The authors constructed WOVEN using video-pretrained generative models to create 36,076 examples spanning 20 scene types, 5 action types, and 8 reasoning types.
Shared visual transition training improves multimodal LLMs across diverse tasks
The WOVEN benchmark and training recipe show a single capability transfers to many external benchmarks and can replace task-specific data.
Big Tech
Zheyu Fan · Yue Zhang · Mingkai Deng · Kangrui Wang · Qineng Wang · Canyu Chen · +6 more
Northwestern University · Carnegie Mellon University · UNC Chapel Hill · Amazon
Research Digest··2 min read
The authors introduce WOVEN, a training source and benchmark for visual transition reasoning, and show that training on it improves multimodal LLMs across many external benchmarks, even with small subsets.
Why this paper
From Amazon and 3 others
In one line
Visual transition reasoning can serve as a shared training primitive for multimodal LLMs, improving performance across diverse tasks.
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
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§