Active View Selection
Overview
Active view selection studies how an agent can move a camera or sensor to gather useful evidence instead of observing from fixed viewpoints or scanning exhaustively. Methods estimate information value using uncertainty, Fisher information, predictive entropy, visibility and coverage, then choose a next-best view or plan a longer trajectory. The approach is used in 3D reconstruction with radiance fields and Gaussian splatting, as well as robot navigation, manipulation, medical imaging, and multi-view reasoning. A central challenge is balancing information gain with task relevance, view feasibility, computation, and the reliability of uncertain or generated observations.
Latest papers
36 papers1 paper in the last four weeks, down 87% on the four weeks before. 0.0% of all new papers.