Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Geometric Context Transformer for Streaming 3D Reconstruction

About

Streaming 3D reconstruction aims to recover 3D information, such as camera poses and point clouds, from a video stream, which necessitates geometric accuracy, temporal consistency, and computational efficiency. Motivated by the principles of Simultaneous Localization and Mapping (SLAM), we introduce LingBot-Map, a feed-forward 3D foundation model for reconstructing scenes from streaming data, built upon a geometric context transformer (GCT) architecture. A defining aspect of LingBot-Map lies in its carefully designed attention mechanism, which integrates an anchor context, a pose-reference window, and a trajectory memory to address coordinate grounding, dense geometric cues, and long-range drift correction, respectively. This design keeps the streaming state compact while retaining rich geometric context, enabling stable efficient inference at around 20 FPS on 518 x 378 resolution inputs over long sequences exceeding 10,000 frames. Extensive evaluations across a variety of benchmarks demonstrate that our approach achieves superior performance compared to both existing streaming and iterative optimization-based approaches.

Lin-Zhuo Chen, Jian Gao, Yihang Chen, Ka Leong Cheng, Yipengjing Sun, Liangxiao Hu, Nan Xue, Xing Zhu, Yujun Shen, Yao Yao, Yinghao Xu• 2026

Related benchmarks

TaskDatasetResultRank
3D Reconstruction7 Scenes--
94
3D ReconstructionNRGBD--
44
Pose EstimationETH3D
AUC @ Threshold 30.2779
41
3D ReconstructionETH3D
F1 Score98.98
25
Camera pose estimationOxford Spires sparse setting
AUC@1561.64
18
Pose and trajectory estimation7 Scenes
AUC312.63
9
Pose and trajectory estimationTanks&Temples
AUC345.8
9
Showing 7 of 7 rows

Other info

GitHub

Follow for update