Problem
Research questions and friction points this paper is trying to address.
Acoustic Pedestrian Detection
Cross-Modal Knowledge Distillation
Class Imbalance
Modality Gap
Cross-Modal Knowledge Distillation for Acoustic Pedestrian Detection
🤖 AI Summary
This study addresses the challenges of weak acoustic cues and the failure of cross-modal distillation under class imbalance in audio-only pedestrian detection. To this end, it proposes a Trust-Filtered Distillation (TFD) strategy that leverages a video teacher model to supervise an audio student model. Theoretically, we reveal the conditional label smoothing nature of TFD, explicitly disentangling operating point shifts from discriminability gains. Experimentally, the effectiveness of logit distillation is validated on the ASPED dataset. Results demonstrate that while the proposed method significantly improves non-pedestrian accuracy, it reduces pedestrian recall without enhancing the average PR-AUC. These findings objectively expose the inherent limitations and trade-offs associated with selective supervision mechanisms in highly imbalanced scenarios.
📝 Abstract
Audio-only pedestrian detection is attractive for urban sensing but limited by weak acoustic cues. An appealing strategy is cross-modal knowledge distillation, in which a video teacher supervises the audio student during training so that the deployed model runs on audio alone. Under this task's severe class imbalance and wide video-audio modality gap, however, what such distillation contributes is unclear. We introduce Trust-Filtered Distillation (TFD), which selectively suppresses teacher supervision on pedestrian samples, and interpret its logit formulation as conditional label smoothing under a shared temperature. Across ten distillation configurations under five-fold cross-session validation on ASPED, most methods yield modest gains in macro accuracy accompanied by small changes in PR-AUC. The main effect is higher no-pedestrian accuracy at the cost of lower pedestrian recall. Adding TFD to logit distillation strengthens this trade-off without improving mean PR-AUC. These findings clarify the benefits and limitations of selective cross-modal supervision by distinguishing operating-point shifts from discrimination gains in imbalanced acoustic detection.
Innovation
Methods, ideas, or system contributions that make the work stand out.
Cross-Modal Knowledge Distillation
Trust-Filtered Distillation
Acoustic Pedestrian Detection
Conditional Label Smoothing
Class Imbalance
🔎 Similar Papers
No similar papers found.
