Table 10 Effect of adversarial training on retrieval performance for normal and adversarial queries across different backbone networks and Hashcode lengths on the K-CT dataset.

Backbone Network	Query Type	Adversarial Training	Hashcode Length
Backbone Network	Query Type	Adversarial Training	8 bits	16 bits	32 bits	64 bits
AlexNet	Normal	No	0.91	0.95	0.93	0.91
	Adversary	No	0.09	0.14	0.10	0.11
	Normal	Yes	0.81	0.86	0.86	0.80
	Adversary	Yes	0.81	0.82	0.75	0.72
VGG 16	Normal	No	0.89	0.92	0.92	0.90
	Adversary	No	0.11	0.12	0.16	0.13
	Normal	Yes	0.82	0.86	0.91	0.89
	Adversary	Yes	0.81	0.85	0.81	0.75
DenseNet 121	Normal	No	0.90	0.92	0.89	0.86
	Adversary	No	0.12	0.17	0.19	0.20
	Normal	Yes	0.89	0.90	0.87	0.86
	Adversary	Yes	0.86	0.89	0.83	0.77
DeneNet 201	Normal	No	0.91	0.94	0.90	0.89
	Adversary	No	0.11	0.11	0.24	0.26
	Normal	Yes	0.90	0.92	0.88	0.87
	Adversary	Yes	0.88	0.88	0.85	0.78
ConvNeXt (Ours)	Normal	No	0.93	0.95	0.92	0.91
	Adversary	No	0.11	0.11	0.11	0.17
	Normal	Yes	0.92	0.94	0.92	0.91
	Adversary	Yes	0.85	0.89	0.83	0.79

Quick links

Search