Multilayer Convolutional Feature Aggregation Algorithm for Image Retrieval

Table 7

Comparison of commonly used image retrieval methods on two common datasets.

CNN-based Method	Dimension	Oxford5k	Paris6k
CNN-ss [8]	32-120k	55.6	69.7
Neural codes [9]	4096	54.5	38.6
OxfordNet [12]	256	53.3	67
SPoC [13]	256	58.9	-
CroW [14]	512	68.2	79.6
R-MAC [15]	512	66.9	83.0
ML-RCroW (one layer)	512	79.3	81.7
ML-CroW (aggregate two layers)	1024	77.3	82.1
ML-RCRoW (aggregate two layers)	1024	79.5	83.5
ML-CroW (aggregate three layers)	1280	76.9	81.9
ML-RCRoW (aggregate three layers)	1280	79.7	83.6