Text-to-Image Retrieval on Retrieval VQA Visual Genome (test)
66.8Count AccuracyGoldiCLIP
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| GoldiCLIPData size=30M, Zero-shot=true2026.03 | 66.8 | 48.5 | 51.5 | 56.4 | 55.9 | 55.8 | |
| SigLIP 2Data size=10B, Zero-shot=true2026.03 | 66.6 | 49.1 | 54.7 | 56.7 | 57.7 | 57 | |
| PE_coreData size=5B, Zero-shot=true2026.03 | 65.1 | 67.4 | 48.9 | 52.4 | 57.5 | 52.9 | |
| SigLIPData size=10B, Zero-shot=true2026.03 | 64.1 | 46.5 | 51.1 | 53.2 | 53.5 | 53.5 | |
| OpenCLIPData size=2B, Zero-shot=true2026.03 | 62.1 | 45 | 47.7 | 51.3 | 48.5 | 50.9 | |
| COSMOSData size=30M, Zero-shot=true2026.03 | 60.7 | 42.4 | 46.1 | 51.6 | 48.9 | 49.9 | |
| FLAIRData size=30M, Zero-shot=true2026.03 | 57.6 | 39.1 | 44.4 | 49 | 47.7 | 47.6 |