ZipMap: Linear-Time Stateful 3D Reconstruction with Test-Time Training
2026-03-04 • Computer Vision and Pattern Recognition
Computer Vision and Pattern RecognitionArtificial IntelligenceMachine Learning
AI summaryⓘ
The authors present ZipMap, a new method for 3D reconstruction from multiple images that is much faster than previous models. Unlike earlier approaches that slow down a lot as more images are added, ZipMap works in a way that scales linearly, allowing it to process hundreds of images quickly without losing accuracy. It does this by creating a compact representation of the scene in one pass and uses special layers that adjust during testing. The authors also show that ZipMap's design helps with tasks like real-time scene updates and handling streaming data.
3D reconstructiontransformer modelsfeed-forward networkscomputational complexitytest-time trainingscene representationreal-time processingstreaming reconstructionVGGTH100 GPU
Authors
Haian Jin, Rundi Wu, Tianyuan Zhang, Ruiqi Gao, Jonathan T. Barron, Noah Snavely, Aleksander Holynski
Abstract
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.