← 研踪 PaperTrailarXiv:2609.02448 · v2
Adapting a Foundation Model for Lunar Surface Height Estimation
首次提交 2026-09-02 19:11
版本更新 2026-11-18 17:03
雷达观测 2026-11-19 19:07
摘要
Digital elevation models (DEMs) can provide accurate height information, making it invaluable for analyzing the lunar surface. Due to recent advancements in computer vision, the focus is on generating DEMs using monocular depth estimation via deep learning. Classic lunar crater detection methods rely mainly on 2D image data. Our goal is to address this issue by developing a relative lunar surface height estimator that could provide additional information for crater localization. In this letter, we present a methodology that builds on the zero-shot relative depth estimation model Depth Anything V2 (DAV2). Other works have been using it as a state-of-the-art comparison for their proposed lunar DEM estimation method, but without adaptations to the target domain. Thus, it may underperform. Therefore, we propose a fine-tuning strategy with publicly available SPG-derived DEM data of the lunar surface, and refer to it as DEM-DAV2. Specifically, we introduce new learnable parameters in the Transformer-based encoder with low-rank adaptation (LoRA), and we unfreeze certain layers in the model's decoder. On our test set, the proposed DEM-DAV2 reduces the average MAE and RMSE by 80\% and 79\%, respectively, compared to the zero-shot DAV2. This demonstrates a significant improvement in performance compared to the zero-shot model, effectively transforming DAV2 into a reliable relative height estimator of the lunar surface. However, the predictions remain relative and require scale recovery from a low-resolution DEM such as SLDEM.
观测到的版本
相邻记录可能跳过中间版本;首次发现 v2 及以上也会标为版本更新。
v2 · 2026-11-18 17:03 · 最新观测版本
Adapting a Foundation Model for Lunar Surface Height Estimation
Digital elevation models (DEMs) can provide accurate height information, making it invaluable for analyzing the lunar surface. Due to recent advancements in computer vision, the focus is on generating DEMs using monocular depth estimation via deep learning. Classic lunar crater detection methods rely mainly on 2D image data. Our goal is to address this issue by developing a relative lunar surface height estimator that could provide additional information for crater localization. In this letter, we present a methodology that builds on the zero-shot relative depth estimation model Depth Anything V2 (DAV2). Other works have been using it as a state-of-the-art comparison for their proposed lunar DEM estimation method, but without adaptations to the target domain. Thus, it may underperform. Therefore, we propose a fine-tuning strategy with publicly available SPG-derived DEM data of the lunar surface, and refer to it as DEM-DAV2. Specifically, we introduce new learnable parameters in the Transformer-based encoder with low-rank adaptation (LoRA), and we unfreeze certain layers in the model's decoder. On our test set, the proposed DEM-DAV2 reduces the average MAE and RMSE by 80\% and 79\%, respectively, compared to the zero-shot DAV2. This demonstrates a significant improvement in performance compared to the zero-shot model, effectively transforming DAV2 into a reliable relative height estimator of the lunar surface. However, the predictions remain relative and require scale recovery from a low-resolution DEM such as SLDEM.
查看此版本解读 →查看此版本原文 ↗