Mem3R: Streaming 3D Reconstruction with Hybrid Memory via Test-Time Training

1Google, 2HKUST, 3USC

Interactive 3D Demo

W/O Mem3R
Loading…
W/ Mem3R
Loading…
Left Drag Orbit Wheel Zoom Right Drag Pan
Pool Pool
Comparison

Abstract

Method

Method overview figure goes here

Results

We quantitatively evaluate camera pose estimation on the ScanNet dataset and the TUM Dynamics dataset. Mem3R achieves significant improvement compared with CUT3R. Moreover, when combined with training-free, plug-and-play state update strategies TTT3R or TTSA3R, Mem3R further yields better results compared to applying these strategies to CUT3R.

Quantitative evaluation of camera pose estimation on ScanNet and TUM Dynamics datasets

Quantitative evaluation of camera pose estimation from the ScanNet dataset (left) and the TUM Dynamics dataset (right).

BibTeX

@article{mem3r2025,
  author    = {Author 1 and Author 2 and Author 3},
  title     = {Mem3R},
  journal   = {arXiv preprint},
  year      = {2025},
}