Harmful Meme Detection on Ex-ToxiCN-MM (test)
96.7AccuracyInternVL3_5-8B
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| InternVL3_5-8BStrategy=RIKE2026.05 | 96.7 | 96.9 | 96.8 | 96.9 | — | |
| InternVL3_5-8BStrategy=SFT+RIR2026.05 | 96.6 | 97.4 | 96.1 | 96.8 | — | |
| Qwen2.5-VL-7BStrategy=RIKE2026.05 | 95.7 | 94.9 | 97.1 | 96 | — | |
| Qwen2.5-VL-7BStrategy=SFT+RIR2026.05 | 95.6 | 96.2 | 95.5 | 95.8 | — | |
| Qwen2.5-VL-3BStrategy=RIKE2026.05 | 94.8 | 94 | 96.4 | 95.2 | — | |
| Qwen2.5-VL-3BStrategy=SFT+RIR2026.05 | 94.3 | 93.7 | 95.6 | 94.6 | — | |
| LLaVa-1.5-7bStrategy=RIKE2026.05 | 91.5 | 96.9 | 86.8 | 91.5 | — | |
| LLaVa-1.5-7bStrategy=SFT+RIR2026.05 | 89.7 | 96.9 | 83.3 | 89.6 | — | |
| Qwen2.5-VL-7BStrategy=SFT2026.05 | 86.9 | 95.3 | 79.3 | 86.6 | — | |
| InternVL3_5-8BStrategy=SFT2026.05 | 86.6 | 96.4 | 77.6 | 86 | — | |
| Qwen2.5-VL-3BStrategy=SFT (Supervised Fine-tuning)2026.05 | 82.6 | 94.5 | 71.4 | 81.3 | — | |
| Qwen-VL-maxStrategy=RIKE2026.05 | 75.7 | 78.4 | 74.6 | 76.5 | — | |
| Qwen-VL-maxStrategy=RIR2026.05 | 73.1 | 78 | 68.6 | 73 | — | |
| LLaVa-1.5-7bStrategy=SFT2026.05 | 72.3 | 98.6 | 48.5 | 65 | — | |
| Qwen-VL-maxStrategy=Base (Direct Reasoning)2026.05 | 71.9 | 87.5 | 54.6 | 67.3 | — | |
| GLM-4v-flashStrategy=RIKE2026.05 | 70.5 | 72.7 | 67.4 | — | 70 | |
| Qwen2.5-VL-7BStrategy=RIR+BK (AKE)2026.05 | 70.2 | 83.4 | 54.6 | 66 | — | |
| InternVL3_5-8BStrategy=RIR+BK (AKE)2026.05 | 67.4 | 84 | 47.7 | 60.8 | — | |
| GLM-4v-flashStrategy=RIR2026.05 | 64.9 | 74.7 | 47 | — | 57.7 | |
| Qwen2.5-VL-7BStrategy=RIR2026.05 | 64.3 | 80.5 | 43.1 | 56.2 | — | |
| Qwen2.5-VL-3BStrategy=RIR+BK (AKE)2026.05 | 61.9 | 59.5 | 87.8 | 71 | — | |
| InternVL3_5-8BStrategy=RIR2026.05 | 61.4 | 81 | 35.5 | 49.4 | — | |
| GLM-4v-flashStrategy=Base (Direct Reasoning)2026.05 | 60.8 | 84.3 | 28.3 | — | 42.3 | |
| InternVL3_5-8BStrategy=Base (Direct Reasoning)2026.05 | 59.6 | 83.7 | 29.6 | 43.7 | — | |
| LLaVa-1.5-7bStrategy=RIR+BK (AKE)2026.05 | 59.2 | 60.7 | 65.5 | 63 | — | |
| Qwen2.5-VL-7BStrategy=Base (Direct Reasoning)2026.05 | 58.3 | 91.2 | 23.6 | 37.5 | — | |
| Qwen2.5-VL-3BStrategy=Base (Direct Reasoning)2026.05 | 56.9 | 67.3 | 36.4 | 47.2 | — | |
| Qwen2.5-VL-3BStrategy=RIR (Relative Intent Reasoning)2026.05 | 56.6 | 56.7 | 77.2 | 65.4 | — | |
| LLaVa-1.5-7bStrategy=RIR2026.05 | 55.7 | 58.3 | 58 | 58.1 | — | |
| LLaVa-1.5-7bStrategy=Base (Direct Reasoning)2026.05 | 50.9 | 57.2 | 29.2 | 38.7 | — |