Class Geometry as Supervision for Sample-Efficient Open-World Detection

New research tackles open-world object detection with novel geometry and dual-perspective frameworks · 2 sources tracked

Two new research papers propose novel approaches to open-world object detection, a challenging task that requires models to identify known objects, reject unknown ones, and adapt to new classes over time. The first pape…