Information-Regularized Attention for Visual-Centric Reasoning
Open paper
Benchmark
Model
Rank
Results
multimodal-reasoning-on-mmmu
InternVL2.5-IRA
–
Accuracy: 47.6
multimodal-reasoning-on-mmmu-pro
InternVL2-IRA
–
Accuracy: 31.2
vision-language-understanding-on-mmstar
InternVL2.5-IRA
–
Accuracy: 58.8
Rank counts only results with a code link.