Image Classification and Multimodal Retrieval on Average
77.3Clean AccuracyFuseLIP
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| FuseLIPTokenizer=clean, Evaluation Protocol=Zero-shot2026.02 | 77.3 | 4 | 0.6 | |
| FuseLIPTokenizer=AT 4/255, Evaluation Protocol=Zero-shot2026.02 | 74.6 | 52.1 | 27 | |
| FuseLIPTokenizer=AT 8/255, Evaluation Protocol=Zero-shot2026.02 | 72.2 | 55.2 | 38.3 | |
| FuseLIPTokenizer=AT 12/255, Evaluation Protocol=Zero-shot2026.02 | 68.1 | 55.3 | 40 | |
| FuseLIPTokenizer=AT 16/255, Evaluation Protocol=Zero-shot2026.02 | 63.5 | 52.6 | 40.6 |