A Symmetric Fusion Learning Model for Detecting Visual Relations and Scene Parsing

Table 4

Ablation studies on the key components of our model. We report the phrase and relationship detection performance in R@n scores (R@50 and R@100).

K	Method	Relation detection		Phrase detection
		R@50	R@100	R@50	R@100
1	Baseline (20% noisy labels)	18.00	22.44	26.37	31.42
	Baseline +  +  (20% noisy labels)	24.09	28.13	29.34	34.51
70	Baseline (20% noisy labels)	21.52	26.38	28.24	35.44
	Baseline +  +  (20% noisy labels)	25.06	30.83	30.47	38.07