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

Table 3

Experimental results with text data augmentation on Flickr8k.

Model	Feature (s)	Image count				Image retrieval
		R@1	R@5	R@10	Med r	R@1	R@5	R@10	Med r
DeViSE [13]	Word2Vec	4.8	16.5	27.3	28	5.9	20.1	29.6	29
DeFrag [14]	R–CNN	12.6	32.9	44.0	14	9.7	29.6	42.5	15
VSA [16]	R–CNN + BRNN	16.5	40.6	54.2	7.6	11.8	32.1	44.7	12.4
UVSE [18]	ConvNet + LSTM	13.5	36.2	45.7	13	10.4	31	43.7	14
UVSE (VGG) [19]	VGG + LSTM	18.0	40.9	55.0	8	12.5	37.0	51.5	10
VSE++ [21]	VGG + GRU + HNM	16.3	37.7	52.5	9	12	33.3	48.1	11
Ours	Aug	20.9	44.1	58.8	7	14.5	39	51.2	10
Ours	Aug + Word2Vec	21.5	49.1	62.3	6	15.1	38.9	53.1	9