Lin et al.
Benchmark and method for true audio-visual joint reasoning
A new benchmark, data engine and learning method demonstrate that omni-modal models can be trained to reason across audio and vision when both modalities are genuinely required.
Chinese Tech
Junming Lin · Yuxuan Wang · Zhenxin Lei · Yuxin Liu · Ruixun Liu · Yinsong Yan · +8 more
Peking University · Alibaba Group
Research Digest··3 min read
The authors introduce OmniReasoningBench, a benchmark of 1,150 questions where both audio and visual evidence are indispensable, and show that existing omni-modal benchmarks often allow single-modality solutions.
Why this paper
From Alibaba Group and Peking University
In one line
OmniReasoning framework improves audio-visual joint reasoning accuracy by 12.8 and 9.3 points on two benchmarks.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ·No stated limitations found
- ✓Reports numbers on named benchmarks (2 benchmarks)
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§