Table 4.
Fine-tuning results compared with state-of-the-art methods on the ALL dataset.
| Author | Method | Batch Size | Labelled Ratio | Precision | Recall | F1-score | Accuracy |
|---|---|---|---|---|---|---|---|
| Chen et al. [6] |
SimCLR (repro.) |
256 |
1% | 54.2% | 50.6% | 45.8% | 50.6% |
| 10% | 62.1% | 52.0% | 50.7% | 55.2% | |||
| 25% | 58.3% | 47.1% | 43.7% | 47.1% | |||
| 50% |
54.6% |
43.6% |
38.3% |
42.3% |
|||
| 500 |
1% | 53.8% | 40.1% | 35.6% | 41.0% | ||
| 10% | 56.7% | 44.8% | 42.5% | 46.0% | |||
| 25% | 56.7% | 45.3% | 43.2% | 46.3% | |||
| 50% | 68.1% | 40.3% | 37.4% | 43.9% | |||
| Chen and He [9] |
SimSiam (repro.) |
256 |
1% | 75.4% | 73.2% | 73.9% | 75.9% |
| 10% | 91.2% | 92.4% | 91.5% | 92.0% | |||
| 25% | 97.4% | 98.2% | 97.8% | 98.0% | |||
| 50% |
98.8% |
98.9% |
98.9% |
98.9% |
|||
| 500 |
1% | 81.9% | 77.1% | 77.5% | 79.7% | ||
| 10% | 94.7% | 94.6% | 94.6% | 95.1% | |||
| 25% | 97.0% | 97.3% | 97.2% | 97.4% | |||
| 50% | 98.2% | 98.3% | 98.2% | 98.3% | |||
| Ren et al. [51] |
UKSSL (repro.) |
256 |
1% | 6.9% | 25.0% | 10.8% | 27.6% |
| 10% | 82.1% | 81.1% | 81.4% | 84.7% | |||
| 25% | 89.2% | 86.2% | 83.6% | 86.2% | |||
| 50% |
89.8% |
86.5% |
84.2% |
86.5% |
|||
| 500 |
1% | 59.1% | 68.6% | 63.4% | 76.4% | ||
| 10% | 82.4% | 80.9% | 81.4% | 82.8% | |||
| 25% | 83.4% | 82.2% | 82.7% | 84.7% | |||
| 50% | 97.0% | 93.0% | 94.4% | 95.4% | |||
| Ours | SimTrip | 256 | 1% | 88.6% | 85.8% | 86.7% | 88.0% |
| 10% | 95.5% | 94.3% | 94.9% | 95.2% | |||
| 25% | 99.5% | 99.6% | 99.5% | 99.5% | |||
| 50% | 99.6% | 99.7% | 99.7% | 99.7% | |||