PerSense: A One-Shot Framework for Personalized Segmentation in Dense Images

PerSense, a model-agnostic, one-shot, training-free framework for customized instance segmentation in dense pictures, is presented in this research. PerSense uses a Vision-Language Model (VLM), a grounding detector, and a few-shot object counter (FSOC) to provide accurate point prompts from density maps, in contrast to conventional bounding-box-based techniques that suffer from occlusion and clutter. To get rid of false positives and automate exemplar selection, the framework incorporates two new modules: the Point Prompt Selection Module (PPSM) and the Instance Detection Module (IDM).