Object Recognition on Office (standard)
95.7Accuracy (A to W)CDAN+TransNorm
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| CDAN+TransNormBackbone=ResNet-502020.02 | 95.7 | 94 | 73.4 | 98.7 | 74.2 | 100 | 89.3 | |
| rRevGrad+CATBackbone=ResNet-502020.02 | 94.4 | 90.8 | 72.2 | 98 | 70.2 | 100 | 87.6 | |
| CDAN+EBackbone=ResNet-502020.02 | 94.1 | 92.9 | 71 | 98.6 | 69.3 | 100 | 87.7 | |
| CDAN+BSPBackbone=ResNet-502020.02 | 93.3 | 93 | 73.6 | 98.2 | 72.6 | 100 | 88.5 | |
| SHOT-IMBackbone=ResNet-50, beta=02020.02 | 91.2 | 90.6 | 72.5 | 98.3 | 71.4 | 99.9 | 87.3 | |
| SAFN+ENTBackbone=ResNet-502020.02 | 90.1 | 90.7 | 73 | 98.6 | 70.2 | 99.8 | 87.1 | |
| SHOTBackbone=ResNet-50, beta=0.32020.02 | 90.1 | 94 | 74.7 | 98.4 | 74.3 | 99.9 | 88.6 | |
| DANNBackbone=ResNet-502020.02 | 82 | 79.7 | 68.2 | 96.9 | 67.4 | 99.1 | 82.2 | |
| DANBackbone=ResNet-502020.02 | 80.5 | 78.6 | 63.6 | 97.1 | 62.8 | 99.6 | 80.4 | |
| Source model onlyBackbone=ResNet-502020.02 | 76.9 | 80.8 | 60.3 | 95.3 | 63.6 | 98.7 | 79.3 | |
| ResNet-50Backbone=ResNet-502020.02 | 68.4 | 68.9 | 62.5 | 96.7 | 60.7 | 99.3 | 76.1 | |
| ReverseGrad2015.11 | 67.3 | — | — | 94 | — | 93.7 | — | |
| D-CORALBackbone=BVLC Reference CaffeNet, Loss=Correlation Alignment (CORAL), Adaptation Protocol=Unsupervised2016.07 | 66.4 | 66.8 | 52.8 | 95.7 | 51.5 | 99.2 | 72.1 | |
| D-CORALSetup=Deep architecture2016.12 | 66.4 | 66.8 | 52.8 | 95.7 | 51.5 | 99.2 | 72.1 | |
| DAN2015.11 | 66 | — | — | 93.5 | — | 95.3 | — | |
| CORALFeature=fc7 fine-tuned, Classifier=linear SVM, Backbone=BVLC Reference CaffeNet, Adaptation Protocol=Unsupervised2016.07 | 64.3 | 65.7 | 48.5 | 96.1 | 48.2 | 99.8 | 70.4 | |
| CORALSetup=Deep features2016.12 | 64.3 | 65.7 | 48.5 | 96.1 | 48.2 | 99.8 | 70.4 | |
| DANFeature layer=fc8, Adaptation=Multi-kernel selection, Backbone=BVLC Reference CaffeNet2016.07 | 63.8 | 65.8 | 52.8 | 94.6 | 51.9 | 98.8 | 71.3 | |
| DANSetup=Deep architecture2016.12 | 63.8 | 65.8 | 52.8 | 94.6 | 51.9 | 98.8 | 71.3 | |
| CORALFeature Layer=fc7, Training Mode=Fine-tuned2015.11 | 61.9 | 62.2 | 48.4 | 96.2 | 48.2 | 99.5 | 69.4 | |
| DDCBackbone=AlexNet, Adaptation Loss=Domain confusion, Fine-tuned=Source and Target2016.07 | 61.8 | 64.4 | 52.1 | 95 | 52.2 | 98.5 | 70.6 | |
| DDCSetup=Deep architecture2016.12 | 61.8 | 64.4 | 52.1 | 95 | 52.2 | 98.5 | 70.6 | |
| CNNAdaptation=No adaptation, Backbone=BVLC Reference CaffeNet2016.07 | 61.6 | 63.8 | 51.1 | 95.4 | 49.8 | 99 | 70.1 | |
| CNNSetup=Deep architecture2016.12 | 61.6 | 63.8 | 51.1 | 95.4 | 49.8 | 99 | 70.1 | |
| CORALFeature Layer=fc6, Training Mode=Fine-tuned2015.11 | 59.8 | 61.2 | 47.4 | 97.1 | 45.8 | 99.5 | 68.5 | |
| DDC2015.11 | 59.4 | — | — | 92.5 | — | 91.7 | — | |
| GFKFeature=fc7 fine-tuned, Classifier=linear SVM, Backbone=BVLC Reference CaffeNet, Adaptation Protocol=Unsupervised2016.07 | 54.7 | 52.4 | 43.2 | 92.1 | 41.8 | 96.2 | 63.4 | |
| GFKSetup=Deep features2016.12 | 54.7 | 52.4 | 43.2 | 92.1 | 41.8 | 96.2 | 63.4 | |
| DECAFFeature Layer=fc72015.11 | 53.9 | — | — | 89.2 | — | — | — | |
| CORALFeature Layer=fc7, Training Mode=Pre-trained2015.11 | 53.1 | 57.1 | 51.1 | 94.6 | 47.3 | 98.2 | 66.9 | |
| NAFeature Layer=fc7, Training Mode=Fine-tuned2015.11 | 53 | 58.5 | 43.8 | 94.8 | 43.7 | 99.1 | 65.5 | |
| GFKFeature Layer=fc7, Training Mode=Fine-tuned2015.11 | 52.3 | 56.4 | 43.2 | 92.2 | 41.5 | 96.6 | 63.7 | |
| DECAFFeature Layer=fc62015.11 | 52.2 | — | — | 91.5 | — | — | — | |
| NAFeature Layer=fc7, Training Mode=Pre-trained2015.11 | 50.6 | 55.7 | 46.5 | 93.1 | 43 | 97.4 | 64.4 | |
| NAFeature Layer=fc6, Training Mode=Pre-trained2015.11 | 48.6 | 53.2 | 40.5 | 92.9 | 39 | 98.8 | 62.2 | |
| CORALFeature Layer=fc6, Training Mode=Pre-trained2015.11 | 48.4 | 53.7 | 44.4 | 96.5 | 41.9 | 99.2 | 64 | |
| GFKFeature Layer=fc7, Training Mode=Pre-trained2015.11 | 48.2 | 52 | 41.8 | 86.5 | 38.6 | 87.5 | 59.1 | |
| NAFeature Layer=fc6, Training Mode=Fine-tuned2015.11 | 48 | 54.5 | 38.9 | 91.2 | 40.7 | 98.9 | 62 | |
| SAFeature=fc7 fine-tuned, Classifier=linear SVM, Backbone=BVLC Reference CaffeNet, Adaptation Protocol=Unsupervised2016.07 | 47.4 | 50.6 | 39.5 | 89.1 | 37.6 | 93.8 | 59.7 | |
| SASetup=Deep features2016.12 | 47.4 | 50.6 | 39.5 | 89.1 | 37.6 | 93.8 | 59.7 | |
| SAFeature Layer=fc7, Training Mode=Fine-tuned2015.11 | 47.2 | 50.5 | 39.6 | 89 | 37.3 | 93 | 59.4 | |
| GFKFeature Layer=fc6, Training Mode=Fine-tuned2015.11 | 45.6 | 48.8 | 40.5 | 90.4 | 36.7 | 96.3 | 59.7 | |
| TCAFeature=fc7 fine-tuned, Classifier=linear SVM, Backbone=BVLC Reference CaffeNet, Adaptation Protocol=Unsupervised2016.07 | 45.5 | 46.8 | 36.4 | 81.1 | 39.5 | 92.2 | 56.9 | |
| TCASetup=Deep features2016.12 | 45.5 | 46.8 | 36.4 | 81.1 | 39.5 | 92.2 | 56.9 | |
| TCAFeature Layer=fc7, Training Mode=Fine-tuned2015.11 | 45.2 | 47.3 | 36.4 | 80.9 | 39.2 | 92 | 56.8 | |
| SAFeature Layer=fc7, Training Mode=Pre-trained2015.11 | 42.5 | 46.2 | 39.3 | 78.9 | 36.3 | 80.6 | 54 | |
| SAFeature Layer=fc6, Training Mode=Fine-tuned2015.11 | 41.1 | 40.5 | 33.8 | 85.4 | 33.4 | 88.2 | 53.7 | |
| TCAFeature Layer=fc7, Training Mode=Pre-trained2015.11 | 40.5 | 45.4 | 36.5 | 78.2 | 34.1 | 84 | 53.1 | |
| GFKFeature Layer=fc6, Training Mode=Pre-trained2015.11 | 37.8 | 44.8 | 34.8 | 81 | 31.4 | 86.9 | 49.1 | |
| TCAFeature Layer=fc6, Training Mode=Fine-tuned2015.11 | 37.2 | 40.8 | 30.6 | 79.5 | 36.7 | 91.8 | 52.8 | |
| TCAFeature Layer=fc6, Training Mode=Pre-trained2015.11 | 36.8 | 40.6 | 32.9 | 82.3 | 28.9 | 84.1 | 50.9 | |
| SAFeature Layer=fc6, Training Mode=Pre-trained2015.11 | 35 | 41.3 | 32.3 | 74.5 | 30.1 | 81.5 | 49.1 | |
| DANN2015.11 | 34.1 | 34 | 20.1 | 62 | 21.2 | 64.4 | 39.3 | |
| DLID2015.11 | 26.1 | — | — | 68.9 | — | 84.9 | — | |
| DA-NBNN2015.11 | 23.3 | — | — | 67.2 | — | 67.4 | — |