Infrared-Visible Object Detection via Distillation-Fermentation Dual Processing

This paper proposes a novel dual-processing framework for infrared-visible object detection, inspired by the fermentation-distillation paradigm in traditional Chinese liquor brewing. To address the complementary characteristics of RGB and thermal modalities, we first design a Dual-stage Feature Complementary Fusion module (DFCF) that sequentially performs coarse and fine processing on cross-modal features. Subsequently, a Polymorphic Convolution module (PCM) is developed by extending the YOLOv11 architecture with variable kernels and channel separation strategies. Furthermore, an Adaptive Semantic Aggregation module (ASA) effectively integrates shallow boundary details with deep semantic features. Extensive experiments on multiple datasets demonstrate that our method achieves superior performance compared to widely adopted approaches, with particularly significant improvements in challenging scenarios like low-light conditions. The ablation studies validate the contributions of each proposed component. Our code is available at https://github.com/Wante-Eren/DFNet_public.