SLIDE 1
Tools for Efficient Object Detection
Organizers:
ICCV 2015 Tutorial
Santiago, Chile, December 2015
Rogerio Feris
Classification Versus Detection
Slide Adapted from Ross Girshick
Classification: WHAT Detection: WHAT and WHERE
Tools for Efficient Object Detection ICCV 2015 Tutorial Santiago, - - PowerPoint PPT Presentation
Tools for Efficient Object Detection ICCV 2015 Tutorial Santiago, - - PowerPoint PPT Presentation
Tools for Efficient Object Detection ICCV 2015 Tutorial Santiago, Chile, December 2015 Organizers: Classification Versus Detection Rogerio Feris Classification: WHAT Detection: WHAT and WHERE Slide Adapted from Ross Girshick Rogerio Feris
SLIDE 2
SLIDE 3
Rogerio Feris
Efficient Object Detection
- Object detection is arguably a harder problem than image classification
- Usually a large number of image sub-windows need to be scanned in
- rder to localize objects, leading to heavy computational processing
- Challenge: In many real-world applications, running a fast object
detector is as critical as running an accurate object detector.
SLIDE 4
Rogerio Feris
Applications
MobilEye Forward Collision Warning [Click for video demo]
SLIDE 5
Rogerio Feris
Applications
Funny Nikon ad: “"The Nikon S60 detects up to 12 faces."
Slide credit: Lana Lazebnik
SLIDE 6
Rogerio Feris
Applications
IBM Intelligent Video Analytics [Click for video demo]
SLIDE 7
Rogerio Feris
Applications
Body-worn Cameras [Click for video demo] (using Fast R-CNN)
SLIDE 8
Rogerio Feris
Applications
Many more applications require real-time object detection…
Augmented Reality Robotics Wildlife Monitoring Self-Driving Cars Mobile
SLIDE 9
Rogerio Feris
Tutorial Overview
SLIDE 10
Rogerio Feris
Goals:
- Cover tools for speeding-up object detection while
maintaining high accuracy
- Focus on the state of the art
- Focus on software tools instead of hardware acceleration
- Provide pointers to publicly available source code
SLIDE 11
Rogerio Feris
How to design a detector running at 100 Hz (CPU only), step by step
(Rodrigo Benenson)
- What makes strong rigid templates
- Integral Channels and Aggregated Features
- Feature Approximation Across Scales
- Cascades
- Geometric Prior
Figure credit: Rodrigo Benenson and Piotr Dollar
SLIDE 12
Rogerio Feris
Region Proposals
(Jan Hosang)
- Grouping proposal methods
- Window scoring proposal
methods
- Metrics and in-depth analysis
Towards generic object detection: candidate region generation
Figure credit: Jan Hosang
SLIDE 13
Rogerio Feris
Regionlets for Generic Object Detection
(Xiaoyu Wang)
- Regionlet representation for handling object deformations
- Classification of region proposals based on boosted detector cascades
- Integration with CNN features
Figure credit: Xiaoyu Wang
SLIDE 14
Rogerio Feris
Tools for fast CNN-based Detection
Figure credit: Kaiming He
Kaiming He (Inference) Ross Girshick (Training) “Slow” R-CNN Fas ast R-CNN Fas aster R-CNN
SLIDE 15
Rogerio Feris
Schedule
14:00 Intr trod
- ductio
ion (Rogerio Feris) 14:15 De Detectin ing g ob
- bjects at
t 100 Hz Hz wi with ri rigid gid templa lates (Rodrigo Benenson) 15:00 Coffee Break 15:30 Regi egion prop
- pos
- sals
ls (Jan Hosang) 16:00 Regi egionle let Obj Object De Detector
- r wi
with Han Hand-crafted an and CN CNN Fea eatures (Xiaoyu Wang) 16:30 Co Convol
- lutio
ional l Fea eature Map Maps: : El Elements of
- f effic
ficie ient CN CNN-based ob
- bject
de detectio ion (Kaiming He) 17:15 Train ainin ing g R-CNNs of
- f vario
ious velo elocit itie ies: Slo Slow, fas ast, and and fas aster (Ross Girshick) 18:00 Concluding Remarks