• No results found

Defocus and Motion Blur Detection with Deep Contextual Features

N/A
N/A
Protected

Academic year: 2022

Share "Defocus and Motion Blur Detection with Deep Contextual Features"

Copied!
46
0
0

Laster.... (Se fulltekst nå)

Fulltekst

(1)

Defocus and Motion Blur Detection with Deep Contextual Features

Supplementary Material

Beomseok Kim

1

, Hyeongseok Son

1

, Seong‐Jin Park

1

, Sunghyun Cho

2

, and Seungyong Lee

1

1

POSTECH       

2

DGIST

(2)

Contents

• More comparisons with a state‐of‐the‐art method in blur detection  on CUHK test set

• Huang et al., “Multiscale blur detection by learning discriminative deep  features”, Neurocomputing, 2018.

• Challenging examples with mixed blur

• Objects outside the depth‐of‐field are motion‐blurred

• Examples of moving object segmentation

• More results with real blurred photographs

2

(3)

More comparisons of blur detection results on CUHK test set

(4)

input

Huang et el. ours

ground-truth

4

(5)

input

Huang et el. ours

ground-truth

5

(6)

input

Huang et el. ours

ground-truth

6

(7)

input

Huang et el. ours

ground-truth

7

(8)

input

Huang et el. ours

ground-truth

8

(9)

input

Huang et el. ours

ground-truth

9

(10)

input

Huang et el. ours

ground-truth

10

(11)

input

Huang et el. ours

ground-truth

11

(12)

input

Huang et el. ours

ground-truth

12

(13)

input

Huang et el. ours

ground-truth

13

(14)

input

Huang et el. ours

ground-truth

14

(15)

input

Huang et el. ours

ground-truth

15

(16)

input

Huang et el. ours

ground-truth

16

(17)

input

Huang et el. ours

ground-truth

17

(18)

input

Huang et el. ours

ground-truth

18

(19)

input

Huang et el. ours

ground-truth

19

(20)

input

Huang et el. ours

ground-truth

20

(21)

input

Huang et el. ours

ground-truth

21

(22)

input

Huang et el. ours

ground-truth

22

(23)

input

Huang et el. ours

ground-truth

23

(24)

input

Huang et el. ours

ground-truth

24

(25)

Challenging examples with mixed blur

(Objects outside the depth‐of‐field are motion‐blurred) (blue: motion blur, red: defocus blur, black: no‐blur)

Our method returns the dominant blur type at each pixel in this case.

(26)

input output

26

(27)

input output

27

(28)

input output

28

(29)

input output

29

(30)

input output

30

(31)

Examples of moving object segmentation with real photographs

(blue: motion blur, red: defocus blur, black: no‐blur)

(32)

input output

32

(33)

input output

33

(34)

input output

34

(35)

input output

35

(36)

input output

36

(37)

input output

37

(38)

input output

38

(39)

input output 39

(40)

More examples with real blurred photographs

(blue: motion blur, red: defocus blur, black: no‐blur)

(41)

41

(42)

42

(43)

43

(44)

44

(45)

45

(46)

Thank you.

http://cg.postech.ac.kr

http://vclab.dgist.ac.kr

Referanser

RELATERTE DOKUMENTER

We then choose, in section 3, a functional form (CES) for the import share functions, and specify their dynamic and stochastic formulation. In section 4 we briefly explain the data:

Waste input-output; circular economy; secondary metal production; multi-regional input-output 11.. analysis; socio-economic impacts

Figure 1: Photographer’s techniques to generate motion blur images while maintaining the clear shape of objects and the hint of motion direction.. choose to use a simplified version

Reference implementation of multiple rendering To have a standard for comparisons, we also implemented motion blur based on temporal supersampling: Images that correspond to N

This paper describes a motion blur technique which can be applied to rendering fluid simulations that are carried out in the Eulerian framework.. Existing motion blur techniques can

Figure 1: Shading cost comparison for a complex scene rendered without motion and defocus blur (left), stochastic motion and defocus blur with decoupled sampling (center),

We show the use of covariance tracing with various applications ranging from motion blur and depth-of-field adaptive sampling and filtering, photon mapping kernel size estimation

Adaptive Shading for Defocus Blur: Figure 8 (middle) shows the CITADEL 3 scene rendered with a post process defocus blur, where shading is evaluated at a lower rate in the