Image-Text Retrieval on Urban 1k long-text
88.6T2I Recall@1GoldiCLIP
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| GoldiCLIPData size=30M, Evaluation setting=Zero-shot2026.03 | 88.6 | 83.2 | |
| COSMOSData size=30M, Evaluation setting=Zero-shot2026.03 | 88 | 82.7 | |
| FLAIRData size=30M, Evaluation setting=Zero-shot2026.03 | 87.7 | 83.6 | |
| Long-CLIPData size=400M, Evaluation setting=Zero-shot2026.03 | 79.5 | 78.9 | |
| GoldiCLIPData size=3M, Evaluation setting=Zero-shot2026.03 | 79.3 | 72.3 | |
| COSMOSData size=3M, Evaluation setting=Zero-shot2026.03 | 73.2 | 67.1 | |
| FLAIRData size=3M, Evaluation setting=Zero-shot2026.03 | 69.5 | 63.5 | |
| OpenCLIPData size=2B, Evaluation setting=Zero-shot2026.03 | 65.8 | 69.5 | |
| SigLIPData size=10B, Evaluation setting=Zero-shot2026.03 | 62.1 | 62.7 | |
| SigLIP 2Data size=10B, Evaluation setting=Zero-shot2026.03 | 57.8 | 60.3 | |
| PEcoreData size=5B, Evaluation setting=Zero-shot2026.03 | 56.7 | 60.3 |