Image Classification on BigEarthNet (val)
87.81mAPSeCo
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| SeCoBackbone=ResNet-50, Unlabeled pre-training set size=1M, Evaluation protocol=Fine-tuning, Supervised training set=100%2021.03 | 87.81 | — | — | |
| SeCoBackbone=ResNet-18, Unlabeled pre-training set size=1M, Evaluation protocol=Fine-tuning, Supervised training set=100%2021.03 | 87.27 | — | — | |
| SeCoBackbone=ResNet-50, Unlabeled pre-training set size=100k, Evaluation protocol=Fine-tuning, Supervised training set=100%2021.03 | 87.12 | — | — | |
| SeCoBackbone=ResNet-18, Unlabeled pre-training set size=100k, Evaluation protocol=Fine-tuning, Supervised training set=100%2021.03 | 87.04 | — | — | |
| ImageNet (sup.)Backbone=ResNet-50, Unlabeled pre-training set size=100k, Evaluation protocol=Fine-tuning, Supervised training set=100%2021.03 | 86.74 | — | — | |
| MoCo-v2+TPBackbone=ResNet-18, Unlabeled pre-training set size=1M, Evaluation protocol=Fine-tuning, Supervised training set=100%2021.03 | 86.59 | — | — | |
| MoCo-v2+TPBackbone=ResNet-50, Unlabeled pre-training set size=1M, Evaluation protocol=Fine-tuning, Supervised training set=100%2021.03 | 86.11 | — | — | |
| MoCo-v2Backbone=ResNet-50, Unlabeled pre-training set size=1M, Evaluation protocol=Fine-tuning, Supervised training set=100%2021.03 | 86.05 | — | — | |
| ImageNet (sup.)Backbone=ResNet-18, Unlabeled pre-training set size=100k, Evaluation protocol=Fine-tuning, Supervised training set=100%2021.03 | 85.9 | — | — | |
| MoCo-v2Backbone=ResNet-50, Unlabeled pre-training set size=100k, Evaluation protocol=Fine-tuning, Supervised training set=100%2021.03 | 85.79 | — | — | |
| MoCo-v2+TPBackbone=ResNet-18, Unlabeled pre-training set size=100k, Evaluation protocol=Fine-tuning, Supervised training set=100%2021.03 | 85.71 | — | — | |
| MoCo-v2+TPBackbone=ResNet-50, Unlabeled pre-training set size=100k, Evaluation protocol=Fine-tuning, Supervised training set=100%2021.03 | 85.35 | — | — | |
| MoCo-v2Backbone=ResNet-18, Unlabeled pre-training set size=1M, Evaluation protocol=Fine-tuning, Supervised training set=100%2021.03 | 85.23 | — | — | |
| MoCo-v2Backbone=ResNet-18, Unlabeled pre-training set size=100k, Evaluation protocol=Fine-tuning, Supervised training set=100%2021.03 | 85.17 | — | — | |
| SeCoBackbone=ResNet-50, Unlabeled pre-training set size=1M, Evaluation protocol=Fine-tuning, Supervised training set=10%2021.03 | 82.62 | — | — | |
| SeCoBackbone=ResNet-18, Unlabeled pre-training set size=1M, Evaluation protocol=Fine-tuning, Supervised training set=10%2021.03 | 81.86 | — | — | |
| SeCoBackbone=ResNet-50, Unlabeled pre-training set size=100k, Evaluation protocol=Fine-tuning, Supervised training set=10%2021.03 | 81.72 | — | — | |
| SeCoBackbone=ResNet-18, Unlabeled pre-training set size=100k, Evaluation protocol=Fine-tuning, Supervised training set=10%2021.03 | 81.49 | — | — | |
| MoCo-v2+TPBackbone=ResNet-18, Unlabeled pre-training set size=1M, Evaluation protocol=Fine-tuning, Supervised training set=10%2021.03 | 80.68 | — | — | |
| SeCoBackbone=ResNet-50, Unlabeled pre-training set size=1M, Evaluation protocol=Linear probing, Supervised training set=100%2021.03 | 80.35 | — | — | |
| MoCo-v2+TPBackbone=ResNet-50, Unlabeled pre-training set size=1M, Evaluation protocol=Fine-tuning, Supervised training set=10%2021.03 | 80.2 | — | — | |
| MoCo-v2Backbone=ResNet-50, Unlabeled pre-training set size=1M, Evaluation protocol=Fine-tuning, Supervised training set=10%2021.03 | 80.08 | — | — | |
| ImageNet (sup.)Backbone=ResNet-50, Unlabeled pre-training set size=100k, Evaluation protocol=Fine-tuning, Supervised training set=10%2021.03 | 80.04 | — | — | |
| Random init.Backbone=ResNet-18, Unlabeled pre-training set size=100k, Evaluation protocol=Fine-tuning, Supervised training set=100%2021.03 | 79.8 | — | — | |
| MoCo-v2+TPBackbone=ResNet-18, Unlabeled pre-training set size=100k, Evaluation protocol=Fine-tuning, Supervised training set=10%2021.03 | 79.8 | — | — | |
| MoCo-v2Backbone=ResNet-50, Unlabeled pre-training set size=100k, Evaluation protocol=Fine-tuning, Supervised training set=10%2021.03 | 79.23 | — | — | |
| SeCoBackbone=ResNet-50, Unlabeled pre-training set size=100k, Evaluation protocol=Linear probing, Supervised training set=100%2021.03 | 79.13 | — | — | |
| MoCo-v2+TPBackbone=ResNet-50, Unlabeled pre-training set size=100k, Evaluation protocol=Fine-tuning, Supervised training set=10%2021.03 | 79.04 | — | — | |
| Random init.Backbone=ResNet-50, Unlabeled pre-training set size=100k, Evaluation protocol=Fine-tuning, Supervised training set=100%2021.03 | 78.98 | — | — | |
| ImageNet (sup.)Backbone=ResNet-18, Unlabeled pre-training set size=100k, Evaluation protocol=Fine-tuning, Supervised training set=10%2021.03 | 78.76 | — | — | |
| MoCo-v2Backbone=ResNet-18, Unlabeled pre-training set size=100k, Evaluation protocol=Fine-tuning, Supervised training set=10%2021.03 | 78.76 | — | — | |
| SeCoBackbone=ResNet-50, Unlabeled pre-training set size=1M, Evaluation protocol=Linear probing, Supervised training set=10%2021.03 | 78.56 | — | — | |
| MoCo-v2Backbone=ResNet-18, Unlabeled pre-training set size=1M, Evaluation protocol=Fine-tuning, Supervised training set=10%2021.03 | 78.33 | — | — | |
| SeCoBackbone=ResNet-50, Unlabeled pre-training set size=100k, Evaluation protocol=Linear probing, Supervised training set=10%2021.03 | 77.49 | — | — | |
| SeCoBackbone=ResNet-18, Unlabeled pre-training set size=1M, Evaluation protocol=Linear probing, Supervised training set=100%2021.03 | 77 | — | — | |
| MoCo-v2+TPBackbone=ResNet-50, Unlabeled pre-training set size=1M, Evaluation protocol=Linear probing, Supervised training set=100%2021.03 | 76.32 | — | — | |
| SeCoBackbone=ResNet-18, Unlabeled pre-training set size=1M, Evaluation protocol=Linear probing, Supervised training set=10%2021.03 | 76.05 | — | — | |
| MoCo-v2Backbone=ResNet-50, Unlabeled pre-training set size=1M, Evaluation protocol=Linear probing, Supervised training set=100%2021.03 | 75.65 | — | — | |
| SeCoBackbone=ResNet-18, Unlabeled pre-training set size=100k, Evaluation protocol=Linear probing, Supervised training set=100%2021.03 | 75.52 | — | — | |
| InDomainReference=Ours, Fine-tuning=ImageNet2019.11 | 75.36 | — | — | |
| SeCoBackbone=ResNet-18, Unlabeled pre-training set size=100k, Evaluation protocol=Linear probing, Supervised training set=10%2021.03 | 74.67 | — | — | |
| MoCo-v2+TPBackbone=ResNet-50, Unlabeled pre-training set size=1M, Evaluation protocol=Linear probing, Supervised training set=10%2021.03 | 74.5 | — | — | |
| MoCo-v2+TPBackbone=ResNet-50, Unlabeled pre-training set size=100k, Evaluation protocol=Linear probing, Supervised training set=100%2021.03 | 73.91 | — | — | |
| MoCo-v2Backbone=ResNet-50, Unlabeled pre-training set size=1M, Evaluation protocol=Linear probing, Supervised training set=10%2021.03 | 73.71 | — | — | |
| MoCo-v2+TPBackbone=ResNet-18, Unlabeled pre-training set size=1M, Evaluation protocol=Linear probing, Supervised training set=100%2021.03 | 73.6 | — | — | |
| MoCo-v2Backbone=ResNet-50, Unlabeled pre-training set size=100k, Evaluation protocol=Linear probing, Supervised training set=100%2021.03 | 73.27 | — | — | |
| MoCo-v2+TPBackbone=ResNet-50, Unlabeled pre-training set size=100k, Evaluation protocol=Linear probing, Supervised training set=10%2021.03 | 72.61 | — | — | |
| MoCo-v2+TPBackbone=ResNet-18, Unlabeled pre-training set size=1M, Evaluation protocol=Linear probing, Supervised training set=10%2021.03 | 72.58 | — | — | |
| MoCo-v2Backbone=ResNet-50, Unlabeled pre-training set size=100k, Evaluation protocol=Linear probing, Supervised training set=10%2021.03 | 71.85 | — | — | |
| ImageNet (sup.)Backbone=ResNet-50, Unlabeled pre-training set size=100k, Evaluation protocol=Linear probing, Supervised training set=100%2021.03 | 71.82 | — | — | |
| MoCo-v2+TPBackbone=ResNet-18, Unlabeled pre-training set size=100k, Evaluation protocol=Linear probing, Supervised training set=100%2021.03 | 71.08 | — | — | |
| MoCo-v2Backbone=ResNet-18, Unlabeled pre-training set size=100k, Evaluation protocol=Linear probing, Supervised training set=100%2021.03 | 70.9 | — | — | |
| MoCo-v2Backbone=ResNet-18, Unlabeled pre-training set size=1M, Evaluation protocol=Linear probing, Supervised training set=100%2021.03 | 70.79 | — | — | |
| ImageNet (sup.)Backbone=ResNet-50, Unlabeled pre-training set size=100k, Evaluation protocol=Linear probing, Supervised training set=10%2021.03 | 70.46 | — | — | |
| MoCo-v2+TPBackbone=ResNet-18, Unlabeled pre-training set size=100k, Evaluation protocol=Linear probing, Supervised training set=10%2021.03 | 70.2 | — | — | |
| MoCo-v2Backbone=ResNet-18, Unlabeled pre-training set size=100k, Evaluation protocol=Linear probing, Supervised training set=10%2021.03 | 69.7 | — | — | |
| Random init.Backbone=ResNet-50, Unlabeled pre-training set size=100k, Evaluation protocol=Fine-tuning, Supervised training set=10%2021.03 | 69.49 | — | — | |
| MoCo-v2Backbone=ResNet-18, Unlabeled pre-training set size=1M, Evaluation protocol=Linear probing, Supervised training set=10%2021.03 | 69.28 | — | — | |
| Random init.Backbone=ResNet-18, Unlabeled pre-training set size=100k, Evaluation protocol=Fine-tuning, Supervised training set=10%2021.03 | 68.11 | — | — | |
| ImageNet (sup.)Backbone=ResNet-18, Unlabeled pre-training set size=100k, Evaluation protocol=Linear probing, Supervised training set=100%2021.03 | 66.4 | — | — | |
| ImageNet (sup.)Backbone=ResNet-18, Unlabeled pre-training set size=100k, Evaluation protocol=Linear probing, Supervised training set=10%2021.03 | 65.69 | — | — | |
| Random init.Backbone=ResNet-50, Unlabeled pre-training set size=100k, Evaluation protocol=Linear probing, Supervised training set=100%2021.03 | 46.92 | — | — | |
| Random init.Backbone=ResNet-18, Unlabeled pre-training set size=100k, Evaluation protocol=Linear probing, Supervised training set=100%2021.03 | 45.95 | — | — | |
| Random init.Backbone=ResNet-50, Unlabeled pre-training set size=100k, Evaluation protocol=Linear probing, Supervised training set=10%2021.03 | 43.95 | — | — | |
| Random init.Backbone=ResNet-18, Unlabeled pre-training set size=100k, Evaluation protocol=Linear probing, Supervised training set=10%2021.03 | 43.05 | — | — | |
| Sumbul et al.Reference=Sumbul et al. (2019)2019.11 | — | 69.93 | 77.1 |