Decoupled Functional Evaluation of Autonomous Driving Models via Feature Map Quality Scoring

2025-08-10

AI Summary

End-to-end autonomous driving models suffer from poor feature map interpretability and difficulty in independent evaluation and optimization due to the absence of explicit supervision on intermediate modules. To address this, we propose a decoupled functional evaluation framework grounded in feature map quality scoring, introducing— for the first time—the Feature Map Convergence Score (FMCS) mechanism and a dual-granularity dynamic weighting scoring system with a unified quality metric. Our method, CLIP-FMQE-Net, employs a CLIP-based feature encoder coupled with a quality prediction head, jointly modeling feature–ground-truth similarity and incorporating dynamic weighting for fine-grained, real-time quality quantification. Integrated into a 3D object detector on NuScenes, our module improves the NuScenes Detection Score (NDS) by 3.89%, demonstrating substantial enhancement in both feature representation quality and end-to-end system performance.

Technology Category

• Computer Vision: Learning & Optimization for CV
• Machine Learning: Feature Construction/Reformulation
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

End-to-end models are emerging as the mainstream in autonomous driving perception and planning. However, the lack of explicit supervision signals for intermediate functional modules leads to opaque operational mechanisms and limited interpretability, making it challenging for traditional methods to independently evaluate and train these modules. Pioneering in the issue, this study builds upon the feature map-truth representation similarity-based evaluation framework and proposes an independent evaluation method based on Feature Map Convergence Score (FMCS). A Dual-Granularity Dynamic Weighted Scoring System (DG-DWSS) is constructed, formulating a unified quantitative metric - Feature Map Quality Score - to enable comprehensive evaluation of the quality of feature maps generated by functional modules. A CLIP-based Feature Map Quality Evaluation Network (CLIP-FMQE-Net) is further developed, combining feature-truth encoders and quality score prediction heads to enable real-time quality analysis of feature maps generated by functional modules. Experimental results on the NuScenes dataset demonstrate that integrating our evaluation module into the training improves 3D object detection performance, achieving a 3.89 percent gain in NDS. These results verify the effectiveness of our method in enhancing feature representation quality and overall model performance.

Problem

• Evaluates autonomous driving models' intermediate feature maps independently
• Proposes a feature map quality score for module evaluation
• Improves 3D object detection via enhanced feature representation

Innovation

• Feature Map Convergence Score evaluation method
• Dual-Granularity Dynamic Weighted Scoring System
• CLIP-based Feature Map Quality Evaluation Network