Thermal-to-Text Retrieval (via Vision pivot) on LLVIP to FLIR (test)
40.2mAPIB+BB
Evaluation Results
| Method | Links | |
|---|---|---|
| IB+BBBackbone=ImageBind, Variant=BrokenBind2026.02 | 40.2 | |
| LB+BBBackbone=LanguageBind, Variant=BrokenBind2026.02 | 39.4 | |
| LB+FTBackbone=LanguageBind, Variant=Fine-tuned2026.02 | 21.3 | |
| IB+FTBackbone=ImageBind, Variant=Fine-tuned2026.02 | 19.6 | |
| IBBackbone=ImageBind, Variant=Zero-shot2026.02 | 14.1 | |
| LBBackbone=LanguageBind, Variant=Zero-shot2026.02 | 13.8 |