OmniJigsaw: Enhancing Omni-Modal Reasoning via Modality-Orchestrated Reordering

Benchmark Model Rank Results
audio-understanding-reasoning-on-mmarOmniJigsaw (CMM)#1Accuracy: 71.0
audio-visual-question-answering-on-omnivideobenchOmniJigsaw (CMM)#4Accuracy: 40.50
audio-understanding-reasoning-on-mmsuOmniJigsaw (CMM)#5Accuracy: 70.7
video-question-answering-on-videoholmesOmniJigsaw (CMM)#8Accuracy: 52.53
audio-visual-question-answering-on-dailyomniOmniJigsaw (CMM)#9Accuracy: 71.09
video-question-answering-on-tempcompass-overallVideoJigsaw#9Accuracy: 72.28
video-question-answering-on-mlvuOmniJigsaw (SMS)#29Accuracy: 72.63