Fast and Generalizable NeRF Architecture Selection for Satellite Scene Reconstruction
Devjyoti Chakraborty, Zaki Sukma, Rakandhiya D. Rachmanto, Kriti Ghosh, In Kee Kim, Suchendra M. Bhandarkar, Lakshmish Ramaswamy, Nancy K. O’Hare, Deepak Mishra · 2026
Read Fast and Generalizable NeRF Architecture Selection for Satellite Scene Reconstruction by Devjyoti Chakraborty, Zaki Sukma, Rakandhiya D. Rachmanto, Kriti Ghosh, In Kee Kim, Suchendra M. Bhandarkar, Lakshmish Ramaswamy, Nancy K. O’Hare, Deepak Mishra (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-18306.txt.
Description
Chakraborty and co-authors present PreSCAN, a fast way to pick NeRF architectures for satellite scene reconstruction. Per-scene training and architecture search can take hours to days of GPU time. Their SHAP analysis shows multi-view consistency, not architecture, drives reconstruction quality. PreSCAN predicts quality from light geometric and photometric descriptors before training starts. It picks architectures in under 30 seconds with under 1 dB error, and cuts edge inference power by 26%.
Catalog record
| CREATOR | Devjyoti Chakraborty, Zaki Sukma, Rakandhiya D. Rachmanto, Kriti Ghosh, In Kee Kim, Suchendra M. Bhandarkar, Lakshmish Ramaswamy, Nancy K. O’Hare, Deepak Mishra |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 34410 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2603.18306