SlideShare a Scribd company logo
/18
EventNet: Asynchronous Recursive Event Processing
Study 2020/09/18
2011136 Ryuta Shitomi
1
Yusuke Sekikawa, Kosuke Hara, and Hideo Saito, CVPR2019
/18
The event camera
2
 High dynamic range
 Do not suffer from motion blur
 No latency
 Recently, frame-based approach have achieved impressive results in scenarios.
/18
The event data from event-based camera
3
 When the new event occurred:
𝑒𝑖 = 𝑥𝑖, 𝑦𝑖, 𝑝𝑖, 𝑡 − 𝑡′
𝑡′
: time of the last event
 Asynchronous
 Sparse
 Many events (1 mega event per sec)
↓
Need a new learning approach to differ from frame-based
approach
/18
Frame-based approach for an event streams
4
 Integration 𝛥𝑡 to create a frame and feed a neural
net.
 Disadvantages:
Cannot use of the sparseness of the data
Redundant computation
A. I. Maqueda, A. Loquercio, G. Gallego, N. N. Garcia, and D. Scaramuzza. Event-based vision meets deep
learning on steering prediction for self-driving cars. CoRR, abs/1804.01310, 2018.
𝜟𝒕
/18
Purpose and Contribution
5
 Develop a novel neural network:
can process an extremely high-rate and variable length.
for real-time processing of an asynchronous event stream.
to output recursively using temporal information.
𝜏′
𝜏
𝑡
𝑝𝑜𝑙𝑎𝑟𝑖𝑡𝑦
/18
Problem Statement
6
 Consider a sequence of events within a 𝜏 ms interval based on the latest time stamp 𝑡𝑗 of
the event:
𝒆𝑗 = 𝑒𝑖 𝑖 = 𝑗 − 𝑛 𝑗 + 1, … , 𝑗}
𝑒𝑖 = 𝑥𝑖, 𝑦𝑖, 𝑝𝑖, Δ𝑡𝑗,𝑖 Δ𝑡𝑗,𝑖 = 𝑡𝑗 − 𝑡𝑖
 Consider a neural network 𝑓 to estimate 𝑦𝑗 given event stream 𝒆𝒋
𝑦𝑗 = 𝑓 𝒆𝑗 𝑦𝑗: Task dependent
𝜏
𝑡
𝒆𝒋
𝜏
𝒆𝒋+𝟏
/18
Problem Statement
7
 To realize a trainable neural network 𝑓 that satisfies following conditions:
1. End-to-End Trainable
2. Event-Wise Processing
3. Efficient Recursive Processing
4. Local Permutation Invariance
𝜏
𝑡
𝒆𝒋
𝜏
𝒆𝒋+𝟏
/18
Using PointNet architecture
8
 PointNet solves the problem by approximating the function 𝑓:
𝑦𝑗 = 𝑓 𝒆𝑗 = 𝑔 max ℎ 𝑒𝑗−𝑛 𝑗 +1 , … , ℎ 𝑒𝑗
where ℎ: ℝ4
→ ℝ 𝐾
, 𝑔: ℝ 𝐾
→ ℝ
 Because of the max(∙) , the permutation of events does not change the output 𝑦𝑗.
MLP
 Problem:
A huge amount of computation of ℎ ∙ and max ∙ .
Should run on the event rate. (could be more than 1MEPS)
When a new j + 1 -th event arrives, Δtj,i = tj − ti changes to tj+1 − ti .
Many previous event has already been processed by h(∙).
/18
EventNet
9
 Introduce the ”Temporal Coding” and “Recursive Processing” to overcome the difficulty.
The Δ𝑡 only changes as the new event is being received.
Split the ℎ(𝑒):
ℎ 𝑒𝑖 = 𝑐(ℎ 𝑒𝑖
−
, Δ𝑡𝑗,𝑖), 𝑒−
: (𝑥, 𝑦, 𝑝)
𝑐 ∙ :temporal coding function, ℎ: ℝ3
→ ℂ 𝐾
𝑦𝑗 become:
𝑓 𝒆𝑗 = 𝑔 max 𝑐 𝑧𝑗−𝑛 𝑗 +1, Δ𝑡𝑗,𝑗−𝑛 𝑗 +1 , … , 𝑐 𝑧𝑗, 0
where 𝑧𝑖 = ℎ 𝑒−
∈ ℂ 𝐾
𝑐 ∙ , max ∙ need to be computed for all events.
Temporal Coding
/18
EventNet
10
• Assume the norm of each element of 𝑧𝑖 is less than 1 (by tanh).
• max ∙ , 𝑐 ∙ need to be recursive.
Example)
𝑠𝑗 = max 𝑠𝑗−1, 𝑐 𝑧𝑗, 0
𝑠𝑗+1 = max 𝑠𝑗, 𝑐 𝑧𝑗+1, 0
• Propose temporal coding function to satisfy this condition:
𝑎𝑗,𝑖 = 𝑐 𝑧𝑖, Δ𝑡𝑗,𝑖 = 𝑧𝑖 −
Δ𝑡𝑗,𝑖
𝜏
+
exp −i
2𝜋Δ𝑡𝑗,𝑖
𝜏
Recursive Processing
Encode the elapsed time
/18
Temporal Coding Function
11
𝑎𝑗,𝑖 = 𝑐 𝑧𝑖, Δ𝑡𝑗,𝑖 = 𝑧𝑖 −
Δ𝑡𝑗,𝑖
𝜏
+
exp −i
2𝜋Δ𝑡𝑗,𝑖
𝜏
When the new event is received:
𝑎𝑗+1,𝑖 = 𝑐 𝑧𝑖, Δ𝑡𝑗+1,𝑖 = 𝑧𝑖 −
Δ𝑡𝑗,𝑖
𝜏
−
𝑡𝑗+1 − 𝑡𝑗
𝜏
+
exp −i
2𝜋(Δ𝑡𝑗,𝑖 − 𝑡𝑗+1 + 𝑡𝑗
𝜏
Recursive Processing
𝑡
|𝑎 𝑗,𝑖|
𝑒𝑗
New event is received
𝑡
|𝑎 𝑗,𝑖|
𝑒𝑗 𝑒𝑗+1
− 𝑡𝑗+1 − 𝑡𝑗
/18
Temporal Coding Function
12
𝑡
|𝑎 𝑗,𝑖|
𝑒𝑗
New event is received
𝑡
|𝑎 𝑗+1,𝑖|
𝑒𝑗 𝑒𝑗+1
− 𝑡𝑗+1 − 𝑡𝑗
𝑎𝑗,𝑗−2 = max 𝑎𝑗,𝑗−5, … , 𝑎𝑗,𝑗
 when you new event is received.
𝑎𝑗+1,𝑗−2 = max 𝑐 𝑎𝑗,𝑗−2, 𝑡𝑗+1 − 𝑡𝑗 , 𝑎𝑗+1,𝑗+1
 Therefore, 𝑦𝑗+1:
𝑦𝑗+1 = 𝑓 𝒆𝒋+𝟏 = 𝑔 max 𝑐 𝑠𝑗, 𝑡𝑗+1 − 𝑡𝑗 , ℎ 𝑒𝑗+1
−
𝑒𝑗−2
𝑒𝑗−5
|𝑎 𝑗,𝑗−2|
/18
Overview of EventNet
13
 First event is arrived:
𝒆0 = {𝑒0}
𝑧0 = ℎ(𝑒0
−
)
𝑦0 = 𝑓 𝒆0 = 𝑔 max 𝑐 𝑧0, 0
𝑡
𝑐(𝑧0, 0)
𝑗 = 0
𝑗 = 1
 Next event is arrived:
𝒆1 = {𝑒0, 𝑒1}
𝑠1 = max(𝑐 𝑧0, 𝑡1 − 𝑡0 , ℎ(𝑒1
−
)
𝑦1 = 𝑓 𝒆1 = 𝑔(𝑠1)
 Next event is arrived:
𝒆2 = {𝑒0, 𝑒1, 𝑒2}
𝑠2 = max 𝑐 𝑠1, 𝑡2 − 𝑡1 , ℎ 𝑒2
−
𝑦3 = 𝑔 𝑠2
𝑗 = 2
/18
EventNet Architecture
14
 Create Look-up table(LUT) for decreasing inference time.
(There are only 𝑊 × 𝐻 × 2 patterns in inputs 𝑒−
)
ℎ(∙)
𝑔(∙)
/18
Experiments
15
 ETHTED+ dataset for target motion estimation and semantic segmentation
 MVSEC dataset for ego-motion estimation
/18
Results for target motion estimation
16
/18
Results for ego-motion
17
 Estimated ego motions are shown as yellow.
 Estimated at the rate of 1000 Hz.
/18
Compare the computation time [μs]
18
 Using newly received event, PointNet need to process the all events again.

More Related Content

What's hot

3D Analyst - Cut and Fill
3D Analyst - Cut and Fill3D Analyst - Cut and Fill
3D Analyst - Cut and Fill
Hartanto Sanjaya
 
Toy Model Overview
Toy Model OverviewToy Model Overview
Toy Model Overview
Brandon McKinzie
 
Global Grid of Grapes
Global Grid of GrapesGlobal Grid of Grapes
Global Grid of Grapes
Derek Groen
 
How to Get the Most Out of LiDAR Data
How to Get the Most Out of LiDAR DataHow to Get the Most Out of LiDAR Data
How to Get the Most Out of LiDAR Data
Safe Software
 
Lesson9
Lesson9Lesson9
Quantum Computer Overview
Quantum Computer OverviewQuantum Computer Overview
Quantum Computer Overview
Yuichiro MInato
 
Superefficient Monte Carlo Simulations
Superefficient Monte Carlo SimulationsSuperefficient Monte Carlo Simulations
Superefficient Monte Carlo Simulations
Cheng-An Yang
 
Lecture15 fsm i_ic
Lecture15 fsm i_icLecture15 fsm i_ic
Lecture15 fsm i_ic
Konsta Anttila
 
Ulrik De Bie - Newtec@ Virtual Wall
Ulrik De Bie - Newtec@ Virtual WallUlrik De Bie - Newtec@ Virtual Wall
Ulrik De Bie - Newtec@ Virtual Wall
imec.archive
 
Energy of Some Simple Graphs: MATLAB Approach
Energy of Some Simple Graphs: MATLAB ApproachEnergy of Some Simple Graphs: MATLAB Approach
Energy of Some Simple Graphs: MATLAB Approach
IJCSIS Research Publications
 
Internship
InternshipInternship
Internship
Ali Akbari
 
2 introduction to computer programming
2 introduction to computer programming2 introduction to computer programming
2 introduction to computer programming
Nataly Medina
 

What's hot (12)

3D Analyst - Cut and Fill
3D Analyst - Cut and Fill3D Analyst - Cut and Fill
3D Analyst - Cut and Fill
 
Toy Model Overview
Toy Model OverviewToy Model Overview
Toy Model Overview
 
Global Grid of Grapes
Global Grid of GrapesGlobal Grid of Grapes
Global Grid of Grapes
 
How to Get the Most Out of LiDAR Data
How to Get the Most Out of LiDAR DataHow to Get the Most Out of LiDAR Data
How to Get the Most Out of LiDAR Data
 
Lesson9
Lesson9Lesson9
Lesson9
 
Quantum Computer Overview
Quantum Computer OverviewQuantum Computer Overview
Quantum Computer Overview
 
Superefficient Monte Carlo Simulations
Superefficient Monte Carlo SimulationsSuperefficient Monte Carlo Simulations
Superefficient Monte Carlo Simulations
 
Lecture15 fsm i_ic
Lecture15 fsm i_icLecture15 fsm i_ic
Lecture15 fsm i_ic
 
Ulrik De Bie - Newtec@ Virtual Wall
Ulrik De Bie - Newtec@ Virtual WallUlrik De Bie - Newtec@ Virtual Wall
Ulrik De Bie - Newtec@ Virtual Wall
 
Energy of Some Simple Graphs: MATLAB Approach
Energy of Some Simple Graphs: MATLAB ApproachEnergy of Some Simple Graphs: MATLAB Approach
Energy of Some Simple Graphs: MATLAB Approach
 
Internship
InternshipInternship
Internship
 
2 introduction to computer programming
2 introduction to computer programming2 introduction to computer programming
2 introduction to computer programming
 

Similar to eventnet asynchronous recursive event processing

Distributed solution of stochastic optimal control problem on GPUs
Distributed solution of stochastic optimal control problem on GPUsDistributed solution of stochastic optimal control problem on GPUs
Distributed solution of stochastic optimal control problem on GPUs
Pantelis Sopasakis
 
Complexity Analysis
Complexity Analysis Complexity Analysis
Complexity Analysis
Shaista Qadir
 
Martin Takac - “Solving Large-Scale Machine Learning Problems in a Distribute...
Martin Takac - “Solving Large-Scale Machine Learning Problems in a Distribute...Martin Takac - “Solving Large-Scale Machine Learning Problems in a Distribute...
Martin Takac - “Solving Large-Scale Machine Learning Problems in a Distribute...
diannepatricia
 
Lecture2a algorithm
Lecture2a algorithmLecture2a algorithm
Lecture2a algorithm
mbadhi barnabas
 
MUMS: Transition & SPUQ Workshop - Practical Bayesian Optimization for Urban ...
MUMS: Transition & SPUQ Workshop - Practical Bayesian Optimization for Urban ...MUMS: Transition & SPUQ Workshop - Practical Bayesian Optimization for Urban ...
MUMS: Transition & SPUQ Workshop - Practical Bayesian Optimization for Urban ...
The Statistical and Applied Mathematical Sciences Institute
 
DSD-INT 2018 Algorithmic Differentiation - Markus
DSD-INT 2018 Algorithmic Differentiation - MarkusDSD-INT 2018 Algorithmic Differentiation - Markus
DSD-INT 2018 Algorithmic Differentiation - Markus
Deltares
 
AINL 2016: Strijov
AINL 2016: StrijovAINL 2016: Strijov
AINL 2016: Strijov
Lidia Pivovarova
 
Mining event streams with BeepBeep 3
Mining event streams with BeepBeep 3Mining event streams with BeepBeep 3
Mining event streams with BeepBeep 3
Sylvain Hallé
 
4.3 real time game physics
4.3 real time game physics4.3 real time game physics
4.3 real time game physics
Sayed Ahmed
 
CEM Workshop Lectures (8/11): Method of moments
CEM Workshop Lectures (8/11):  Method of momentsCEM Workshop Lectures (8/11):  Method of moments
CEM Workshop Lectures (8/11): Method of moments
Abhishek Jain
 
Algorithm Analysis.pdf
Algorithm Analysis.pdfAlgorithm Analysis.pdf
Algorithm Analysis.pdf
MemMem25
 
Power systemsilablri
Power systemsilablriPower systemsilablri
Power systemsilablri
Olivier Teytaud
 
Introduction to PyTorch
Introduction to PyTorchIntroduction to PyTorch
Introduction to PyTorch
Jun Young Park
 
Particle filter
Particle filterParticle filter
Particle filter
Mohammad Reza Jabbari
 
Augmented Arithmetic Operations Proposed for IEEE-754 2018
Augmented Arithmetic Operations Proposed for IEEE-754 2018Augmented Arithmetic Operations Proposed for IEEE-754 2018
Augmented Arithmetic Operations Proposed for IEEE-754 2018
Jason Riedy
 
Deep time-to-failure: predicting failures, churns and customer lifetime with ...
Deep time-to-failure: predicting failures, churns and customer lifetime with ...Deep time-to-failure: predicting failures, churns and customer lifetime with ...
Deep time-to-failure: predicting failures, churns and customer lifetime with ...
Data Science Milan
 
Type and proof structures for concurrency
Type and proof structures for concurrencyType and proof structures for concurrency
Type and proof structures for concurrency
Facultad de Informática UCM
 
QMC: Transition Workshop - Importance Sampling the Union of Rare Events with ...
QMC: Transition Workshop - Importance Sampling the Union of Rare Events with ...QMC: Transition Workshop - Importance Sampling the Union of Rare Events with ...
QMC: Transition Workshop - Importance Sampling the Union of Rare Events with ...
The Statistical and Applied Mathematical Sciences Institute
 
Ke yi small summaries for big data
Ke yi small summaries for big dataKe yi small summaries for big data
Ke yi small summaries for big data
jins0618
 
19 - Neural Networks I.pptx
19 - Neural Networks I.pptx19 - Neural Networks I.pptx
19 - Neural Networks I.pptx
EmanAl15
 

Similar to eventnet asynchronous recursive event processing (20)

Distributed solution of stochastic optimal control problem on GPUs
Distributed solution of stochastic optimal control problem on GPUsDistributed solution of stochastic optimal control problem on GPUs
Distributed solution of stochastic optimal control problem on GPUs
 
Complexity Analysis
Complexity Analysis Complexity Analysis
Complexity Analysis
 
Martin Takac - “Solving Large-Scale Machine Learning Problems in a Distribute...
Martin Takac - “Solving Large-Scale Machine Learning Problems in a Distribute...Martin Takac - “Solving Large-Scale Machine Learning Problems in a Distribute...
Martin Takac - “Solving Large-Scale Machine Learning Problems in a Distribute...
 
Lecture2a algorithm
Lecture2a algorithmLecture2a algorithm
Lecture2a algorithm
 
MUMS: Transition & SPUQ Workshop - Practical Bayesian Optimization for Urban ...
MUMS: Transition & SPUQ Workshop - Practical Bayesian Optimization for Urban ...MUMS: Transition & SPUQ Workshop - Practical Bayesian Optimization for Urban ...
MUMS: Transition & SPUQ Workshop - Practical Bayesian Optimization for Urban ...
 
DSD-INT 2018 Algorithmic Differentiation - Markus
DSD-INT 2018 Algorithmic Differentiation - MarkusDSD-INT 2018 Algorithmic Differentiation - Markus
DSD-INT 2018 Algorithmic Differentiation - Markus
 
AINL 2016: Strijov
AINL 2016: StrijovAINL 2016: Strijov
AINL 2016: Strijov
 
Mining event streams with BeepBeep 3
Mining event streams with BeepBeep 3Mining event streams with BeepBeep 3
Mining event streams with BeepBeep 3
 
4.3 real time game physics
4.3 real time game physics4.3 real time game physics
4.3 real time game physics
 
CEM Workshop Lectures (8/11): Method of moments
CEM Workshop Lectures (8/11):  Method of momentsCEM Workshop Lectures (8/11):  Method of moments
CEM Workshop Lectures (8/11): Method of moments
 
Algorithm Analysis.pdf
Algorithm Analysis.pdfAlgorithm Analysis.pdf
Algorithm Analysis.pdf
 
Power systemsilablri
Power systemsilablriPower systemsilablri
Power systemsilablri
 
Introduction to PyTorch
Introduction to PyTorchIntroduction to PyTorch
Introduction to PyTorch
 
Particle filter
Particle filterParticle filter
Particle filter
 
Augmented Arithmetic Operations Proposed for IEEE-754 2018
Augmented Arithmetic Operations Proposed for IEEE-754 2018Augmented Arithmetic Operations Proposed for IEEE-754 2018
Augmented Arithmetic Operations Proposed for IEEE-754 2018
 
Deep time-to-failure: predicting failures, churns and customer lifetime with ...
Deep time-to-failure: predicting failures, churns and customer lifetime with ...Deep time-to-failure: predicting failures, churns and customer lifetime with ...
Deep time-to-failure: predicting failures, churns and customer lifetime with ...
 
Type and proof structures for concurrency
Type and proof structures for concurrencyType and proof structures for concurrency
Type and proof structures for concurrency
 
QMC: Transition Workshop - Importance Sampling the Union of Rare Events with ...
QMC: Transition Workshop - Importance Sampling the Union of Rare Events with ...QMC: Transition Workshop - Importance Sampling the Union of Rare Events with ...
QMC: Transition Workshop - Importance Sampling the Union of Rare Events with ...
 
Ke yi small summaries for big data
Ke yi small summaries for big dataKe yi small summaries for big data
Ke yi small summaries for big data
 
19 - Neural Networks I.pptx
19 - Neural Networks I.pptx19 - Neural Networks I.pptx
19 - Neural Networks I.pptx
 

Recently uploaded

AI assisted telemedicine KIOSK for Rural India.pptx
AI assisted telemedicine KIOSK for Rural India.pptxAI assisted telemedicine KIOSK for Rural India.pptx
AI assisted telemedicine KIOSK for Rural India.pptx
architagupta876
 
Rainfall intensity duration frequency curve statistical analysis and modeling...
Rainfall intensity duration frequency curve statistical analysis and modeling...Rainfall intensity duration frequency curve statistical analysis and modeling...
Rainfall intensity duration frequency curve statistical analysis and modeling...
bijceesjournal
 
IEEE Aerospace and Electronic Systems Society as a Graduate Student Member
IEEE Aerospace and Electronic Systems Society as a Graduate Student MemberIEEE Aerospace and Electronic Systems Society as a Graduate Student Member
IEEE Aerospace and Electronic Systems Society as a Graduate Student Member
VICTOR MAESTRE RAMIREZ
 
学校原版美国波士顿大学毕业证学历学位证书原版一模一样
学校原版美国波士顿大学毕业证学历学位证书原版一模一样学校原版美国波士顿大学毕业证学历学位证书原版一模一样
学校原版美国波士顿大学毕业证学历学位证书原版一模一样
171ticu
 
Data Driven Maintenance | UReason Webinar
Data Driven Maintenance | UReason WebinarData Driven Maintenance | UReason Webinar
Data Driven Maintenance | UReason Webinar
UReason
 
Curve Fitting in Numerical Methods Regression
Curve Fitting in Numerical Methods RegressionCurve Fitting in Numerical Methods Regression
Curve Fitting in Numerical Methods Regression
Nada Hikmah
 
Optimizing Gradle Builds - Gradle DPE Tour Berlin 2024
Optimizing Gradle Builds - Gradle DPE Tour Berlin 2024Optimizing Gradle Builds - Gradle DPE Tour Berlin 2024
Optimizing Gradle Builds - Gradle DPE Tour Berlin 2024
Sinan KOZAK
 
Unit-III-ELECTROCHEMICAL STORAGE DEVICES.ppt
Unit-III-ELECTROCHEMICAL STORAGE DEVICES.pptUnit-III-ELECTROCHEMICAL STORAGE DEVICES.ppt
Unit-III-ELECTROCHEMICAL STORAGE DEVICES.ppt
KrishnaveniKrishnara1
 
Certificates - Mahmoud Mohamed Moursi Ahmed
Certificates - Mahmoud Mohamed Moursi AhmedCertificates - Mahmoud Mohamed Moursi Ahmed
Certificates - Mahmoud Mohamed Moursi Ahmed
Mahmoud Morsy
 
artificial intelligence and data science contents.pptx
artificial intelligence and data science contents.pptxartificial intelligence and data science contents.pptx
artificial intelligence and data science contents.pptx
GauravCar
 
An improved modulation technique suitable for a three level flying capacitor ...
An improved modulation technique suitable for a three level flying capacitor ...An improved modulation technique suitable for a three level flying capacitor ...
An improved modulation technique suitable for a three level flying capacitor ...
IJECEIAES
 
ITSM Integration with MuleSoft.pptx
ITSM  Integration with MuleSoft.pptxITSM  Integration with MuleSoft.pptx
ITSM Integration with MuleSoft.pptx
VANDANAMOHANGOUDA
 
LLM Fine Tuning with QLoRA Cassandra Lunch 4, presented by Anant
LLM Fine Tuning with QLoRA Cassandra Lunch 4, presented by AnantLLM Fine Tuning with QLoRA Cassandra Lunch 4, presented by Anant
LLM Fine Tuning with QLoRA Cassandra Lunch 4, presented by Anant
Anant Corporation
 
Generative AI leverages algorithms to create various forms of content
Generative AI leverages algorithms to create various forms of contentGenerative AI leverages algorithms to create various forms of content
Generative AI leverages algorithms to create various forms of content
Hitesh Mohapatra
 
BRAIN TUMOR DETECTION for seminar ppt.pdf
BRAIN TUMOR DETECTION for seminar ppt.pdfBRAIN TUMOR DETECTION for seminar ppt.pdf
BRAIN TUMOR DETECTION for seminar ppt.pdf
LAXMAREDDY22
 
Material for memory and display system h
Material for memory and display system hMaterial for memory and display system h
Material for memory and display system h
gowrishankartb2005
 
Properties Railway Sleepers and Test.pptx
Properties Railway Sleepers and Test.pptxProperties Railway Sleepers and Test.pptx
Properties Railway Sleepers and Test.pptx
MDSABBIROJJAMANPAYEL
 
Use PyCharm for remote debugging of WSL on a Windo cf5c162d672e4e58b4dde5d797...
Use PyCharm for remote debugging of WSL on a Windo cf5c162d672e4e58b4dde5d797...Use PyCharm for remote debugging of WSL on a Windo cf5c162d672e4e58b4dde5d797...
Use PyCharm for remote debugging of WSL on a Windo cf5c162d672e4e58b4dde5d797...
shadow0702a
 
Comparative analysis between traditional aquaponics and reconstructed aquapon...
Comparative analysis between traditional aquaponics and reconstructed aquapon...Comparative analysis between traditional aquaponics and reconstructed aquapon...
Comparative analysis between traditional aquaponics and reconstructed aquapon...
bijceesjournal
 
Data Control Language.pptx Data Control Language.pptx
Data Control Language.pptx Data Control Language.pptxData Control Language.pptx Data Control Language.pptx
Data Control Language.pptx Data Control Language.pptx
ramrag33
 

Recently uploaded (20)

AI assisted telemedicine KIOSK for Rural India.pptx
AI assisted telemedicine KIOSK for Rural India.pptxAI assisted telemedicine KIOSK for Rural India.pptx
AI assisted telemedicine KIOSK for Rural India.pptx
 
Rainfall intensity duration frequency curve statistical analysis and modeling...
Rainfall intensity duration frequency curve statistical analysis and modeling...Rainfall intensity duration frequency curve statistical analysis and modeling...
Rainfall intensity duration frequency curve statistical analysis and modeling...
 
IEEE Aerospace and Electronic Systems Society as a Graduate Student Member
IEEE Aerospace and Electronic Systems Society as a Graduate Student MemberIEEE Aerospace and Electronic Systems Society as a Graduate Student Member
IEEE Aerospace and Electronic Systems Society as a Graduate Student Member
 
学校原版美国波士顿大学毕业证学历学位证书原版一模一样
学校原版美国波士顿大学毕业证学历学位证书原版一模一样学校原版美国波士顿大学毕业证学历学位证书原版一模一样
学校原版美国波士顿大学毕业证学历学位证书原版一模一样
 
Data Driven Maintenance | UReason Webinar
Data Driven Maintenance | UReason WebinarData Driven Maintenance | UReason Webinar
Data Driven Maintenance | UReason Webinar
 
Curve Fitting in Numerical Methods Regression
Curve Fitting in Numerical Methods RegressionCurve Fitting in Numerical Methods Regression
Curve Fitting in Numerical Methods Regression
 
Optimizing Gradle Builds - Gradle DPE Tour Berlin 2024
Optimizing Gradle Builds - Gradle DPE Tour Berlin 2024Optimizing Gradle Builds - Gradle DPE Tour Berlin 2024
Optimizing Gradle Builds - Gradle DPE Tour Berlin 2024
 
Unit-III-ELECTROCHEMICAL STORAGE DEVICES.ppt
Unit-III-ELECTROCHEMICAL STORAGE DEVICES.pptUnit-III-ELECTROCHEMICAL STORAGE DEVICES.ppt
Unit-III-ELECTROCHEMICAL STORAGE DEVICES.ppt
 
Certificates - Mahmoud Mohamed Moursi Ahmed
Certificates - Mahmoud Mohamed Moursi AhmedCertificates - Mahmoud Mohamed Moursi Ahmed
Certificates - Mahmoud Mohamed Moursi Ahmed
 
artificial intelligence and data science contents.pptx
artificial intelligence and data science contents.pptxartificial intelligence and data science contents.pptx
artificial intelligence and data science contents.pptx
 
An improved modulation technique suitable for a three level flying capacitor ...
An improved modulation technique suitable for a three level flying capacitor ...An improved modulation technique suitable for a three level flying capacitor ...
An improved modulation technique suitable for a three level flying capacitor ...
 
ITSM Integration with MuleSoft.pptx
ITSM  Integration with MuleSoft.pptxITSM  Integration with MuleSoft.pptx
ITSM Integration with MuleSoft.pptx
 
LLM Fine Tuning with QLoRA Cassandra Lunch 4, presented by Anant
LLM Fine Tuning with QLoRA Cassandra Lunch 4, presented by AnantLLM Fine Tuning with QLoRA Cassandra Lunch 4, presented by Anant
LLM Fine Tuning with QLoRA Cassandra Lunch 4, presented by Anant
 
Generative AI leverages algorithms to create various forms of content
Generative AI leverages algorithms to create various forms of contentGenerative AI leverages algorithms to create various forms of content
Generative AI leverages algorithms to create various forms of content
 
BRAIN TUMOR DETECTION for seminar ppt.pdf
BRAIN TUMOR DETECTION for seminar ppt.pdfBRAIN TUMOR DETECTION for seminar ppt.pdf
BRAIN TUMOR DETECTION for seminar ppt.pdf
 
Material for memory and display system h
Material for memory and display system hMaterial for memory and display system h
Material for memory and display system h
 
Properties Railway Sleepers and Test.pptx
Properties Railway Sleepers and Test.pptxProperties Railway Sleepers and Test.pptx
Properties Railway Sleepers and Test.pptx
 
Use PyCharm for remote debugging of WSL on a Windo cf5c162d672e4e58b4dde5d797...
Use PyCharm for remote debugging of WSL on a Windo cf5c162d672e4e58b4dde5d797...Use PyCharm for remote debugging of WSL on a Windo cf5c162d672e4e58b4dde5d797...
Use PyCharm for remote debugging of WSL on a Windo cf5c162d672e4e58b4dde5d797...
 
Comparative analysis between traditional aquaponics and reconstructed aquapon...
Comparative analysis between traditional aquaponics and reconstructed aquapon...Comparative analysis between traditional aquaponics and reconstructed aquapon...
Comparative analysis between traditional aquaponics and reconstructed aquapon...
 
Data Control Language.pptx Data Control Language.pptx
Data Control Language.pptx Data Control Language.pptxData Control Language.pptx Data Control Language.pptx
Data Control Language.pptx Data Control Language.pptx
 

eventnet asynchronous recursive event processing

  • 1. /18 EventNet: Asynchronous Recursive Event Processing Study 2020/09/18 2011136 Ryuta Shitomi 1 Yusuke Sekikawa, Kosuke Hara, and Hideo Saito, CVPR2019
  • 2. /18 The event camera 2  High dynamic range  Do not suffer from motion blur  No latency  Recently, frame-based approach have achieved impressive results in scenarios.
  • 3. /18 The event data from event-based camera 3  When the new event occurred: 𝑒𝑖 = 𝑥𝑖, 𝑦𝑖, 𝑝𝑖, 𝑡 − 𝑡′ 𝑡′ : time of the last event  Asynchronous  Sparse  Many events (1 mega event per sec) ↓ Need a new learning approach to differ from frame-based approach
  • 4. /18 Frame-based approach for an event streams 4  Integration 𝛥𝑡 to create a frame and feed a neural net.  Disadvantages: Cannot use of the sparseness of the data Redundant computation A. I. Maqueda, A. Loquercio, G. Gallego, N. N. Garcia, and D. Scaramuzza. Event-based vision meets deep learning on steering prediction for self-driving cars. CoRR, abs/1804.01310, 2018. 𝜟𝒕
  • 5. /18 Purpose and Contribution 5  Develop a novel neural network: can process an extremely high-rate and variable length. for real-time processing of an asynchronous event stream. to output recursively using temporal information. 𝜏′ 𝜏 𝑡 𝑝𝑜𝑙𝑎𝑟𝑖𝑡𝑦
  • 6. /18 Problem Statement 6  Consider a sequence of events within a 𝜏 ms interval based on the latest time stamp 𝑡𝑗 of the event: 𝒆𝑗 = 𝑒𝑖 𝑖 = 𝑗 − 𝑛 𝑗 + 1, … , 𝑗} 𝑒𝑖 = 𝑥𝑖, 𝑦𝑖, 𝑝𝑖, Δ𝑡𝑗,𝑖 Δ𝑡𝑗,𝑖 = 𝑡𝑗 − 𝑡𝑖  Consider a neural network 𝑓 to estimate 𝑦𝑗 given event stream 𝒆𝒋 𝑦𝑗 = 𝑓 𝒆𝑗 𝑦𝑗: Task dependent 𝜏 𝑡 𝒆𝒋 𝜏 𝒆𝒋+𝟏
  • 7. /18 Problem Statement 7  To realize a trainable neural network 𝑓 that satisfies following conditions: 1. End-to-End Trainable 2. Event-Wise Processing 3. Efficient Recursive Processing 4. Local Permutation Invariance 𝜏 𝑡 𝒆𝒋 𝜏 𝒆𝒋+𝟏
  • 8. /18 Using PointNet architecture 8  PointNet solves the problem by approximating the function 𝑓: 𝑦𝑗 = 𝑓 𝒆𝑗 = 𝑔 max ℎ 𝑒𝑗−𝑛 𝑗 +1 , … , ℎ 𝑒𝑗 where ℎ: ℝ4 → ℝ 𝐾 , 𝑔: ℝ 𝐾 → ℝ  Because of the max(∙) , the permutation of events does not change the output 𝑦𝑗. MLP  Problem: A huge amount of computation of ℎ ∙ and max ∙ . Should run on the event rate. (could be more than 1MEPS) When a new j + 1 -th event arrives, Δtj,i = tj − ti changes to tj+1 − ti . Many previous event has already been processed by h(∙).
  • 9. /18 EventNet 9  Introduce the ”Temporal Coding” and “Recursive Processing” to overcome the difficulty. The Δ𝑡 only changes as the new event is being received. Split the ℎ(𝑒): ℎ 𝑒𝑖 = 𝑐(ℎ 𝑒𝑖 − , Δ𝑡𝑗,𝑖), 𝑒− : (𝑥, 𝑦, 𝑝) 𝑐 ∙ :temporal coding function, ℎ: ℝ3 → ℂ 𝐾 𝑦𝑗 become: 𝑓 𝒆𝑗 = 𝑔 max 𝑐 𝑧𝑗−𝑛 𝑗 +1, Δ𝑡𝑗,𝑗−𝑛 𝑗 +1 , … , 𝑐 𝑧𝑗, 0 where 𝑧𝑖 = ℎ 𝑒− ∈ ℂ 𝐾 𝑐 ∙ , max ∙ need to be computed for all events. Temporal Coding
  • 10. /18 EventNet 10 • Assume the norm of each element of 𝑧𝑖 is less than 1 (by tanh). • max ∙ , 𝑐 ∙ need to be recursive. Example) 𝑠𝑗 = max 𝑠𝑗−1, 𝑐 𝑧𝑗, 0 𝑠𝑗+1 = max 𝑠𝑗, 𝑐 𝑧𝑗+1, 0 • Propose temporal coding function to satisfy this condition: 𝑎𝑗,𝑖 = 𝑐 𝑧𝑖, Δ𝑡𝑗,𝑖 = 𝑧𝑖 − Δ𝑡𝑗,𝑖 𝜏 + exp −i 2𝜋Δ𝑡𝑗,𝑖 𝜏 Recursive Processing Encode the elapsed time
  • 11. /18 Temporal Coding Function 11 𝑎𝑗,𝑖 = 𝑐 𝑧𝑖, Δ𝑡𝑗,𝑖 = 𝑧𝑖 − Δ𝑡𝑗,𝑖 𝜏 + exp −i 2𝜋Δ𝑡𝑗,𝑖 𝜏 When the new event is received: 𝑎𝑗+1,𝑖 = 𝑐 𝑧𝑖, Δ𝑡𝑗+1,𝑖 = 𝑧𝑖 − Δ𝑡𝑗,𝑖 𝜏 − 𝑡𝑗+1 − 𝑡𝑗 𝜏 + exp −i 2𝜋(Δ𝑡𝑗,𝑖 − 𝑡𝑗+1 + 𝑡𝑗 𝜏 Recursive Processing 𝑡 |𝑎 𝑗,𝑖| 𝑒𝑗 New event is received 𝑡 |𝑎 𝑗,𝑖| 𝑒𝑗 𝑒𝑗+1 − 𝑡𝑗+1 − 𝑡𝑗
  • 12. /18 Temporal Coding Function 12 𝑡 |𝑎 𝑗,𝑖| 𝑒𝑗 New event is received 𝑡 |𝑎 𝑗+1,𝑖| 𝑒𝑗 𝑒𝑗+1 − 𝑡𝑗+1 − 𝑡𝑗 𝑎𝑗,𝑗−2 = max 𝑎𝑗,𝑗−5, … , 𝑎𝑗,𝑗  when you new event is received. 𝑎𝑗+1,𝑗−2 = max 𝑐 𝑎𝑗,𝑗−2, 𝑡𝑗+1 − 𝑡𝑗 , 𝑎𝑗+1,𝑗+1  Therefore, 𝑦𝑗+1: 𝑦𝑗+1 = 𝑓 𝒆𝒋+𝟏 = 𝑔 max 𝑐 𝑠𝑗, 𝑡𝑗+1 − 𝑡𝑗 , ℎ 𝑒𝑗+1 − 𝑒𝑗−2 𝑒𝑗−5 |𝑎 𝑗,𝑗−2|
  • 13. /18 Overview of EventNet 13  First event is arrived: 𝒆0 = {𝑒0} 𝑧0 = ℎ(𝑒0 − ) 𝑦0 = 𝑓 𝒆0 = 𝑔 max 𝑐 𝑧0, 0 𝑡 𝑐(𝑧0, 0) 𝑗 = 0 𝑗 = 1  Next event is arrived: 𝒆1 = {𝑒0, 𝑒1} 𝑠1 = max(𝑐 𝑧0, 𝑡1 − 𝑡0 , ℎ(𝑒1 − ) 𝑦1 = 𝑓 𝒆1 = 𝑔(𝑠1)  Next event is arrived: 𝒆2 = {𝑒0, 𝑒1, 𝑒2} 𝑠2 = max 𝑐 𝑠1, 𝑡2 − 𝑡1 , ℎ 𝑒2 − 𝑦3 = 𝑔 𝑠2 𝑗 = 2
  • 14. /18 EventNet Architecture 14  Create Look-up table(LUT) for decreasing inference time. (There are only 𝑊 × 𝐻 × 2 patterns in inputs 𝑒− ) ℎ(∙) 𝑔(∙)
  • 15. /18 Experiments 15  ETHTED+ dataset for target motion estimation and semantic segmentation  MVSEC dataset for ego-motion estimation
  • 16. /18 Results for target motion estimation 16
  • 17. /18 Results for ego-motion 17  Estimated ego motions are shown as yellow.  Estimated at the rate of 1000 Hz.
  • 18. /18 Compare the computation time [μs] 18  Using newly received event, PointNet need to process the all events again.