List your company

Video Question Answering on Video-MME short and medium videos

50.1Temporal AccuracyOracle-RLAIF
38.34841.39944.4547.501Oct 2, 2025

Evaluation results

MethodLinks
Oracle-RLAIFOptimization=Rank-based RLAIF2025.1050.138.14943.442.329.537.84843.2
Oracle-DPOOptimization=DPO2025.1049.534.846.341.440.227.138.148.541.4
VLM SFTOptimization=Supervised Fine-Tuning2025.1048.232.939.543.339.519.634.550.137.5
VLM-RLAIFOptimization=RLAIF (scalar reward + PPO)2025.1038.829.636.845.139.419.434.649.436.9