a DINO-informed design strategy to enhance implicit (Neural Radiance Field -- NeRF) and explicit representations (Gaussian Splatting -- GS) in SLAM systems through the more comprehensive semantics understanding enabled by DINO.
This latter alone, however, lacks proper 3D geometry understanding, allowing only for marginal improvements. Therefore, we rely on a Scene Geometry Encoder (SGE) to enrich DINO features into geometry-aware DINO features (geoDINO), to better understand those geometric relationships that vanilla DINO features fail to capture. Building upon it, we propose two foundational paradigms for NeRF and GS SLAM systems integrating geoDINO features.
Compared to state-of-the-art methods, our DINO-informed pipelines achieve superior performance on the Replica, ScanNet, and TUM datasets.
Scene Geometry Encoder (SGE): Our SGE has two outputs: 1) DINO features, simply extracted from RGB frames by the DINO model, 2) geoDINO features obtained from the joint processing of DINO features and depth maps, projected respectively into appearance and geometric features, and used as keys and queries to attend over.
Framework of Nueral Implicit and Explicit Representations: In the NeRF-based pipeline (left), we leverage EDINO features to supervise the tri-plane (red double arrow) while DINO features guide the optimization of the estimated DINO feature map (red double arrow). In the 3DGS-based pipeline (right), we incorporate the corresponding EDINO feature into the parameters of each Gaussian (blue arrow). Meanwhile, the DINO features are leveraged to guide the learning of DINO information within the representation (blue double arrow).
Visualization of DINO-SLAM and baselines on the Replica dataset.: We present details of reconstruction and rendering quality with red boxes and blue boxes. The two left columns present the reconstruction performance of our DINO-SLAM (NeRF). Our method yields superior mesh results with detailed textures and better completeness. The two right columns show the rendering results of our DINO-SLAM (GS). Our DINO-SLAM demonstrates high-fidelity and realistic images, especially on text tags and object textures.
More results and performance on Replica. DINO-SLAM achieves superior camera tracking (left, evaluated on scannet) and 3D reconstruction quality (middle, evaluated on Replica). Our approach is general: it can be integrated into existing NeRF/GS SLAM systems and boost their results (right, showing some representative GS methods being improved by our proposal).
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.
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.
@article{gong2025dino,
title={DINO-SLAM: DINO-informed RGB-D SLAM for Neural Implicit and Explicit Representations},
author={Gong, Ziren and Li, Xiaohan and Tosi, Fabio and Zhang, Youmin and Mattoccia, Stefano and Wu, Jun and Poggi, Matteo},
journal={arXiv preprint arXiv:2507.19474},
year={2025}
}