Reproduce “MMTA: Benchmarking Multimodal Temporal Analysis with Time Series, Text, and Vision” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Ziyang Zhang, Shenyi Li, Yilin Wang, Ziyun Cui, Bowen Zhou, Wen Wu, Chao Zhang
TimeOmni-v encodes time series as a separate modality and aligns their tokens with video timelines, addressing the prompt length and alignment limits of writing measurements as text. The authors evaluate it on MMTA, a benchmark combining visual, language, and time-series evidence. On AgiBot classification, it achieves 0.844 swF1 versus 0.126 for the strongest baseline, but it does not lead forecasting.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.