Skip to main content

Table 8 Comparison of network accuracy (mAP and rank-1) and complexity (M: Parameters and T: GFLOPS) with different pruning scenarios on all the person ReID datasets. BFE backbone feature extractor; LC, local convolutional layer; FC, fully connected layer

From: Exploiting prunability for person re-identification

   Market-1501 DukeMTMC-reID
Methods Backbone mAP rank-1 T M mAP rank-1 T M
PCB (Baseline)   77.3 92.4 6.1 27.2 65.3 81.9 6.1 27.2
PSFP + PCB (BFE) ResNet50 69.4 90.4 2.6 12.0 60.7 78.5 2.6 12.0
PSFP + PCB (BFE+ LC + FC)   69.1 89.1 2.6 11.16 59.4 77.8 2.6 11.16
PCB (Baseline)   77.3 91.7 5.9 27.7 65.3 81.2 5.9 27.7
PSFP + PCB (BFE) SE-ResNet50 70.0 88.4 2.59 14.5 61.9 78.9 2.59 14.5
PSFP + PCB (BFE+ LC + FC)   69.2 87.2 2.58 13.69 59.4 77.4 2.58 13.69