MOMO：面向火星轨道应用的火星轨道模型基础模型
MOMO: Mars Orbital MOdel Foundation Model for Mars Orbital Applications
用模型融合（task arithmetic）把不同分辨率传感器上独立预训练的表示合并成一个基础模型，并给出按验证损失相似度对齐检查点的 EVL 策略；做多源、多分辨率遥感或多模态预训练的人可以直接复用这套「先各自预训练、再按收敛阶段对齐融合」的流程，避免从头联合训练。
原文依据 MOMO uses model merge to integrate representations learned independently from three key Martian sensors
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论我们提出 MOMO，这是首个面向火星遥感的多传感器基础模型。MOMO 采用模型融合，将分别从三种关键火星传感器（HiRISE、CTX 和 THEMIS）独立学习到的表示整合起来，覆盖从 0.25 米/像素到 100 米/像素的分辨率范围。本方法的核心是我们提出的等验证损失（EVL）策略：在通过任务算术进行融合之前，该策略依据验证损失相似度在不同传感器之间对齐检查点。这确保模型在相互兼容的收敛阶段被融合，从而带来更好的稳定性与泛化能力。我们在从火星轨道数据中整理出的、包含 1200 万样本的大规模高质量语料上训练 MOMO，并在 Mars-Bench 的 9 个下游任务上进行评测。与 ImageNet 预训练、地球观测基础模型、传感器特定预训练以及全监督基线相比，MOMO 取得了更好的整体性能。尤其在分割任务上，MOMO 表现出持续且显著的性能提升。我们的结果表明，通过最优检查点选择策略进行模型融合，为构建面向多分辨率数据的基础模型提供了一条有效途径。模型权重、预训练代码、预训练数据与评测代码已发布于：github.com/kerner-lab/MOMO。
英文原文摘要 1302 字符
We introduce MOMO, the first multi-sensor foundation model for Mars remote sensing. MOMO uses model merge to integrate representations learned independently from three key Martian sensors (HiRISE, CTX, and THEMIS), spanning resolutions from 0.25 m/pixel to 100 m/pixel. Central to our method is our novel Equal Validation Loss (EVL) strategy, which aligns checkpoints across sensors based on validation loss similarity before fusion via task arithmetic. This ensures models are merged at compatible convergence stages, leading to improved stability and generalization. We train MOMO on a large-scale, high-quality corpus of 12 million samples curated from Mars orbital data and evaluate it on 9 downstream tasks from Mars-Bench. MOMO achieves better overall performance compared to ImageNet pre-trained, earth observation foundation model, sensor-specific pre-training, and fully-supervised baselines. Particularly on segmentation tasks, MOMO shows consistent and significant performance improvement. Our results demonstrate that model merging through an optimal checkpoint selection strategy provides an effective approach for building foundation models for multi-resolution data. The model weights, pretraining code, pretraining data, and evaluation code are available at: github.com/kerner-lab/MOMO.