- 🐶 means that model is initialized with imagenet pretrained weights.
- Results are presented using the format of
Rank1 (mAP)in the following tables. - CUHK03: detected, new protocol (767/700).
- Classification layer is ignored when computing model size.
| Model | # param (M) | Loss | Market1501 | CUHK03 | DukeMTMC-reID | MSMT17 |
|---|---|---|---|---|---|---|
| ResNet50:dog: | 23.5 | xent | 88.5 (71.3) | 47.9 (46.8) | 77.7 (58.8) | 63.4 (34.2) |
| DenseNet121:dog: | 7.0 | xent | 88.2 (69.2) | 41.0 (40.1) | 78.6 (58.5) | 66.0 (34.6) |
| DenseNet121:dog: | 7.0 | htri | 86.6 (70.8) | - | - | - |
| DenseNet121:dog: | 7.0 | xent+htri | 90.1 (74.0) | - | - | - |
| ResNet50M:dog: | 27.7 | xent | 89.0 (74.6) | 55.4 (52.7) | 81.0 (64.1) | 64.6 (35.9) |
| NasnetMobile:dog: | 4.2 | xent | 83.8 (64.9) | 42.4 (42.4) | 74.0 (53.7) | 57.1 (30.2) |
| SqueezeNet | 1.1 | xent | 72.2 (47.9) | 26.9 (25.8) | 58.8 (37.8) | 30.6 (13.0) |
| MobileNetV2 | 2.2 | xent | 84.2 (65.8) | 41.0 (40.3) | 73.2 (52.5) | 44.9 (21.1) |
| ShuffleNet | 0.9 | xent | 80.0 (58.4) | 31.9 (31.7) | 69.3 (46.8) | 39.6 (17.8) |
| HACNN | 2.9 | xent | 90.6 (75.3) | 48.0 (47.6) | 80.7 (64.4) | 61.8 (34.6) |