LatentOmni: Rethinking Omni-Modal Understanding via Unified Audio-Visual Latent Reasoning

Benchmark Model Rank Results
audio-visual-question-answering-on-lvomnibenchLatentOmni#2Accuracy: 35.1
audio-visual-question-answering-on-omnivideobenchLatentOmni#7Accuracy: 35.4
audio-visual-question-answering-on-worldsenseLatentOmni#9Accuracy: 48.9
audio-visual-question-answering-on-dailyomniLatentOmni#12Accuracy: 67.4