Deep Visual Semantic Embedding with Text Data Augmentation and Word Embedding Initialization

Table 4

Experimental results with text data augmentation on Flickr30k.

Model	Feature (s)	Image count				Image retrieval
		R@1	R@5	R@10	Med r	R@1	R@5	R@10	Med r
DeViSE	Word2Vec	4.5	18.1	29.2	26	6.7	21.9	32.7	25
DeFrag	R–CNN	16.4	40.2	54.7	8	10.3	31.4	44.5	13
VSA	R–CNN + BRNN	22.2	48.2	61.4	4.8	15.2	37.7	50.5	9.2
UVSE	ConvNet + LSTM	14.8	39.2	50.9	10	11.8	34.0	46.3	13
UVSE (VGG)	VGG + LSTM	23.0	50.7	62.9	5	16.8	42.0	56.5	8
VSE++	VGG + GRU + HNM	29.0	54.4	66.5	4	20.3	48	59.9	6
Ours	Aug	30.6	57.9	68.5	4	21.4	49.3	61.4	6
Ours	Aug + Word2Vec	33.4	59.2	69.6	3	23.3	49.9	61.7	6