Zero-shot Classification on DTD
84.5AccuracyLLaVA-OV HEC-T
Evaluation Results
| Method | Links | |
|---|---|---|
| LLaVA-OV HEC-TBackbone=SigLIP+LLM (LLaVA-OV), Protocol=10-way 0-shot2026.03 | 84.5 | |
| Qwen2-VL HEC-TBackbone=DFN+LLM (Qwen2-VL), Protocol=10-way 0-shot2026.03 | 84.2 | |
| SigLIP Zero-ShotBackbone=SigLIP, Protocol=10-way 0-shot2026.03 | 84.1 | |
| DFN Zero-ShotBackbone=DFN, Protocol=10-way 0-shot2026.03 | 78.4 | |
| LLaVA-OV BaselineBackbone=SigLIP+LLM (LLaVA-OV), Protocol=10-way 0-shot2026.03 | 77.8 | |
| Qwen2-VL STBackbone=DFN+LLM (Qwen2-VL), Protocol=10-way 0-shot2026.03 | 77.3 | |
| Qwen2-VL BaselineBackbone=DFN+LLM (Qwen2-VL), Protocol=10-way 0-shot2026.03 | 70.9 | |
| BiFTABackbone=ViT-B/16, Zero-shot=true2026.01 | 54.64 | |
| WCABackbone=ViT-B/16, Zero-shot=true2026.01 | 52.87 | |
| CuPLBackbone=ViT-B/16, Zero-shot=true2026.01 | 50.53 | |
| CLIP-DBackbone=ViT-B/16, Zero-shot=true2026.01 | 46.17 | |
| CLIP-EBackbone=ViT-B/16, Zero-shot=true2026.01 | 45.27 | |
| WaffleBackbone=ViT-B/16, Zero-shot=true2026.01 | 44.68 | |
| LLaVA-OV STBackbone=SigLIP+LLM (LLaVA-OV), Protocol=10-way 0-shot2026.03 | 44.2 | |
| CLIPBackbone=ViT-B/16, Zero-shot=true2026.01 | 42.98 |