MAGiSt3R: Multi-Agent Feed-forward 3D Reconstruction from Monocular RGB Videos

ECCV 2026

Ziren Gong1, Xiaohan Li2, Fabio Tosi1, Ninghui Xu3, Stefano Mattocia1, Jianfei Cai4, Matteo Poggi1,
1University of Bologna, 2University of Hongkong 3Southeast University 4Monash University
Paper Video Code (coming soon) Poster (coming soon) Results
NIS-SLAM architecture.

Pipeline of MAGiSt3R. Given RGB sequences from multiple agents, our method simultaneously predicts local point maps and camera poses. Then, the model merges submaps at the intra-agent level. On the server side, when loops are detected between agents, the same module fuses the local maps of multiple agents into a global map. Finally, pose graph optimization further mitigates cumulative drift.

Abstract

We present MAGiSt3R, a multi-agent 3D reconstruction framework performing reconstruction and camera tracking for monocular RGB videos at almost 10 FPS.

MAGiSt3R relies on a feed-forward model from the 3R family to process RGB videos and regress local point maps, and on a merging model, MAGMA, that combines local maps at both intra-agent and inter-agent levels to obtain the final global point map. Furthermore, MAGiSt3R performs pose graph optimization to mitigate cumulative camera drift occurring along the feed-forward pipeline.

We evaluate MAGiSt3R on both synthetic and real-world datasets, demonstrating its superior reconstruction and camera tracking accuracy compared to state-of-the-art approaches.

MAGMA

Reconstruction Result.

Annotations: Given a global submap, a keyframe set, and a local submap, the MAGMA module merges the local submap into a global map with poses in a unified coordinate system. First, we use retrieval to obtain the most relevant frames to form the inputs, reference set and registering set. Next, we encode their point maps and aggregate these geometric features with visual tokens, spatial tokens, and camera tokens. Then, we decode and predict global point maps.

Visualization

Reconstruction Result.

Annotations: We present 3D reconstruction results with multi-agent settings on ReplicaMultiagent. To compare with the existing baselines, we use RANSAC + ICP (RICP) to combine multi-agent reconstructions for these baselines when detecting loop closures. MAGiSt3R achieves more accurate recosntruction quality with less drifts between multiple agents.

Reconstruction Result.

Annotations: To test on more challenging scenes, we collect two sequences with less overlaps and dynamic objects in an indoor scene to simulate the multi-agent reconstruction. MAGiSt3R filters out dynamic objects and achieve accurate reconstructions.

Reconstruction & Tracking Results

Our final reconstruction and segmentation results are available now. You can find it in the above Results button or you can download it through Google Drive.

Related Links

There's a lot of excellent work that was introduced around the same time as ours.

DINO-SLAM: DINO-informed RGB-D SLAM for Neural Implicit and Explicit Representations provides two foundational paradigms for NeRF and 3DGS SLAM systems integrating geometry-aware DINO features.

HS-SLAM: Hybrid Representation with Structural Supervision for Improved Dense SLAM introduce HS-SLAM to enhance scene representations, capture structural information, and maintain global consistency.

How NeRFs and 3D Gaussian Splatting are Reshaping SLAM: a Survey provides the first comprehensive survey of SLAM progress through the lens of the latest advancements in radiance fields (NeRF and 3DGS).

Go-SLAM introduces a deep-learning-based dense visual SLAM framework that achieves real-time global optimization of poses and 3D reconstruction.

BibTeX

@inproceedings{gong2026magist3r,
  title={MAGiSt3R: Multi-Agent Feed-forward 3D Reconstruction from Monocular RGB Videos},
  author={Gong, Ziren and Li, Xiaohan and Tosi, Fabio and Xu, Ninghui and Mattoccia, Stefano and Cai, Jianfei and Poggi, Matteo},
  booktitle={European Conference on Computer Vision},
  year={2026}
}