Let VLMs Grade Their Own Thoughts: A Self-Quantification Approach to Reasoning-Aware Reward Modeling

Benchmark Model Rank Results
spatial-reasoning-on-vsi-benchVideo-RAISE-7B–Accuracy: 38.7
video-question-answering-on-mmvuVideo-RAISE-7B–Accuracy: 66.9
video-question-answering-on-mvbenchVideo-RAISE-7B–Avg.: 65.9
video-question-answering-on-tempcompass-overallVideo-RAISE-7B–Accuracy: 65.9
video-question-answering-on-videommmuVideo-RAISE-7B–Accuracy: 53.6