D2-CDIG: Controlled Diffusion Remote Sensing Image Generation with Dual Priors of DEM and Cloud-Fog

2026-05-15 · Source: cs.CV updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Expert, extended

Summary

The D2-CDIG framework introduces a novel approach to remote sensing image generation, integrating diffusion models with a dual-prior control mechanism to enhance accuracy and naturalness. Developed by researchers at China University of Mining and Technology, D2-CDIG utilizes Digital Elevation Model (DEM) and cloud-fog information as dual prior knowledge to precisely control ground features and atmospheric phenomena. It decouples terrain and atmospheric generation processes through independent ground and atmospheric branches, and incorporates a refined cloud-fog slider for flexible adjustment of cloud thickness and distribution. During training, control signals are injected in layers to ensure seamless transitions. This method significantly improves image quality, detail richness, and realism compared to traditional segmentation or edge detection techniques, providing high-quality data for training large remote sensing models and various downstream tasks.

Key takeaway

For Computer Vision Engineers developing remote sensing applications, D2-CDIG offers a robust solution for generating highly realistic and controllable synthetic imagery. You should consider integrating this dual-prior diffusion model to create diverse training datasets, especially for scenarios requiring precise control over terrain and atmospheric conditions. This approach can significantly improve the performance of downstream tasks like environmental monitoring and disaster response by providing semantically rich data.

Key insights

D2-CDIG uses dual DEM and cloud-fog priors with a diffusion model for precise, natural remote sensing image generation.

Principles

• Decouple terrain and atmospheric generation.
• Align control signals with network layer functionality.
• Balance terrain fidelity and atmospheric control in loss.

Method

D2-CDIG extends Stable Diffusion v1.5 with a dual-branch ControlNet, injecting DEM features into high-resolution encoder blocks and cloud-fog features into lower-resolution decoder blocks, optimized with a joint loss function.

In practice

• Use D2-CDIG for high-quality synthetic remote sensing data.
• Adjust cloud coverage with the cloud-density slider.
• Apply generated data for downstream segmentation model training.

Topics

• D2-CDIG
• Remote Sensing Image Generation
• Diffusion Models
• Digital Elevation Model
• Cloud-Fog Control

Best for: Computer Vision Engineer, AI Scientist, Research Scientist, Machine Learning Engineer

Related on AIssential

• AnyBand-Diff: A Unified Remote Sensing Image Generation and Band Repair Framework with Spectral Priors — AnyBand-Diff generates physically consistent remote sensing images by integrating spectral priors and multi-scale physical constraints into a diffusion model.
• Single Image Defogging Using a Fourth-Order Telegraph PDE Guided by Physical Haze Modeling — This paper introduces a novel hybrid defogging model that integrates a fourth-order nonlinear Telegraph Partial Differential Equation (PDE) with a physical haze formation model for single image defogging.
• DVFace: Spatio-Temporal Dual-Prior Diffusion for Video Face Restoration — DVFace uses a one-step, dual-prior diffusion framework for efficient, high-quality video face restoration.
• AsyncPatch Diffusion: spatially-flexible image generation — AsyncPatch Diffusion enables spatially-flexible image generation by applying distinct noise levels to different input regions.
• DiffuSAM: Diffusion Guided Zero-Shot Object Grounding for Remote Sensing Imagery — DiffuSAM combines diffusion models with segmentation models for superior zero-shot object grounding in remote sensing.
• Estimation of instrument and noise parameters for inverse problem based on prior diffusion model — A Gibbs-Diffusion Posterior Sampling (G-DPS) based method efficiently estimates observation parameters and images in inverse problems with diffusion priors.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.CV updates on arXiv.org.