ZipMap: Linear-Time Stateful 3D Reconstruction via Test-Time Training
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2026
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866908948771110912 |
|---|---|
| author | Jin, Haian Wu, Rundi Zhang, Tianyuan Gao, Ruiqi Barron, Jonathan T. Snavely, Noah Holynski, Aleksander |
| author_facet | Jin, Haian Wu, Rundi Zhang, Tianyuan Gao, Ruiqi Barron, Jonathan T. Snavely, Noah Holynski, Aleksander |
| contents | Feed-forward transformer models have driven rapid progress in 3D vision, but state-of-the-art methods such as VGGT and $π^3$ have a computational cost that scales quadratically with the number of input images, making them inefficient when applied to large image collections. Sequential-reconstruction approaches reduce this cost but sacrifice reconstruction quality. We introduce ZipMap, a stateful feed-forward model that achieves linear-time, bidirectional 3D reconstruction while matching or surpassing the accuracy of quadratic-time methods. ZipMap employs test-time training layers to zip an entire image collection into a compact hidden scene state in a single forward pass, enabling reconstruction of over 700 frames in under 10 seconds on a single H100 GPU, more than $20\times$ faster than state-of-the-art methods such as VGGT. Moreover, we demonstrate the benefits of having a stateful representation in real-time scene-state querying and its extension to sequential streaming reconstruction. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2603_04385 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | ZipMap: Linear-Time Stateful 3D Reconstruction via Test-Time Training Jin, Haian Wu, Rundi Zhang, Tianyuan Gao, Ruiqi Barron, Jonathan T. Snavely, Noah Holynski, Aleksander Computer Vision and Pattern Recognition Artificial Intelligence Machine Learning Feed-forward transformer models have driven rapid progress in 3D vision, but state-of-the-art methods such as VGGT and $π^3$ have a computational cost that scales quadratically with the number of input images, making them inefficient when applied to large image collections. Sequential-reconstruction approaches reduce this cost but sacrifice reconstruction quality. We introduce ZipMap, a stateful feed-forward model that achieves linear-time, bidirectional 3D reconstruction while matching or surpassing the accuracy of quadratic-time methods. ZipMap employs test-time training layers to zip an entire image collection into a compact hidden scene state in a single forward pass, enabling reconstruction of over 700 frames in under 10 seconds on a single H100 GPU, more than $20\times$ faster than state-of-the-art methods such as VGGT. Moreover, we demonstrate the benefits of having a stateful representation in real-time scene-state querying and its extension to sequential streaming reconstruction. |
| title | ZipMap: Linear-Time Stateful 3D Reconstruction via Test-Time Training |
| topic | Computer Vision and Pattern Recognition Artificial Intelligence Machine Learning |
| url | https://arxiv.org/abs/2603.04385 |