Towards Real-Time Explainable AI: Using Class Activation Mapping for Brake Prediction in YOLO-Based Systems - Figure 2

Advancements in object detection have resulted in more efficient YOLO-based systems, outperforming alternatives such as RetinaNet, Fast R-CNN, and SSD in terms of speed, accuracy, and learning capability (Alqarqaz et al., 2023). Figure 2 provides a visual representation of the comparative performance of these algorithms.

To enhance detection accuracy while maintaining real-time performance, the authors in (Menaka et al., 2020) have explored a hybrid method that merges YOLO with Faster R-CNN. In this framework, YOLO quickly identifies potential object regions by drawing bounding boxes, and Faster R-CNN refines these results using its RoI pooling for accurate classification and segmentation.