1 of 19

ANNOUNCEMENT OF WINNERS

Sponsored by:

2 of 19

Interest in the dataset

2

3 of 19

24 Submissions

3

4 of 19

Christopher Boehnen

Walter Scheirer

Adam Czajka

4

Judges

5 of 19

Test Dataset

Challenge 1

300 Images

Challenge 2

8,910 images → 9,187 objects

5

Collection

UAV

Glider

Ground

Total

Videos

19

15

21

55

Frames

2,814

2,909

3,187

8,910

Extracted objects

3,000

3,000

3,187

9,187

ImageNet superclasses

28

17

21

42

6 of 19

CHALLENGE 1:

IMAGE ENHANCEMENT TO FACILITATE MANUAL INSPECTION

7 of 19

Perceived deterioration/improvement

7

8 of 19

Perceived deterioration/improvement

8

9 of 19

Scoring

9

Organization

Overall Score

Honeywell - ACST

126

Northwestern University

59

National Tsing Hua University

22

Johns Hopkins

16

Noblis

14

Texas A&M University, Peking University

10

1

2

10 of 19

CHALLENGE 2:

IMAGE ENHANCEMENT TO IMPROVE AUTOMATIC OBJECT RECOGNITION

11 of 19

Baseline Classification

11

12 of 19

Classification results

12

13 of 19

Classification results

13

14 of 19

Classification results

14

15 of 19

Participant vs. previously tested algorithms

15

16 of 19

Participant vs. previously tested algorithms

16

17 of 19

Scoring

17

Metric

UAV

Glider

Ground

Inception-M1

TexasAM&Peking

Inception-M2

TexasAM&Peking

ResNet-M1

Honeywell

ResNet-M2

Honeywell

Honeywell

VGG16-M1

VGG16-M2

Northwestern

NationalTsingHua

VGG19-M1

Honeywell

VGG19-M2

Northwestern

TexasAM&Peking

18 of 19

Scoring

18

Organization

Overall Score

Honeywell - ACST

4

Texas A&M University, Peking University

3

Northwestern University

2

National Tsing Hua University

1

Noblis

0

Johns Hopkins

0

1

2

19 of 19

Thank You

Support for the research and challenge workshop is provided under IARPA contract #2016-16070500002. This research is supported in part by the Office of the Director of National Intelligence (ODNI), Intelligence Advanced Research Projects Activity (IARPA). The views and conclusions contained herein are those of the organizers and should not be interpreted as necessarily representing the official policies, either expressed or implied, of ODNI, IARPA, or the U.S. Government. The U.S. Government is authorized to reproduce and distribute reprints for governmental purposes notwithstanding any copyright annotation therein.