This paper formulates Slides2MindMap, a task for reconstructing cognitively efficient knowledge hierarchies from collections of lecture slides. It introduces S2M-Bench, containing 12,774 slide pages from 24 university courses with expert-annotated mind maps and an evaluation framework combining reference-based comparison, structural conformity, and VLM-as-a-Judge. The proposed AutoMindMap agent builds a global skeleton, iteratively integrates summarized knowledge, and applies local-global dual-stage refinement. The abstract reports gains over baselines and robustness across models and scenarios, but provides no numerical results.
No heat snapshots are available in the last 24 hours.