Sat3R: Satellite DSM Reconstruction via RPC-Aware Depth Fine-tuning

2026-05-08

AI Summary

Existing methods for digital surface model (DSM) reconstruction from satellite imagery struggle to balance accuracy and efficiency: optimization-based approaches are computationally expensive, while general-purpose geometric foundation models fail to generalize due to discrepancies between RPC camera models and depth scale. This work proposes Sat3R, the first framework to adapt the monocular depth foundation model Depth Anything V2 to the satellite domain via RPC-aware metric depth fine-tuning. By leveraging RPC geometry to generate physically consistent pseudo-depth supervision, Sat3R enables feed-forward DSM reconstruction without per-scene optimization. Combining SiLog loss with RPC geometric constraints, the method reduces MAE error by 38% over zero-shot baselines on the DFC2019 benchmark, achieving accuracy comparable to optimization-based methods while accelerating inference by over 300×.

Abstract

Accurate Digital Surface Model (DSM) reconstruction from satellite imagery is critical for applications such as disaster response, urban planning, and large-scale geographic mapping. Existing approaches face a fundamental trade-off: optimization-based methods achieve strong accuracy but require hours of per-scene computation, while generalizable geometry foundation models offer near-instant inference but fail to generalize to satellite imagery due to the domain gap introduced by the Rational Polynomial Camera (RPC) model and mismatched depth scale distributions. We present Sat3R, a feed-forward framework that bridges this gap via RPC-aware metric depth fine-tuning of Depth Anything V2 using the Scale-Invariant Logarithmic (SiLog) loss. By constructing physically consistent pseudo depth supervision from RPC geometry, Sat3R adapts a monocular depth foundation model to the satellite domain without per-scene optimization. Experiments on the DFC2019 benchmark demonstrate that Sat3R reduces MAE by 38% over zero-shot feed-forward baselines and achieves competitive accuracy against optimization-based methods, while delivering over 300x speedup. Sat3R demonstrates that feed-forward models, when properly adapted to the satellite domain, can match optimization-based accuracy at a fraction of the computational cost, paving the way for practical large-scale satellite DSM reconstruction.

Problem

• Digital Surface Model
• satellite imagery
• RPC model
• domain gap
• depth estimation

Innovation

• RPC-aware depth fine-tuning
• Satellite DSM reconstruction
• Depth Anything V2
• SiLog loss
• feed-forward framework