Real-GDSR: Real-World Guided DSM Super-Resolution via Edge-Enhancing Residual Network

TL;DR

Real-World Guided DSM Super-Resolution (Real-GDSR) tackles the ill-posed problem of upscaling real DSM data by leveraging multimodal guidance from high-resolution optical imagery. The method decomposes the task into two stages: a local residual refinement network that operates on a coarse bicubic-upsampled DSM to repair missing or ambiguous areas, and an edge-enhancing diffusion step that globally smooths while preserving height discontinuities guided by the optical image. Trained on real LR–HR DSM pairs from Switzerland, Real-GDSR employs a lightweight local network with residual learning and a diffusion process inspired by anisotropic diffusion, without the adjustment step used in prior diffusion-guided methods. Empirical results show Real-GDSR outperforms bicubic, DADA, D-SRGAN, and other baselines in RMSE, NMAD, and MedAE, achieving sharp building outlines and realistic urban structures, and demonstrating the feasibility of 10x SR in real-world scenarios. The work offers a practical and robust approach for large-scale urban DSM enhancement with potential impact on terrain analysis, city-scale modeling, and resource management.

Abstract

A low-resolution digital surface model (DSM) features distinctive attributes impacted by noise, sensor limitations and data acquisition conditions, which failed to be replicated using simple interpolation methods like bicubic. This causes super-resolution models trained on synthetic data does not perform effectively on real ones. Training a model on real low and high resolution DSMs pairs is also a challenge because of the lack of information. On the other hand, the existence of other imaging modalities of the same scene can be used to enrich the information needed for large-scale super-resolution. In this work, we introduce a novel methodology to address the intricacies of real-world DSM super-resolution, named REAL-GDSR, breaking down this ill-posed problem into two steps. The first step involves the utilization of a residual local refinement network. This strategic approach departs from conventional methods that trained to directly predict height values instead of the differences (residuals) and utilize large receptive fields in their networks. The second step introduces a diffusion-based technique that enhances the results on a global scale, with a primary focus on smoothing and edge preservation. Our experiments underscore the effectiveness of the proposed method. We conduct a comprehensive evaluation, comparing it to recent state-of-the-art techniques in the domain of real-world DSM super-resolution (SR). Our approach consistently outperforms these existing methods, as evidenced through qualitative and quantitative assessments.

Figures (5)

• Figure 1: Guided super-resolution: given a low-resolution dsm and a high-resolution guide image, our method predicts a high-resolution . The figure shows an example output of the proposed method on low-resolution with a factor of 10.
• Figure 2: Examples of real low-resolution in comparison to the bicubic-downsampled and high-resolution . Note that real low-resolution preserve less information in comparison to their bicubic-dowsampled counterparts.
• Figure 3: Summary of the proposed architecture. Real-GDSR comprises mainly a two-step process: Initially, high-dimensional features are extracted from both bicubic-upsampled low-resolution DSM and high-resolution optical image by a pre-trained model. Subsequently, a local refinement network refines the upsampled DSM by incorporating residual blocks and upsampling operations, followed by a diffusion network, which iteratively enhancing the refined upsampled DSMs, emphasizing edge features from the high-resolution optical image.
• Figure 4: Visual comparison of RealGDSR with selected baselines. Our approach demonstrates its accuracy while producing regularized and smooth DSMs. All examples are taken from the test set.
• Figure 5: Line profile analysis of RealGDSR and other baselines.