Skip to main content
Account

Table 6 Comparison of prototype memory and allocation variants on PredCls

From: AMCA-SGG: Adaptive multi-prototype learning and cross-modal alignment for unbiased scene graph generation

Variants

mR@20

mR@50

mR@100

w/o M

32.3

38.4

40.7

Single prototype (\(K=1\))

32.8

38.8

40.9

Fixed multi-prototype (\(K=4\))

35.1

40.0

42.4

Fixed multi-prototype (\(K=7\))

35.5

40.6

42.2

Fixed multi-prototype (\(K=10\))

33.3

39.2

41.6

Frequency-based allocation

33.9

39.2

41.4

Entropy-based allocation

36.5

41.9

43.9

Random permutation

36.7

41.3

43.7

Adaptive allocation (Ours)

37.2

42.8

44.5

  1. The best results are highlighted in bold