全局学习扩展的可见光-红外行人重识别

郭子强,肖璇,陶浩然,王少荣

Global learning-expanded visible-infrared person re-identification

Ziqiang GUO,Xuan XIAO,Haoran TAO,Shaorong WANG

表 2 RegDB数据集上GLE与先进方法的性能比较

Tab.2 Performance comparison of GLE and state-of-the-art methods on RegDB dataset

方法 可见光检索红外 红外检索可见光 R-1/% R-10/% R-20/% mAP/% R-1/% R-10/% R-20/% mAP/% DART[30] 83.6 — — 75.7 82.0 — — 73.8 CAJ[31] 85.0 95.5 97.5 79.1 84.8 95.3 97.5 77.8 MPANet[32] 82.8 — — 80.7 83.7 — — 80.9 MMN[33] 91.6 97.7 98.9 84.1 87.5 96.0 98.1 80.5 DCLNet[34] 81.2 — — 74.3 78.0 — — 70.6 MAUM[4] 87.9 — — 85.1 87.0 — — 84.3 DEEN[16] 91.1 97.8 98.9 85.1 89.5 96.8 98.4 83.4 HOS-Net[35] 94.7 — — 90.4 93.3 — — 89.2 SAAI[36] 91.1 — — 91.5 92.1 — — 92.0 CMT[21] 95.2 98.8 — 87.3 92.0 97.9 99.1 84.5 MUN[37] 95.2 98.9 — 87.2 91.9 98.0 — 85.0 IDKL[18] 94.7 — — 90.2 94.2 — — 90.4 GLE 94.9 98.8 99.6 90.5 90.2 98.0 99.1 89.2