Bottom-up and Layerwise Domain Adaptation for Pedestrian Detection in Thermal Images

Pedestrian detection is a canonical problem for safety and security applications, it remains challenging due to the highly variable lighting conditions in which pedestrians must be detected. This article investigates several domain adaptation approaches adapt RGB-trained detectors thermal domain. Building on our earlier work privacy-preserving pedestrian detection, we conducted an extensive experimental evaluation comparing top-down bottom-up also propose two new strategies. For adaptation, leverage detector pre-trained RGB imagery efficiently perform Our include steps: first, training adapter segment corresponding initial layers of adapts input distribution; then, reconnect original final with loss. To best knowledge, outperform best-performing single-modality results KAIST state art FLIR.