Image Classification on 12 Downstream Classification Tasks (test)
78.7Food-101 AccuracyDnC-4.5k
Evaluation results
| Method | Links | |||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| DnC-4.5kPre-training Dataset=JFT-300M, Evaluation Protocol=Linear Classification, Training Schedule=4,500 epochs2021.05 | 78.7 | 91.7 | 74.9 | 42.1 | 65 | 75.3 | 54.1 | 83.1 | 76.6 | 86.1 | 90.2 | 98.2 | 76.3 | |
| BYOL-5kPre-training Dataset=JFT-300M, Evaluation Protocol=Linear Classification, Training Schedule=5,000 epochs2021.05 | 73.3 | 89.8 | 72.4 | 38.2 | 61.8 | 64.4 | 54.4 | 81.3 | 75.5 | 77 | 90.1 | 94.3 | 72.7 | |
| MoCLR-5kPre-training Dataset=JFT-300M, Evaluation Protocol=Linear Classification, Training Schedule=5,000 epochs2021.05 | 72.8 | 90.7 | 72.5 | 33.8 | 62.2 | 60.6 | 50.9 | 81.9 | 75.3 | 75.8 | 89.5 | 93.8 | 71.7 | |
| DnC-4.5kPre-training Dataset=YFCC100M, Evaluation Protocol=Linear Classification, Training Schedule=4,500 epochs2021.05 | 72.1 | 88 | 71.1 | 35.5 | 67.2 | 52.6 | 49.2 | 83.7 | 76.5 | 75.9 | 87 | 97.8 | 71.4 | |
| BYOL-5kPre-training Dataset=YFCC100M, Evaluation Protocol=Linear Classification, Training Schedule=5,000 epochs2021.05 | 69.1 | 85.8 | 66.8 | 35.5 | 64.1 | 50.1 | 51.9 | 82.5 | 74.5 | 74 | 87.6 | 95.8 | 69.8 | |
| MoCLR-5kPre-training Dataset=YFCC100M, Evaluation Protocol=Linear Classification, Training Schedule=5,000 epochs2021.05 | 68.4 | 87.6 | 69.7 | 30.5 | 63.9 | 41 | 46.7 | 82.4 | 76.2 | 68.5 | 86 | 93 | 67.8 |