InterPruner: Interactive Structured Pruning via Taylor-Implicit Criterion and Language-Prior Modulator for Multimodal Object Detection

2026-08-11

AI Summary

This work addresses the challenges of channel redundancy and high computational cost in RGB-infrared multimodal object detection arising from parallel feature extraction, where existing pruning methods overlook cross-modal interactions and scene-level dynamic redundancy. To this end, we propose the first interactive structured channel pruning framework tailored for this task, which innovatively integrates three key components: a Taylor-based implicit criterion to quantify channel importance, a Modality Interaction Redundancy Analysis (MIRA) module to model cross-modal complementarity, and a language prior-guided Scene-level Pruning with Contextual Awareness (SPCA) mechanism to enable dynamic, context-aware channel pruning. Evaluated on the FLIR dataset, our method achieves a 0.6% increase in mAP after pruning 50% of channels, demonstrating simultaneous reductions in computational overhead and improvements—or at least preservation—of detection performance.

Abstract

Multimodal object detection proves effective in remote sensing, especially the RGB-Infrared paradigm. The parallel feature extractors provide rich multimodal information for robust detection, yet introduce substantial channel redundancy and computational overhead. Existing pruning methods can reduce channel redundancy, but they are designed for unimodal backbones, overlooking cross-modal interactions and dynamic scene-wise redundancy. In this paper, we propose InterPruner, the first interactive structured channel pruning framework for RGB-infrared object detectors. Specifically, we first derive a Taylor-Implicit Criterion(TIC) to quantify channel importance via high-order Taylor expansion and the implicit function theorem. Then, a Modality Interaction Redundancy Analyzer (MIRA) identifies redundant channels via mutual compensability assessment. Finally, a Scene-Prior Channel Anchor (SPCA) uses language priors as semantic anchors to measure channel-scene relevance for dynamic channel importance estimation. Cross-modality channel pruning for RGB-Infrared detection is yet unexplored. Extensive experiments on RGB-infrared object detection dataset demonstrate that InterPruner maintains high performance with negligible degradation. Specifically, it even achieves a 0.6% mAP increase on the FLIR dataset when pruning 50% of the channels. Code will be available on GitHub to facilitate future work.

Problem

• multimodal object detection
• structured pruning
• channel redundancy
• cross-modal interaction
• RGB-Infrared

Innovation

• structured pruning
• multimodal object detection
• Taylor-Implicit Criterion
• cross-modal interaction
• language-prior modulation