PexAce: A Method for Image Metadata Acquisition - Semantic Acquisition Games: Harnessing Manpower for Creating Semantics - page 69

01234567890 Tags
1 00000000000
2 265 143 33 9 1 0 0 0 0 0 0
3 133 113 29 12 1 0 0 0 0 0 0
4 56 62 26 6 1 2 0 0 0 0 0
5 23 33 22 5 0 7 2 0 0 0 0
6 13 23 15 13 3 11 3 0 0 0 0
7 1 7833 440000
8 4 9 11 7 1 30 5 1 0 0 0
9 1 0 3 55 451000
10 0 6 1 24 533100
11 1 8 15 5 0 37 10 5 1 0 0
12 15352 9 50100
13 03122 7 33000
14 00220 5 35000
15 0 26 83 155 175 75 22 10 4 1 0
16 1 8 12 32 39 24 13 7 0 0 0
17 044 4 8714100
18 00254321020
19 01522213000
20 0126 0330000
21 0007 1201000
22 00335201000
23 00131143000
Annotations

Fig. 5.3 Number of cases ( images ) where specific number of annotations ( rows )resultedin
a specific number of tags ( columns ). With the increasing number of annotations, the tradeoff for
tags tends to slow. The bottom of the reverse “L” shaped cluster of higher values suggests, that for
some images a relatively high number of annotations results only in a low number of tags (usually
due to the few dominant concepts)

Fig. 5.4 There was a variety
of tag counts for images that
were annotated by the same
number of players (in this
case, 15). An effect caused by
topic diversity

100

80

60

40

20

0

1 2 3 4 5 6 7 8

Number of tags per image