Upload
others
View
0
Download
0
Embed Size (px)
Citation preview
Sensing and Sensibility: a quest to visual intelligence
Silvio Savarese
May 17th, 2016
New Frontiers in Computing
Sensing is the future
4
Sensing is the future
5
Sensing is the future
6
Sensing is the future
Sensing is the future
Everything is a sensor…
Everything is a sensor…
Everything is a sensor…
Modern vision sensors
night
Kinect
thermal
w/ gravity
Sensing is not the hard problem
Intelligent understanding of the sensing data is the challenge!
What does it mean “intelligent understanding of the sensing data”?
car
car
Car
Street
Building
image labels
Image-to-labels
Turk & Pentland, 91Poggio et al., 93Belhumeur et al., 97LeCun et al. 98Amit and Geman, 99Shi & Malik, 00Viola & Jones, 00Felzenszwalb & Huttenlocher 00Belongie & Malik, 02Ullman et al. 02
Argawal & Roth, 02Ramanan & Forsyth, 03Weber et al., 00Vidal-Naquet & Ullman 02Fergus et al., 03Torralba et al., 03Vogel & Schiele, 03Barnard et al., 03Fei-Fei et al., 04Kumar & Hebert ‘04
He et al. 06Gould et al. 08Maire et al. 08Felzenszwalb et al., 08Kohli et al. 09L.-J. Li et al. 09Ladicky et al. 10,11Gonfaus et al. 10Farhadi et al., 09 Lampert et al., 09
But…
• It is just one ingredient of a much more complex problem.
14
15
Chocolate bar
Tree
Road
Sky
Wheels
16
Sky
Chocolate bar
Tree
Road
Wheels
18
Car-right
Car-right
Car-left
Car- 3/4 right
toy car
Building facade
Road
Car-3/4 right
Road
19
Building facade
20
21
22
chasing
Intelligent understanding of sensory data is not just a labeling problem!
It implies recognizing objects, their physical properties and their relationship with the environment within which the objects live.
24
Biederman, Mezzanotte and Rabinowitz, 1982
25
Humans perceive the world in 3D
V1
where pathway(dorsal stream)
what pathway(ventral stream)
26
Humans perceive the world in 3D
V1Pre-frontal
cortex
27
Humans perceive the world in 3D
where pathway(dorsal stream)
what pathway(ventral stream)
My group’s research
Space understanding
• 3D shape recovery • 3D scene reconstruction • Camera localization• Pose estimation
Recognition
• Object detection• Texture classification• Target tracking• Activity recognition
28
• Objects
• Space
• Activities
From sensing intelligence• Images
• Videos
• RGB-Depth
From sensing to the 3D objects
CHAIR
BED
TABLE
Xiang & Savarese, 2012-2014
CAR
30
Car Person Tree Sky
Street Building Else
From sensing to the 3D scenes
…Bao & Savarese, 2011-2013
31
Car Person Tree Sky
Street Building Else
… Bao
& S
avar
ese,
20
11
From sensing to the 3D scenesBao & Savarese, 2011-2013
Interactions between:- Objects-space
- Object-object- Object-scene class
From sensing to the 3D scenes
32
Choi, Chao, Pantofaru, Savarese, CVPR 13
From 3D point clouds to 3D dynamic scenes
33
Held, Thrun & Savarese, RSS 2014
• Interactions among humans
34
queuing
talking
From sensing to activities
• Interaction human-3D space
Choi et al., VSWS 09Choi et al., CVPR 11Choi & Savarese, ECCV 2012
Queuing
Crossing streetChoi, Savarese, 2012-2014
From sensing to activities
Sensors
Objects
Sensing and sensibility paradigm
3D physical environment
Applications
37
Social Robotics
Mobile vision
Large scale information management
Safe driving
Visual intelligence and large scale information management
Golparvar-Fard, Pena-Mora, Savarese , 2008-2012
James R. Croes Medal, October 2013 (from the American Society of Civil of Engineers)
Automatic coordination of construction progress can lead to huge savings (10 billions USD/year) in construction business!
[Census Bureau, www.census.gov, 2007]
39
Opportunity to modernize age-old process in a profound way
freeing up critical human resources
12/02/2006; 1:13:00 PM (As-built)
Our revolution
- Thousands of images- Computer vision
Our revolution
Images are cheap!
42
Our revolution
Our revolution
12/02/2006; 1:13:00 PM (As-built)
Our revolution
45
Behind ScheduleOn ScheduleAhead of Schedule
12/02/2006; 1:13:00 PM (As-planned)
12/02/2006; 1:13:00 PM (As-planned)12/02/2006; 1:13:00 PM (As-built)
46
Behind ScheduleOn ScheduleAhead of Schedule
12/02/2006; 1:13:00 PM (As-planned)
12/02/2006; 1:13:00 PM (As-planned)12/02/2006; 1:13:00 PM (As-built)
Large impact in the civil engineering community
47
• James R. Croes Medal, October 2013 (from the American Society of Civil of Engineers)• Best paper award from journal of CEM, 2011 • Best paper award at AEC/FM 2010• Best paper award at Construction Research Congress 2009
• Better tools for visualization• Automatic communication of performance deviations• Reduction in delivery time• Safety management• Potential to identify unsafe locations/components
Applications
48
Social Robotics
Mobile vision
Large scale information management
Safe driving
The Jackrabbot
49An Autonomous Perambulating Robot for On-campus Deliveries
The Jackrabbot project
Social affinity
Social affinity
The Jackrabbot project
The Jackrabbot project
Thank you!
TOYOTA
53
54