Thermal-to-Text Retrieval (via Vision pivot) on FLIR to LLVIP (test)
45.2mAPLB+BB
Evaluation Results
| Method | Links | |
|---|---|---|
| LB+BBBackbone=LanguageBind, Variant=BrokenBind2026.02 | 45.2 | |
| IB+BBBackbone=ImageBind, Variant=BrokenBind2026.02 | 42.5 | |
| LB+FTBackbone=LanguageBind, Variant=Fine-tuned2026.02 | 23.4 | |
| IB+FTBackbone=ImageBind, Variant=Fine-tuned2026.02 | 21.2 | |
| IBBackbone=ImageBind, Variant=Zero-shot2026.02 | 15.8 | |
| LBBackbone=LanguageBind, Variant=Zero-shot2026.02 | 14.7 |