驯服生成式扩散模型用于任务导向红外成像
Taming Generative Diffusion Model for Task-Oriented Infrared Imaging
把扩散模型的多步采样压缩成单步、并用动态时间步估计对齐潜变量轨迹，这一思路可直接迁移到其他退化严重、需要物理约束的低层复原任务（如低光、去雾、医学成像）；同时用任务感知低秩适配把复原模型快速接到下游语义任务，是「复原+下游」联合设计的可复用范式。
原文依据 we reformulate infrared imaging as a single-step diffusion process, aligning degraded observations with trajectory latent states via dynamic timestep estimation
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论红外成像对于恶劣环境下的感知至关重要。然而，动态耦合的退化因素严重损害了视觉质量和下游语义精度。尽管生成式扩散模型提供了强大的图像复原先验，但其高昂的计算成本与物理不一致性限制了它们在红外传感中的应用。为弥合这些差距，我们将红外成像重新表述为单步扩散过程，通过动态时间步估计将退化观测与轨迹潜变量状态对齐，从而利用时间步特定的扩散先验实现高保真重建。同时，我们引入光谱正则化项以施加热辐射约束并确保物理一致性。随后，通过动态提示设计了一种任务感知的低秩适配机制，以实现跨下游红外任务的高效迁移。实验表明，我们的方法在复原质量、语义结构保持和任务泛化方面均优于现有方法。代码可在 https://github.com/csmty/InfraredIR 获取。
英文原文摘要 1114 字符
Infrared imaging is essential for perception in harsh environments. However, dynamically coupled degradation factors severely impair visual quality and downstream semantic accuracy. Although generative diffusion models provide strong image restoration priors, high computational cost and physical inconsistency limit their application in infrared sensing. To bridge these gaps, we reformulate infrared imaging as a single-step diffusion process, aligning degraded observations with trajectory latent states via dynamic timestep estimation to leverage timestep-specific diffusion priors for high-fidelity reconstruction. Meanwhile, we introduce a spectral regularization term to enforce thermal radiation constraints and ensure physical consistency. Subsequently, a task-aware low-rank adaptation mechanism is devised through dynamic prompting to enable efficient transfer across downstream infrared tasks. Experiments demonstrate our method surpasses existing approaches in restoration quality, semantic structure preservation, and task generalization. The code is available at https://github.com/csmty/InfraredIR.