LabInstruct: Benchmarking Situated Instructional Video Generation for Lab Procedures
LabInstruct shows frontier image-to-video models often look realistic but miss real lab procedures.
LabInstruct is a benchmark for generating situated instructional videos of real laboratory procedures from a workspace image and a natural-language instruction. It contains 204 tasks across five scientific disciplines, with real reference executions and structured annotations of objects, actions, contacts, and state transitions. Evaluating eight frontier image-to-video models, the authors find that visually plausible generations often remain procedurally incorrect, leaving a gap for reliable experimental guidance in self-driving laboratories.
- LabInstruct covers 204 laboratory tasks across five scientific disciplines.
- Tasks include real executions and annotations of objects, actions, and states.
- Eight frontier image-to-video models were evaluated.
- Visually plausible clips frequently remain procedurally incorrect.
From the original article
1 min read at huggingface.co
Self-driving laboratories (SDLs) aim to automate the full experimental loop, from scientific decision-making to physical execution.
By Yuming Fu, Weijia Wu, Jing Chen · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.