Semantic Stress Evaluation on S3E
96.7Original PerformanceQwen3VL-2B
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| Qwen3VL-2BL=24, N=9,6612026.06 | 96.7 | 86.1 | 83.8 | 0.63 | 2.54 | 1.9 | 2.12 | |
| Qwen3VL-4BL=24, N=9,2262026.06 | 93.3 | 85 | 80.1 | 6.51 | 13.32 | 6.81 | 4.28 | |
| InternVL3-2BL=23, N=9,7852026.06 | 92.7 | 90.3 | 84.9 | 0.32 | 4.13 | 3.81 | 2.01 | |
| Qwen3VL-8BL=32, N=9,6692026.06 | 90.4 | 91.3 | 83.9 | 13.62 | 26.55 | 12.93 | 2.85 | |
| Gemma3-4BL=27, N=9,0682026.06 | 84 | 93.4 | 78.7 | 0.03 | 0.96 | 0.93 | 5.45 |