Explaining YOLO: Leveraging Grad-CAM to Explain Object Detections

TL;DR

This paper explores how to enhance the explainability of the YOLO object detector by integrating Grad-CAM, analyzing attribution-based explanations, and emphasizing the importance of normalization for interpretation.

Contribution

It introduces a method to incorporate Grad-CAM into YOLO for better explanations of detections and highlights the impact of normalization on interpretability.

Findings

Normalization significantly affects explanation clarity

Grad-CAM can be integrated into YOLO for attribution analysis

Attribution explanations help understand detection decisions

Abstract

We investigate the problem of explainability for visual object detectors. Specifically, we demonstrate on the example of the YOLO object detector how to integrate Grad-CAM into the model architecture and analyze the results. We show how to compute attribution-based explanations for individual detections and find that the normalization of the results has a great impact on their interpretation.