithmer Inc.
 
Arithmer OCR Introduction
Arithmer OCR IntroductionArithmer OCR Introduction
Arithmer OCR IntroductionArithmer Inc.
 
Arithmer Dynamics Introduction
Arithmer Dynamics Introduction Arithmer Dynamics Introduction
Arithmer Dynamics Introduction Arithmer Inc.
 
ArithmerDB Introduction
ArithmerDB IntroductionArithmerDB Introduction
ArithmerDB IntroductionArithmer Inc.
 
Summarizing videos with Attention
Summarizing videos with AttentionSummarizing videos with Attention
Summarizing videos with AttentionArithmer Inc.
 
3D human body modeling from RGB images
3D human body modeling from RGB images3D human body modeling from RGB images
3D human body modeling from RGB imagesArithmer Inc.
 

More from Arithmer Inc. (20)

コーディネートレコメンド
コーディネートレコメンドコーディネートレコメンド
コーディネートレコメンド
 
Test for AI model
Test for AI modelTest for AI model
Test for AI model
 
最適化
最適化最適化
最適化
 
Arithmerソリューション紹介 流体予測システム
Arithmerソリューション紹介 流体予測システムArithmerソリューション紹介 流体予測システム
Arithmerソリューション紹介 流体予測システム
 
Weakly supervised semantic segmentation of 3D point cloud
Weakly supervised semantic segmentation of 3D point cloudWeakly supervised semantic segmentation of 3D point cloud
Weakly supervised semantic segmentation of 3D point cloud
 
Arithmer NLP 自然言語処理 ソリューション紹介
Arithmer NLP 自然言語処理 ソリューション紹介Arithmer NLP 自然言語処理 ソリューション紹介
Arithmer NLP 自然言語処理 ソリューション紹介
 
Arithmer Robo Introduction
Arithmer Robo IntroductionArithmer Robo Introduction
Arithmer Robo Introduction
 
Arithmer AIチャットボット
Arithmer AIチャットボットArithmer AIチャットボット
Arithmer AIチャットボット
 
Arithmer R3 Introduction
Arithmer R3 Introduction Arithmer R3 Introduction
Arithmer R3 Introduction
 
VIBE: Video Inference for Human Body Pose and Shape Estimation
VIBE: Video Inference for Human Body Pose and Shape EstimationVIBE: Video Inference for Human Body Pose and Shape Estimation
VIBE: Video Inference for Human Body Pose and Shape Estimation
 
Arithmer Inspection Introduction
Arithmer Inspection IntroductionArithmer Inspection Introduction
Arithmer Inspection Introduction
 
全力解説!Transformer
全力解説!Transformer全力解説!Transformer
全力解説!Transformer
 
Arithmer NLP Introduction
Arithmer NLP IntroductionArithmer NLP Introduction
Arithmer NLP Introduction
 
Introduction of Quantum Annealing and D-Wave Machines
Introduction of Quantum Annealing and D-Wave MachinesIntroduction of Quantum Annealing and D-Wave Machines
Introduction of Quantum Annealing and D-Wave Machines
 
Arithmer OCR Introduction
Arithmer OCR IntroductionArithmer OCR Introduction
Arithmer OCR Introduction
 
Arithmer Dynamics Introduction
Arithmer Dynamics Introduction Arithmer Dynamics Introduction
Arithmer Dynamics Introduction
 
ArithmerDB Introduction
ArithmerDB IntroductionArithmerDB Introduction
ArithmerDB Introduction
 
Summarizing videos with Attention
Summarizing videos with AttentionSummarizing videos with Attention
Summarizing videos with Attention
 
3D human body modeling from RGB images
3D human body modeling from RGB images3D human body modeling from RGB images
3D human body modeling from RGB images
 
YOLACT
YOLACTYOLACT
YOLACT
 

Recently uploaded

Laying the Data Foundations for Artificial Intelligence!
Laying the Data Foundations for Artificial Intelligence!Laying the Data Foundations for Artificial Intelligence!
Laying the Data Foundations for Artificial Intelligence!Memoori
 
Generative AI - Gitex v1Generative AI - Gitex v1.pptx
Generative AI - Gitex v1Generative AI - Gitex v1.pptxGenerative AI - Gitex v1Generative AI - Gitex v1.pptx
Generative AI - Gitex v1Generative AI - Gitex v1.pptxfnnc6jmgwh
 
Potential of AI (Generative AI) in Business: Learnings and Insights
Potential of AI (Generative AI) in Business: Learnings and InsightsPotential of AI (Generative AI) in Business: Learnings and Insights
Potential of AI (Generative AI) in Business: Learnings and InsightsRavi Sanghani
 
Connecting the Dots for Information Discovery.pdf
Connecting the Dots for Information Discovery.pdfConnecting the Dots for Information Discovery.pdf
Connecting the Dots for Information Discovery.pdfNeo4j
 
Design pattern talk by Kaya Weers - 2024 (v2)
Design pattern talk by Kaya Weers - 2024 (v2)Design pattern talk by Kaya Weers - 2024 (v2)
Design pattern talk by Kaya Weers - 2024 (v2)Kaya Weers
 
Long journey of Ruby standard library at RubyConf AU 2024
Long journey of Ruby standard library at RubyConf AU 2024Long journey of Ruby standard library at RubyConf AU 2024
Long journey of Ruby standard library at RubyConf AU 2024Hiroshi SHIBATA
 
Kuma Meshes Part I - The basics - A tutorial
Kuma Meshes Part I - The basics - A tutorialKuma Meshes Part I - The basics - A tutorial
Kuma Meshes Part I - The basics - A tutorialJoão Esperancinha
 
UiPath Community: Communication Mining from Zero to Hero
UiPath Community: Communication Mining from Zero to HeroUiPath Community: Communication Mining from Zero to Hero
UiPath Community: Communication Mining from Zero to HeroUiPathCommunity
 
WomenInAutomation2024: AI and Automation for eveyone
WomenInAutomation2024: AI and Automation for eveyoneWomenInAutomation2024: AI and Automation for eveyone
WomenInAutomation2024: AI and Automation for eveyoneUiPathCommunity
 
Accelerating Enterprise Software Engineering with Platformless
Accelerating Enterprise Software Engineering with PlatformlessAccelerating Enterprise Software Engineering with Platformless
Accelerating Enterprise Software Engineering with PlatformlessWSO2
 
QMMS Lesson 2 - Using MS Excel Formula.pdf
QMMS Lesson 2 - Using MS Excel Formula.pdfQMMS Lesson 2 - Using MS Excel Formula.pdf
QMMS Lesson 2 - Using MS Excel Formula.pdfROWELL MARQUINA
 
The Future Roadmap for the Composable Data Stack - Wes McKinney - Data Counci...
The Future Roadmap for the Composable Data Stack - Wes McKinney - Data Counci...The Future Roadmap for the Composable Data Stack - Wes McKinney - Data Counci...
The Future Roadmap for the Composable Data Stack - Wes McKinney - Data Counci...Wes McKinney
 
Decarbonising Buildings: Making a net-zero built environment a reality
Decarbonising Buildings: Making a net-zero built environment a realityDecarbonising Buildings: Making a net-zero built environment a reality
Decarbonising Buildings: Making a net-zero built environment a realityIES VE
 
Modern Roaming for Notes and Nomad – Cheaper Faster Better Stronger
Modern Roaming for Notes and Nomad – Cheaper Faster Better StrongerModern Roaming for Notes and Nomad – Cheaper Faster Better Stronger
Modern Roaming for Notes and Nomad – Cheaper Faster Better Strongerpanagenda
 
All These Sophisticated Attacks, Can We Really Detect Them - PDF
All These Sophisticated Attacks, Can We Really Detect Them - PDFAll These Sophisticated Attacks, Can We Really Detect Them - PDF
All These Sophisticated Attacks, Can We Really Detect Them - PDFMichael Gough
 
Testing tools and AI - ideas what to try with some tool examples
Testing tools and AI - ideas what to try with some tool examplesTesting tools and AI - ideas what to try with some tool examples
Testing tools and AI - ideas what to try with some tool examplesKari Kakkonen
 
Deliver Latency Free Customer Experience
Deliver Latency Free Customer ExperienceDeliver Latency Free Customer Experience
Deliver Latency Free Customer ExperienceOpsTree solutions
 
Generative Artificial Intelligence: How generative AI works.pdf
Generative Artificial Intelligence: How generative AI works.pdfGenerative Artificial Intelligence: How generative AI works.pdf
Generative Artificial Intelligence: How generative AI works.pdfIngrid Airi González
 
[Webinar] SpiraTest - Setting New Standards in Quality Assurance
[Webinar] SpiraTest - Setting New Standards in Quality Assurance[Webinar] SpiraTest - Setting New Standards in Quality Assurance
[Webinar] SpiraTest - Setting New Standards in Quality AssuranceInflectra
 
Landscape Catalogue 2024 Australia-1.pdf
Landscape Catalogue 2024 Australia-1.pdfLandscape Catalogue 2024 Australia-1.pdf
Landscape Catalogue 2024 Australia-1.pdfAarwolf Industries LLC
 

Recently uploaded (20)

Laying the Data Foundations for Artificial Intelligence!
Laying the Data Foundations for Artificial Intelligence!Laying the Data Foundations for Artificial Intelligence!
Laying the Data Foundations for Artificial Intelligence!
 
Generative AI - Gitex v1Generative AI - Gitex v1.pptx
Generative AI - Gitex v1Generative AI - Gitex v1.pptxGenerative AI - Gitex v1Generative AI - Gitex v1.pptx
Generative AI - Gitex v1Generative AI - Gitex v1.pptx
 
Potential of AI (Generative AI) in Business: Learnings and Insights
Potential of AI (Generative AI) in Business: Learnings and InsightsPotential of AI (Generative AI) in Business: Learnings and Insights
Potential of AI (Generative AI) in Business: Learnings and Insights
 
Connecting the Dots for Information Discovery.pdf
Connecting the Dots for Information Discovery.pdfConnecting the Dots for Information Discovery.pdf
Connecting the Dots for Information Discovery.pdf
 
Design pattern talk by Kaya Weers - 2024 (v2)
Design pattern talk by Kaya Weers - 2024 (v2)Design pattern talk by Kaya Weers - 2024 (v2)
Design pattern talk by Kaya Weers - 2024 (v2)
 
Long journey of Ruby standard library at RubyConf AU 2024
Long journey of Ruby standard library at RubyConf AU 2024Long journey of Ruby standard library at RubyConf AU 2024
Long journey of Ruby standard library at RubyConf AU 2024
 
Kuma Meshes Part I - The basics - A tutorial
Kuma Meshes Part I - The basics - A tutorialKuma Meshes Part I - The basics - A tutorial
Kuma Meshes Part I - The basics - A tutorial
 
UiPath Community: Communication Mining from Zero to Hero
UiPath Community: Communication Mining from Zero to HeroUiPath Community: Communication Mining from Zero to Hero
UiPath Community: Communication Mining from Zero to Hero
 
WomenInAutomation2024: AI and Automation for eveyone
WomenInAutomation2024: AI and Automation for eveyoneWomenInAutomation2024: AI and Automation for eveyone
WomenInAutomation2024: AI and Automation for eveyone
 
Accelerating Enterprise Software Engineering with Platformless
Accelerating Enterprise Software Engineering with PlatformlessAccelerating Enterprise Software Engineering with Platformless
Accelerating Enterprise Software Engineering with Platformless
 
QMMS Lesson 2 - Using MS Excel Formula.pdf
QMMS Lesson 2 - Using MS Excel Formula.pdfQMMS Lesson 2 - Using MS Excel Formula.pdf
QMMS Lesson 2 - Using MS Excel Formula.pdf
 
The Future Roadmap for the Composable Data Stack - Wes McKinney - Data Counci...
The Future Roadmap for the Composable Data Stack - Wes McKinney - Data Counci...The Future Roadmap for the Composable Data Stack - Wes McKinney - Data Counci...
The Future Roadmap for the Composable Data Stack - Wes McKinney - Data Counci...
 
Decarbonising Buildings: Making a net-zero built environment a reality
Decarbonising Buildings: Making a net-zero built environment a realityDecarbonising Buildings: Making a net-zero built environment a reality
Decarbonising Buildings: Making a net-zero built environment a reality
 
Modern Roaming for Notes and Nomad – Cheaper Faster Better Stronger
Modern Roaming for Notes and Nomad – Cheaper Faster Better StrongerModern Roaming for Notes and Nomad – Cheaper Faster Better Stronger
Modern Roaming for Notes and Nomad – Cheaper Faster Better Stronger
 
All These Sophisticated Attacks, Can We Really Detect Them - PDF
All These Sophisticated Attacks, Can We Really Detect Them - PDFAll These Sophisticated Attacks, Can We Really Detect Them - PDF
All These Sophisticated Attacks, Can We Really Detect Them - PDF
 
Testing tools and AI - ideas what to try with some tool examples
Testing tools and AI - ideas what to try with some tool examplesTesting tools and AI - ideas what to try with some tool examples
Testing tools and AI - ideas what to try with some tool examples
 
Deliver Latency Free Customer Experience
Deliver Latency Free Customer ExperienceDeliver Latency Free Customer Experience
Deliver Latency Free Customer Experience
 
Generative Artificial Intelligence: How generative AI works.pdf
Generative Artificial Intelligence: How generative AI works.pdfGenerative Artificial Intelligence: How generative AI works.pdf
Generative Artificial Intelligence: How generative AI works.pdf
 
[Webinar] SpiraTest - Setting New Standards in Quality Assurance
[Webinar] SpiraTest - Setting New Standards in Quality Assurance[Webinar] SpiraTest - Setting New Standards in Quality Assurance
[Webinar] SpiraTest - Setting New Standards in Quality Assurance
 
Landscape Catalogue 2024 Australia-1.pdf
Landscape Catalogue 2024 Australia-1.pdfLandscape Catalogue 2024 Australia-1.pdf
Landscape Catalogue 2024 Australia-1.pdf
 

Centernet

  • 1. 2020/08/28 Centernet: Object as Points Arithmer Dynamics Div. Christian Saravia
  • 2. Self-Introduction Christian Martin Saravia Hernandez ● Graduate School ○ Universidad Peruana de Ciencias Aplicadas ○ Bs. Software Engineering ● Former Job ○ Ficha Inc ■ Algorithm Development Engineer ● Research & application of deep learning in computer vision problems ● Current Job ○ Application of machine learning / deep learning to computer vision problems ■ Object detection ■ Object classification
  • 3. Purpose Purpose of this material ● Understand an anchor free approach object detection algorithm
  • 4. Agenda Agenda ● Current object detection approaches ● Centernet approach ● Object as Points ● Training ○ Keypoint heatmap ○ Local offset ○ Size prediction ○ Loss function ● Network Architecture ○ DLA ○ Modified DLA ● Inference ● Results
  • 5. Background Current approaches ● Object detections model (such as Yolo, SSD, etc.) rely on the usage of anchor boxes ● Anchor boxes are not completely optimal: ○ Wasteful: SSD300 does 8732 detections per class, and yolo448 does 98 detections per class, which means that most of the box are discarded ○ Inefficient: We have to process all the boxes (even we will discard them later), which comes with more processing time ○ Require post processing: like non-max suppression algorithm ○ Fixed: SSD requires fixed scale and steps of boxes, while yolov3 fixes the size of the anchors per detection level
  • 6. Centernet Centernet approach ● End-to-end differentiable solution ● Relies on keypoint estimation to find the center points and regress all other object properties(such as size) ● As a result, the model is simpler, faster and more accurate than bounding-box based detectors Speed-accuracy trade-off on COCO dataset
  • 10. Keypoint Heatmap * If two gaussians of the same class overlap, take the element-wise maximum
  • 17. Network Architecture ● Authors experiment with different backbone architectures, obtaining different results: ● The backbone that produces best speed/accuracy tradeoff is DLA-34 (modified by authors) Results without test augmentation(N.A.), flip testing(F), and multi-scale augmentation(MS). HW: Intel Core i7-8086k CPU, Titan Xp GPU
  • 18. Network Architecture ● Deep Layer Aggregation: Iterative deep aggregation Upsample 2x Aggregation node: Conv2d(1x1) Batch Normalization Relu Stage node: Conv2d(3x3) Batch Normalization Relu Conv2d(3x3) Batch Normalization Relu
  • 19. Network Architecture ● Modified Deep Layer Aggregation: Iterative deep aggregation Upsample 2x Deformable convolution Aggregation node: Conv2d(1x1) Batch Normalization Relu Stage node: Conv2d(3x3) Batch Normalization Relu Conv2d(3x3) Batch Normalization Relu
  • 21. Results ● Object detection ● Human pose estimation
  • 22. Results Experimental results on Pascal VOC 2007 test. The results are shown in mAP@0.5. Flip test is used for Centernet. The FPSs for other methods are copied from the original publications COCO test-dev. Frame-per-second (FPS) were measured on the same machine whenever possible. Italic FPS highlight the cases, where the performance measure was copied from the original publication
  • 23. 23