deep-person-reid/docs/MODEL_ZOO.md

4.9 KiB

Model Zoo

  • Results are presented in the format of Rank-1 (mAP).
  • Classification layer is ignored when computing the model size.
  • Unless specified otherwise, only Random2DTranslation and RandomHorizontalFlip are used for data augmentation.
  • Asterisk (*) means the model is trained from scratch.
  • Model weights can be downloaded by clicking the highlighted numbers.
Model # param (M) Loss Input market1501 dukemtmcreid msmt17
resnet50 23.5 softmax (256, 128) 87.9 (70.4) 78.3 (58.9) 63.2 (33.9)
resnet50_fc512 24.6 softmax (256, 128) 90.8 (75.3) 81.0 (64.0) 69.6 (38.4)
densenet121_fc512 7.5 softmax (256, 128) 87.8 (68.0) 79.7 (58.8) 67.6 (35.0)
se_resnet50_fc512 27.1 softmax (256, 128) 91.9 (75.8) 81.5 (63.7) 71.1 (39.8)
shufflenet 0.9 softmax (256, 128) 84.1(64.1) 73.4(51.9) 51.3(24.2)
squeezenet1_0_fc512 1.0 softmax (256, 128) 79.3 (52.2) 66.6 (42.6) 44.1 (17.1)
mobilenetv2_1dot0 2.2 softmax (256, 128) 85.6 (67.3) 74.2 (54.7) 57.4 (29.3)
mobilenetv2_1dot4 4.3 softmax (256, 128) 87.0 (68.5) 76.2 (55.8) 60.1 (31.5)
resnet50mid 27.7 softmax (256, 128) 90.2 (76.0) 81.6 (64.0) 69.0 (38.0)
mlfn 32.5 softmax (256, 128) 90.1 (74.3) 81.1 (63.2) 66.4 (37.2)
hacnn* 3.7 softmax (160, 64) 90.9 (75.6) 80.1 (63.2) 64.7 (37.2)