Hierarchical diffusion models for generating various pattern vehicles in infrared aerial images

Nan Zhang, Youmeng Liu, Hao Liu, Tian Tian, Jiayi Ma, Jinwen Tian

Abstract: Highlights•Hierarchical Diffusion Model (HDM) is proposed for vehicle augmentation in infrared aerial images.•The hierarchical architecture of HDM ensures high-quality vehicle generation across multiple scales.•Refined prompts are explored, which improves the fidelity and diversity of the generated infrared vehicles.•The inpainting technique is adopted during sampling stage, which achieves a seamless synthesis of vehicles into the background.