P e o p l e A very sparse semantical vector for frame: - - PowerPoint PPT Presentation

▶
p e o p l e a very sparse semantical vector for frame
SMART_READER_LITE
LIVE PREVIEW

P e o p l e A very sparse semantical vector for frame: - - PowerPoint PPT Presentation

Jan 09, 2024 399 likes •638 views

Time Frame Baby Cake People P e o p l e A very sparse semantical vector for frame: Emphasize primary object Vector sparser Overlook small size regional object Vocabulary larger max pooling Only around 40% of Regional Discriminatory power

slide-1
SLIDE 1
slide-2
SLIDE 2

Frame

P e

  • p

l e

Baby Cake People Time

slide-3
SLIDE 3

A very sparse semantical vector for frame: Emphasize primary object Overlook small size regional object

Vector sparser Vocabulary larger

slide-4
SLIDE 4
slide-5
SLIDE 5

Only around 40% of Regional Information left Discriminatory power of deep features consistently improves

max pooling

slide-6
SLIDE 6
  • Framework-1
  • Framework-2
slide-7
SLIDE 7
slide-8
SLIDE 8

Selective search Candidate Objects Frame

On average, each frame has 20 candidate object regions.

slide-9
SLIDE 9

Observations

  • Possible reasons:
  • Alternative method:
slide-10
SLIDE 10
slide-11
SLIDE 11
slide-12
SLIDE 12

VLAD

slide-13
SLIDE 13

Selective search Candidate Objects Spatial & temporal features clustering Regional objects Deep features K-means & VLAD

Zhongwen Xu, Yi Yang, Alexander G. Hauptmann (CVPR’15)

slide-14
SLIDE 14

Deep Feature Map Extraction

Feature Map 7 X 7

Spatial Pyramid Pooling

7 X 7 6 X 6 5 X 5 2 X 2 50 descriptors

VLAD

Max pool filter: 50 descriptors

Feature Spatial Pyramid Pooling filter: Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun (ECCV’14)

slide-15
SLIDE 15

VLAD

slide-16
SLIDE 16
slide-17
SLIDE 17

20 25 30 35 40 45 MED14-Test (mAP%) MED16-EvalSub (MinfAP200%) MED16-EvalFull (MinfAP200%)

PS-10Ex

CNN-VLAD Object-VLAD

20 25 30 35 40 45 MED14-Test (mAP%) MED16-EvalSub (MinfAP200%) MED16-EvalFull (MinfAP200%)

PS-100Ex

CNN-VLAD Object-VLAD

slide-18
SLIDE 18
slide-19
SLIDE 19

20 22 24 26 28 30 32 34 36 38 40 MED14-Test (mAP%) MED16-EvalSub (MinfAP200%) MED16-EvalFull (MinfAP200%)

PS-10Ex

Concept-Bank_N2 Object-VLAD Visual-System (Concept-Bank_N2 + Object-VLAD)

slide-20
SLIDE 20

25 27 29 31 33 35 37 39 41 43 45 VIREO Team2 Team3 Team4

MED16-EvalFull-As- ProgressSubset (MinfAP200%)

25 30 35 40 45 50 Team2 VIREO Team3 Team4

MED16-EvalFull (MinfAP200%)

5 10 15 20 25 30 35 40 45 Team2 VIREO Team3 Team4 Team5 Team6 Team7 Team8 Team9 Team10 Team11

MED16-EvalSub (MinfAP200%)

25 27 29 31 33 35 37 39 41 Team2 VIREO Team3 Team4

MED16-EvalFull (MinfAP200%)

slide-21
SLIDE 21

5 10 15 20 25 30 35 40 45 50 Team2 Team3 Team4 Team5 VIREO Team6 Team7 Team8 Team9

MED16-EvalSub (MinfAP200%)

5 10 15 20 25 30 35 40 45 Team2 VIREO Team4 Team5

MED16-EvalFull (MinfAP200%)

slide-22
SLIDE 22