ESS-PoI: Energy-Driven Semantic Segmentation with Points of Interest

In this paper, we propose ESS-PoI, a novel approach to semantic image segmentation leveraging energy-based models to identify points of interest (PoI). Our approach uses an energy function to highlight salient image features, and then employs an iterative process to attract points to high energy regions. These points of interest guide the segmentation process, potentially capturing semantically meaningful areas more effectively than traditional methods. The study evaluated ESS-PoI in two scenarios: single image segmentation and a comprehensive experiment using the CityScapes dataset for training and testing. Performance metrics including accuracy, precision, recall, F1 score, and Intersection over Union (IoU) were calculated to assess the method's effectiveness. The results show the potential of ESS-Pol in identifying semantic image regions. This work opens new avenues for combining energy-based models with semantic segmentation techniques, potentially leading to more robust and interpretable image understanding algorithms.