Vision-Language Reasoning on SugarCrepe (test)
62.75Simple AccuracyQ4 system redistr (prop)
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Q4 system redistr (prop)Intervention=Q4 system redistribution, strategy=proportional2026.01 | 62.75 | 25.93 | 85.71 | |
| Q4 text redistr (prop)Intervention=Q4 text redistribution, strategy=proportional2026.01 | 58.79 | 17.58 | 90.66 | |
| Image×2.0Intervention=Image×2.02026.01 | 58.18 | 16.37 | 91.37 | |
| Q4 system ablIntervention=Q4 system ablation2026.01 | 57.97 | 16.04 | 91.59 | |
| No intervention baselineIntervention=None2026.01 | 57.53 | 15.16 | 92.03 | |
| PAIIntervention=PAI2026.01 | 57.47 | 15.05 | 91.98 | |
| AD-HHIntervention=AD-HH2026.01 | 57.14 | 14.4 | 92.53 |