Cross-domain few-shot classification on CD-FSL benchmark
79.5Mean AccuracyFlorence
Evaluation results
| Method | Links | |||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| FlorenceShots=50-shot2021.11 | 79.5 | 78.3 | 97.1 | 99.6 | 42.8 | — | — | — | — | — | — | — | — | |
| CWShots=50-shot2021.11 | 78.8 | 74.1 | 96.9 | 99.7 | 44.4 | — | — | — | — | — | — | — | — | |
| FlorenceShots=20-shot2021.11 | 76.4 | 72.9 | 95.8 | 99.3 | 37.5 | — | — | — | — | — | — | — | — | |
| CWShots=20-shot2021.11 | 75.1 | 68.1 | 94.7 | 99.2 | 38.3 | — | — | — | — | — | — | — | — | |
| FlorenceShots=5-shot2021.11 | 68.5 | 57.1 | 90 | 97.7 | 29.3 | — | — | — | — | — | — | — | — | |
| CWShots=5-shot2021.11 | 68 | 57.4 | 88.1 | 96.6 | 29.7 | — | — | — | — | — | — | — | — | |
| P>M>FBackbone=ViT-small, Pre-training=DINO/IN1K, Few-shot protocol (Way/Shot)=5w50s2022.04 | 41.39 | — | — | — | — | — | — | — | — | — | — | — | — | |
| STARTUPBackbone=RN10, Few-shot protocol (Way/Shot)=5w50s2022.04 | 36.91 | — | — | — | — | — | — | — | — | — | — | — | — | |
| FinetuneBackbone=RN10, Few-shot protocol (Way/Shot)=5w50s2022.04 | 35.49 | — | — | — | — | — | — | — | — | — | — | — | — | |
| P>M>FBackbone=ViT-small, Pre-training=DINO/IN1K, Few-shot protocol (Way/Shot)=5w20s2022.04 | 35.33 | — | — | — | — | — | — | — | — | — | — | — | — | |
| DeepCluster2Backbone=RN50, Pre-training=IN1K, Few-shot protocol (Way/Shot)=5w50s2022.04 | 34.17 | — | — | — | — | — | — | — | — | — | — | — | — | |
| P>M>FBackbone=ResNet50, Pre-training=DINO/IN1K, Few-shot protocol (Way/Shot)=5w50s2022.04 | 34.17 | — | — | — | — | — | — | — | — | — | — | — | — | |
| STARTUPBackbone=RN10, Few-shot protocol (Way/Shot)=5w20s2022.04 | 33.19 | — | — | — | — | — | — | — | — | — | — | — | — | |
| P>M>FBackbone=ResNet50, Pre-training=DINO/IN1K, Few-shot protocol (Way/Shot)=5w20s2022.04 | 31.57 | — | — | — | — | — | — | — | — | — | — | — | — | |
| DeepCluster2Backbone=RN50, Pre-training=IN1K, Few-shot protocol (Way/Shot)=5w20s2022.04 | 31.51 | — | — | — | — | — | — | — | — | — | — | — | — | |
| FinetuneBackbone=RN10, Few-shot protocol (Way/Shot)=5w20s2022.04 | 31.32 | — | — | — | — | — | — | — | — | — | — | — | — | |
| CHEFBackbone=RN10, Few-shot protocol (Way/Shot)=5w50s2022.04 | 31.25 | — | — | — | — | — | — | — | — | — | — | — | — | |
| CHEFBackbone=RN10, Few-shot protocol (Way/Shot)=5w20s2022.04 | 29.71 | — | — | — | — | — | — | — | — | — | — | — | — | |
| MetaOptNetBackbone=RN10, Few-shot protocol (Way/Shot)=5w50s2022.04 | 29.35 | — | — | — | — | — | — | — | — | — | — | — | — | |
| ProtoNetBackbone=RN10, Few-shot protocol (Way/Shot)=5w50s2022.04 | 29.32 | — | — | — | — | — | — | — | — | — | — | — | — | |
| RelationNetBackbone=RN10, Few-shot protocol (Way/Shot)=5w50s2022.04 | 28.45 | — | — | — | — | — | — | — | — | — | — | — | — | |
| ProtoNetBackbone=RN10, Few-shot protocol (Way/Shot)=5w20s2022.04 | 28.21 | — | — | — | — | — | — | — | — | — | — | — | — | |
| P>M>FBackbone=ViT-small, Pre-training=DINO/IN1K, Few-shot protocol (Way/Shot)=5w5s2022.04 | 27.27 | — | — | — | — | — | — | — | — | — | — | — | — | |
| P>M>FBackbone=ResNet50, Pre-training=DINO/IN1K, Few-shot protocol (Way/Shot)=5w5s2022.04 | 27.13 | — | — | — | — | — | — | — | — | — | — | — | — | |
| STARTUPBackbone=RN10, Few-shot protocol (Way/Shot)=5w5s2022.04 | 26.94 | — | — | — | — | — | — | — | — | — | — | — | — | |
| RelationNetBackbone=RN10, Few-shot protocol (Way/Shot)=5w20s2022.04 | 26.63 | — | — | — | — | — | — | — | — | — | — | — | — | |
| DeepCluster2Backbone=RN50, Pre-training=IN1K, Few-shot protocol (Way/Shot)=5w5s2022.04 | 26.51 | — | — | — | — | — | — | — | — | — | — | — | — | |
| FinetuneBackbone=RN10, Few-shot protocol (Way/Shot)=5w5s2022.04 | 25.97 | — | — | — | — | — | — | — | — | — | — | — | — | |
| MetaOptNetBackbone=RN10, Few-shot protocol (Way/Shot)=5w20s2022.04 | 25.53 | — | — | — | — | — | — | — | — | — | — | — | — | |
| CHEFBackbone=RN10, Few-shot protocol (Way/Shot)=5w5s2022.04 | 24.72 | — | — | — | — | — | — | — | — | — | — | — | — | |
| ProtoNetBackbone=RN10, Few-shot protocol (Way/Shot)=5w5s2022.04 | 24.05 | — | — | — | — | — | — | — | — | — | — | — | — | |
| RelationNetBackbone=RN10, Few-shot protocol (Way/Shot)=5w5s2022.04 | 22.96 | — | — | — | — | — | — | — | — | — | — | — | — | |
| MetaOptNetBackbone=RN10, Few-shot protocol (Way/Shot)=5w5s2022.04 | 22.53 | — | — | — | — | — | — | — | — | — | — | — | — | |
| ATAWay=5, Shot=12022.03 | — | 33.21 | 61.35 | 67.47 | 22.1 | — | — | — | — | — | — | — | — | |
| ATAWay=5, Shot=52022.03 | — | 44.91 | 83.75 | 90.59 | 24.32 | — | — | — | — | — | — | — | — | |
| C³LRN-way=5, Backbone=Conv4, Pre-training=mini-ImageNet, Supervision=Self-Supervised2022.02 | — | — | — | — | — | 26 | 33.39 | 45.93 | 59.95 | 80.32 | 88.09 | 87.9 | 95.38 | |
| FWTWay=5, Shot=12022.03 | — | 31.58 | 62.36 | 66.36 | 22.04 | — | — | — | — | — | — | — | — | |
| FWTWay=5, Shot=52022.03 | — | 43.17 | 83.01 | 87.11 | 25.18 | — | — | — | — | — | — | — | — | |
| FWT+wave-SANWay=5, Shot=12022.03 | — | 33.09 | 65.5 | 69.65 | 22.39 | — | — | — | — | — | — | — | — | |
| FWT+wave-SANWay=5, Shot=52022.03 | — | 46 | 84.84 | 91.23 | 25.27 | — | — | — | — | — | — | — | — | |
| GNNWay=5, Shot=12022.03 | — | 32.02 | 63.69 | 64.48 | 22 | — | — | — | — | — | — | — | — | |
| GNNWay=5, Shot=52022.03 | — | 43.94 | 83.64 | 87.96 | 25.27 | — | — | — | — | — | — | — | — | |
| GNN+wave-SANWay=5, Shot=12022.03 | — | 33.35 | 69.64 | 70.8 | 22.93 | — | — | — | — | — | — | — | — | |
| GNN+wave-SANWay=5, Shot=52022.03 | — | 44.93 | 85.22 | 89.7 | 25.63 | — | — | — | — | — | — | — | — | |
| LRPWay=5, Shot=12022.03 | — | 30.94 | 54.99 | 59.23 | 22.11 | — | — | — | — | — | — | — | — | |
| LRPWay=5, Shot=52022.03 | — | 44.14 | 77.14 | 86.15 | 24.53 | — | — | — | — | — | — | — | — | |
| MatchingNetWay=5, Shot=12022.03 | — | 29.46 | 50.67 | 48.47 | 20.91 | — | — | — | — | — | — | — | — | |
| MatchingNetWay=5, Shot=52022.03 | — | 34.38 | 66.8 | 58.73 | 22.56 | — | — | — | — | — | — | — | — | |
| Meta-FDMixupWay=5, Shot=12022.03 | — | 32.48 | 62.97 | 66.23 | 22.26 | — | — | — | — | — | — | — | — | |
| Meta-FDMixupWay=5, Shot=52022.03 | — | 44.28 | 80.48 | 87.27 | 24.52 | — | — | — | — | — | — | — | — | |
| Pre+Linear (sup.)N-way=5, Backbone=Conv4, Pre-training=mini-ImageNet, Supervision=Supervised2022.02 | — | — | — | — | — | 25.97 | 31.32 | 48.11 | 59.31 | 79.08 | 87.64 | 89.25 | 95.51 | |
| Pre+Mean-Cent. (sup.)N-way=5, Backbone=Conv4, Pre-training=mini-ImageNet, Supervision=Supervised2022.02 | — | — | — | — | — | 26.31 | 30.41 | 47.16 | 56.4 | 82.21 | 87.62 | 87.61 | 93.87 | |
| ProtoNet (sup.)N-way=5, Backbone=Conv4, Pre-training=mini-ImageNet, Supervision=Supervised2022.02 | — | — | — | — | — | 24.05 | 28.21 | 39.57 | 49.5 | 73.29 | 82.27 | 79.72 | 88.15 | |
| ProtoTransferN-way=5, Backbone=Conv4, Pre-training=mini-ImageNet, Supervision=Unsupervised2022.02 | — | — | — | — | — | 26.71 | 33.82 | 45.19 | 59.07 | 75.62 | 86.8 | 86.53 | 95.06 | |
| RelationNetWay=5, Shot=12022.03 | — | 29.69 | 56.28 | 56.18 | 21.94 | — | — | — | — | — | — | — | — | |
| RelationNetWay=5, Shot=52022.03 | — | 37.07 | 68.08 | 75.33 | 24.34 | — | — | — | — | — | — | — | — | |
| UMTRA-ProtoNetN-way=5, Backbone=Conv4, Pre-training=mini-ImageNet, Supervision=Unsupervised2022.02 | — | — | — | — | — | 24.94 | 28.04 | 39.21 | 44.62 | 74.91 | 80.42 | 79.81 | 86.84 | |
| UMTRA-ProtoTuneN-way=5, Backbone=Conv4, Pre-training=mini-ImageNet, Supervision=Unsupervised2022.02 | — | — | — | — | — | 25 | 30.41 | 38.47 | 51.6 | 68.11 | 81.56 | 82.67 | 92.04 |