Figure 8. HiResCAM-based combined heatmap visualisation for layers 15, 18, and 21, focusing on brake light regions inside bounding boxes. - https://doi.org/10.70594/brain/17.1/1

Read full text Summarise with VIRAT
Authors: Amil Rohani Dar, Faisal Riaz
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

The explainability analysis targeted key convolutional layers (layers 15, 18, and 21), systematically selected based on their spatial resolution and semantic richness within the model's architecture. By aggregating the activation maps from these layers, as shown in Figure 8, we generated composite heatmaps that effectively highlighted the critical regions influencing the model’s predictions. The HiResCAM-generated visualisations consistently demonstrated that the model’s attention was predominantly focused within the regions bounded by the detection boxes corresponding to vehicles, particularly emphasising the taillights and rear areas where brake lights are located. This alignment with human visual reasoning is crucial, as brake light status is alocalised visual cue critical for safe driving decisions. The results validate the model's reasoning process by showing its focus on contextually relevant areas, thereby bridging the gap between opaque AI systems and human-interpretable insights.

Fields of Study & Subjects

Explainable Artificial Intelligence (XAI) Advanced Neural Network Applications Autonomous Vehicle Technology and Safety

Publication details

Journal / source
Zenodo (CERN European Organization for Nuclear Research)
Publication date
26 March 2026
Language
EN