MOMO: Mars Orbital Model Foundation Model for Mars Orbital Applications
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , , , , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2026
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866917382652428288 |
|---|---|
| author | Purohit, Mirali Gajera, Bimal Mehta, Irish Tokas, Bhanu Adler, Jacob Lu, Steven Dickenshied, Scott Diniega, Serina Bue, Brian Rebbapragada, Umaa Kerner, Hannah |
| author_facet | Purohit, Mirali Gajera, Bimal Mehta, Irish Tokas, Bhanu Adler, Jacob Lu, Steven Dickenshied, Scott Diniega, Serina Bue, Brian Rebbapragada, Umaa Kerner, Hannah |
| contents | 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 $\sim 12$ 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: https://github.com/kerner-lab/MOMO. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2604_02719 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | MOMO: Mars Orbital Model Foundation Model for Mars Orbital Applications Purohit, Mirali Gajera, Bimal Mehta, Irish Tokas, Bhanu Adler, Jacob Lu, Steven Dickenshied, Scott Diniega, Serina Bue, Brian Rebbapragada, Umaa Kerner, Hannah Computer Vision and Pattern Recognition Artificial Intelligence Machine Learning 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 $\sim 12$ 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: https://github.com/kerner-lab/MOMO. |
| title | MOMO: Mars Orbital Model Foundation Model for Mars Orbital Applications |
| topic | Computer Vision and Pattern Recognition Artificial Intelligence Machine Learning |
| url | https://arxiv.org/abs/2604.02719 |