Global-shape recognition on Shape-Biased Stimuli modified ImageNet (val)
10.8Accuracy DifferenceCLIP
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| CLIPPre-training source=Image, Evaluation protocol=Linear probing2022.10 | 10.8 | 61.2 | 48.2 | |
| VITOPre-training source=Video, Evaluation protocol=Linear probing2022.10 | 15.7 | 56.4 | 42.2 | |
| L2-RobustPre-training source=Image, Evaluation protocol=Linear probing2022.10 | 17.8 | 54.4 | 38.9 | |
| MoCLRPre-training source=Image, Evaluation protocol=Linear probing2022.10 | 19 | 53.6 | 33.5 | |
| SIN+IN1KPre-training source=Image, Evaluation protocol=Linear probing2022.10 | 20.3 | 52.7 | 33 | |
| VINCEPre-training source=Video, Evaluation protocol=Linear probing2022.10 | 21 | 50.1 | 26.9 | |
| SupervisedPre-training source=Image, Evaluation protocol=Linear probing2022.10 | 21.5 | 51.1 | 32.9 | |
| DINOPre-training source=Image, Evaluation protocol=Linear probing2022.10 | 23.6 | 50.4 | 29.1 | |
| CycleConPre-training source=Video, Evaluation protocol=Linear probing2022.10 | 23.7 | 48.4 | 25.8 | |
| R3MPre-training source=Video, Evaluation protocol=Linear probing2022.10 | 39.2 | 35.9 | 5.4 |