Compressing Multi-Task Model for Autonomous Driving via Pruning and Knowledge Distillation

2025-11-03

AI Summary

To address the high parameter count and computational overhead of multi-task autonomous driving panoptic perception models (joint object detection, drivable area segmentation, and lane line segmentation) for onboard deployment, this paper proposes a lightweight framework integrating task-aware safe pruning and head-free feature distillation. We innovatively introduce a gradient conflict penalty to guide channel importance estimation via Taylor expansion, enabling precise retention of critical channels. Additionally, task-agnostic intermediate-layer feature distillation mitigates performance degradation induced by pruning. Evaluated on BDD100K, our method reduces model parameters by 32.7% with negligible accuracy loss in drivable area and lane line segmentation, a marginal drop in detection mAP, and an inference speed of 32.7 FPS—satisfying real-time deployment requirements.

Technology Category

• Computer Vision: Learning & Optimization for CV
• Machine Learning: Learning on the Edge & Model Compression
• Search and Optimization: Learning to Search

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Autonomous driving systems rely on panoptic perception to jointly handle object detection, drivable area segmentation, and lane line segmentation. Although multi-task learning is an effective way to integrate these tasks, its increasing model parameters and complexity make deployment on on-board devices difficult. To address this challenge, we propose a multi-task model compression framework that combines task-aware safe pruning with feature-level knowledge distillation. Our safe pruning strategy integrates Taylor-based channel importance with gradient conflict penalty to keep important channels while removing redundant and conflicting channels. To mitigate performance degradation after pruning, we further design a task head-agnostic distillation method that transfers intermediate backbone and encoder features from a teacher to a student model as guidance. Experiments on the BDD100K dataset demonstrate that our compressed model achieves a 32.7% reduction in parameters while segmentation performance shows negligible accuracy loss and only a minor decrease in detection (-1.2% for Recall and -1.8% for mAP50) compared to the teacher. The compressed model still runs at 32.7 FPS in real-time. These results show that combining pruning and knowledge distillation provides an effective compression solution for multi-task panoptic perception.

Problem

• Compressing multi-task autonomous driving models to reduce parameters and complexity
• Maintaining panoptic perception accuracy while enabling real-time deployment
• Balancing pruning efficiency with knowledge distillation to minimize performance loss

Innovation

• Combining pruning with knowledge distillation for compression
• Using safe pruning strategy to remove redundant channels
• Applying task head-agnostic distillation to transfer features