Object Detection on ILSVRC weakly labeled categories 2013 (val2)
29.46mAPOracle: Full Detection Network (BB reg.)
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Oracle: Full Detection Network (BB reg.)Backbone=AlexNet, Bounding-box regression=true2018.01 | 29.46 | — | |
| mixture transfer modelBackbone=ResNet-50, Adaptation Method=mixed, Bounding-box Regression=true2018.01 | 28.3 | — | |
| mixture transfer modelBackbone=ResNet-34, Adaptation Method=mixed, Bounding-box Regression=true2018.01 | 28.05 | — | |
| mixture transfer modelBackbone=VGG-19, Adaptation Method=mixed, Bounding-box Regression=true2018.01 | 27.32 | — | |
| mixture transfer modelBackbone=ResNet-50, Adaptation Method=mixed2018.01 | 26.77 | — | |
| mixture transfer modelBackbone=GoogLeNet, Adaptation Method=mixed, Bounding-box Regression=true2018.01 | 26.62 | — | |
| Oracle: Full Detection Network (no BB reg.)Backbone=AlexNet2018.01 | 26.25 | — | |
| mixture transfer modelBackbone=ResNet-34, Adaptation Method=mixed2018.01 | 26.17 | — | |
| mixture transfer modelBackbone=VGG-19, Adaptation Method=mixed2018.01 | 25.07 | — | |
| mixture transfer modelBackbone=GoogLeNet, Adaptation Method=mixed2018.01 | 25.06 | — | |
| visual similarity based methodBackbone=ResNet-50, Adaptation Method=visual2018.01 | 24.94 | — | |
| mixture transfer modelBackbone=VGG-16, Adaptation Method=mixed, Bounding-box Regression=true2018.01 | 24.91 | — | |
| semantic similarity based methodBackbone=ResNet-50, Adaptation Method=semantic2018.01 | 24.75 | — | |
| semantic similarity based methodBackbone=ResNet-34, Adaptation Method=semantic2018.01 | 24.64 | — | |
| visual similarity based methodBackbone=ResNet-34, Adaptation Method=visual2018.01 | 24.28 | — | |
| visual similarity based methodBackbone=VGG-19, Adaptation Method=visual2018.01 | 23.89 | — | |
| visual similarity based methodBackbone=GoogLeNet, Adaptation Method=visual2018.01 | 23.62 | — | |
| semantic similarity based methodBackbone=GoogLeNet, Adaptation Method=semantic2018.01 | 23.46 | — | |
| mixture transfer modelBackbone=VGG-16, Adaptation Method=mixed2018.01 | 23.21 | — | |
| semantic similarity based methodBackbone=VGG-19, Adaptation Method=semantic2018.01 | 23.1 | — | |
| mixture transfer modelBackbone=Alex-Net, Adaptation Method=mixed, Bounding-box Regression=true2018.01 | 21.88 | — | |
| Ours (mixture transfer + BB reg.)Backbone=AlexNet, Bounding-box regression=true2018.01 | 21.88 | — | |
| visual similarity based methodBackbone=VGG-16, Adaptation Method=visual2018.01 | 21.75 | — | |
| semantic similarity based methodBackbone=VGG-16, Adaptation Method=semantic2018.01 | 21.07 | — | |
| LSDABackbone=VGG-19, Adaptation Method=class invariant & specific2018.01 | 21.02 | — | |
| LSDABackbone=VGG-19, Adaptation Method=class invariant2018.01 | 20.38 | — | |
| mixture transfer modelBackbone=Alex-Net, Adaptation Method=mixed2018.01 | 20.03 | — | |
| Ours (mixture transfer)Backbone=AlexNet2018.01 | 20.03 | — | |
| Ours (semantic transfer)Backbone=AlexNet, Representation=Sparse rep. - <202018.01 | 19.04 | — | |
| visual similarity based methodBackbone=Alex-Net, Adaptation Method=visual2018.01 | 19.02 | — | |
| LSDABackbone=VGG-16, Adaptation Method=class invariant & specific2018.01 | 18.86 | — | |
| semantic similarity based methodBackbone=Alex-Net, Adaptation Method=semantic2018.01 | 18.32 | — | |
| LSDABackbone=VGG-16, Adaptation Method=class invariant2018.01 | 18.24 | — | |
| Ours (visual transfer)Backbone=AlexNet, Number of Nearest Neighbors=102018.01 | 17.62 | 18.41 | |
| Ours (visual transfer)Backbone=AlexNet, Number of Nearest Neighbors=52018.01 | 17.42 | 17.59 | |
| Ours (visual transfer)Backbone=AlexNet, Number of Nearest Neighbors=1002018.01 | 17.38 | 19.02 | |
| Only classificationBackbone=ResNet-50, Adaptation Method=None2018.01 | 17.34 | — | |
| Ours (semantic transfer)Backbone=AlexNet, Number of Nearest Neighbors=52018.01 | 17.32 | 17.53 | |
| Only classificationBackbone=ResNet-34, Adaptation Method=None2018.01 | 17.09 | — | |
| Ours (semantic transfer)Backbone=AlexNet, Number of Nearest Neighbors=1002018.01 | 17.04 | 18.32 | |
| Ours (semantic transfer)Backbone=AlexNet, Number of Nearest Neighbors=102018.01 | 16.67 | 17.5 | |
| LSDABackbone=Alex-Net, Adaptation Method=class invariant & specific2018.01 | 16.33 | — | |
| Only classificationBackbone=VGG-19, Adaptation Method=None2018.01 | 16.22 | — | |
| LSDA (class invariant & specific adapt)Backbone=AlexNet, Number of Nearest Neighbors=102018.01 | 16.15 | 16.28 | |
| Only classificationBackbone=GoogLeNet, Adaptation Method=None2018.01 | 16.12 | — | |
| LSDA (class invariant & specific adapt)Backbone=AlexNet, Number of Nearest Neighbors=52018.01 | 15.97 | 16.12 | |
| LSDA (class invariant & specific adapt)Backbone=AlexNet, Number of Nearest Neighbors=1002018.01 | 15.96 | 16.33 | |
| LSDABackbone=Alex-Net, Adaptation Method=class invariant2018.01 | 15.85 | — | |
| LSDA (only class invariant adaptation)Backbone=AlexNet2018.01 | 15.85 | — | |
| Only classificationBackbone=VGG-16, Adaptation Method=None2018.01 | 14.89 | — | |
| Only classificationBackbone=Alex-Net, Adaptation Method=None2018.01 | 10.31 | — | |
| Classification NetworkBackbone=AlexNet2018.01 | 10.31 | — |