Lab2Sim benchmark turns science videos into interactive experiments
Original title: Lab2Sim: Turning Real-World Experiment Videos into Interactive Virtual Experiments
Lab2Sim provides 218 videos and EduVideo2Code improves apparatus fidelity by 24.15 points.
Science experiment videos lack direct interaction for learners. Lab2Sim is a benchmark of 218 videos with 11,881 rubric items. EduVideo2Code extracts procedures as state machines to guide code generation and refines pages through execution verification, improving fidelity across multimodal models.
- Benchmark includes 218 science experiment videos
- EduVideo2Code uses hierarchical extraction and state machines
- Apparatus fidelity gains of 24.15 points
- Works on two backbone models
From the original article
1 min read at huggingface.co
Science experiment videos demonstrate how experiments are performed, but offer no direct interaction for learners. Turning them into virtual experiments requires not only reproducing the observed apparatus and state changes, but also enforcing the preconditions governing each operation. We introduce Lab2Sim, a benchmark of 218 science experiment videos with evidence-grounded specifications and 11,881 rubric items.
By Xusen Hei, Jiali Chen, Jinyu Yang · 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.