Cross-Modal Search for Social Networks via Adversarial Learning

Table 3

MAP evaluations on the NUS-WIDE dataset.

		Traditionally-based	DNN-based		GAN-based
		JFSSL	CMDN	DCCA	ACMR	CMGAN	CMSAL
MAP@5	txt2img	0.2464	0.4187	0.5110	0.6397	0.6872	0.8023
	img2txt	0.2197	0.3313	0.4682	0.4838	0.4797	0.6329
	average	0.1337	0.3908	0.5877	0.6884	0.6248	0.7474
MAP@20	txt2img	0.2511	0.5469	0.4792	0.6077	0.6245	0.6796
	img2txt	0.2231	0.4965	0.5050	0.5284	0.5042	0.5887
	average	0.2450	0.3699	0.4012	0.6249	0.5314	0.6640
MAP@50	txt2img	0.3015	0.4011	0.5670	0.6500	0.6301	0.6798
	img2txt	0.1001	0.4700	0.3910	0.4801	0.5466	0.4836
	average	0.2587	0.4697	0.5035	0.6621	0.6653	0.7976