Skip to main content

Table 7 Comparison of network accuracy (mAP and rank-1) and complexity (M: memory (parameters) and time: T (GFLOPS)) with different pruning compression ratios of different scenarios on all the ReID datasets

From: Exploiting prunability for person re-identification

Datasets Scenario M T L1 Auto-Balanced PSFP
  (compression)    mAP rank-1 mAP rank-1 mAP rank-1
Market-1501 1 (C1) 11.90 2.96 67.04 84.71 65.46 83.64 65.92 83.72
  2 (C1) 11.90 2.96 47.46 69.36 47.57 70.46 65.55 82.69
  3 (C1) 11.90 2.96 53.22 72.21 54.73 74.05 65.88 82.19
  1 (C2) 8.95 2.09 63.63 81.95 63.16 81.29 63.58 82.47
  2 (C2) 8.95 2.09 49.18 70.67 47.36 68.74 65.03 80.85
  3 (C2) 8.95 2.09 41.93 62.62 48.30 69.00 65.88 82.91
Duke-MTMC 1 (C1) 11.90 2.96 57.51 75.00 56.64 74.64 56.96 74.66
  2 (C1) 11.90 2.96 40.60 59.47 41.09 61.09 53.71 71.90
  3 (C1) 11.90 2.96 46.88 65.93 45.54 66.16 56.62 74.09
  1 (C2) 8.95 2.09 55.35 74.69 55.10 72.94 55.22 73.57
  2 (C2) 8.95 2.09 40.91 61.67 43.19 64.14 53.05 73.20
  3 (C2) 8.95 2.09 39.17 58.71 34.80 54.58 56.77 73.38
CUHK03-NP 1 (C1) 11.90 2.96 44.08 46.50 41.85 44.21 42.38 45.58
  2 (C1) 11.90 2.96 27.23 28.43 27.44 29.57 40.47 45.00
  3 (C1) 11.90 2.96 33.57 36.07 33.34 35.29 40.66 44.57
  1 (C2) 8.95 2.09 37.83 39.71 38.51 40.57 38.76 40.52
  2 (C2) 8.95 2.09 26.29 27.36 26.69 27.36 42.19 43.86
  3 (C2) 8.95 2.09 26.79 28.29 26.50 27.14 40.31 40.14