Multi-class Classification on ToxiCN_MM Task B
81.84Macro F1 ScoreKID
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| KID2026.01 | 81.84 | 86.42 | |
| Qwen2.5-VL-7BArchitecture=Dual-Head, Input=Caption2026.01 | 78.76 | 80.67 | |
| Qwen2.5-VL-7BArchitecture=Dual-Head2026.01 | 78.29 | 81.38 | |
| Qwen2.5-VL-7BProtocol=SFT2026.01 | 74.23 | 79.16 | |
| Fusion-MKE2026.01 | 62.17 | — | |
| CLIP2026.01 | 57.85 | — | |
| Gemini-2.5-FlashProtocol=Few-shot2026.01 | 56.32 | 63.72 | |
| Gemini-2.5-FlashProtocol=Zero-shot2026.01 | 40.63 | 38.06 |