Martin R. Oswald
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Martin R. Oswald has a total of 56 co-authors. Here are the top co-authors by number of shared publications:
- Marc Pollefeys
ETH Zurich
6 shared publications - Denys Rozumnyi
Meta Reality Labs Zurich
2 shared publications Fatemeh Karimi Nejadasl
University of Amsterdam
2 shared publications- Weijie Wei
University of Amsterdam
2 shared publications Cristian Sminchisescu
Google Research Zurich, Zurich, Switzerland
1 shared publicationDaniel Cremers
University of Bonn
1 shared publication
Latest Publications
- Open Access
- 1 October 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Robotics and Automation Letters
- Vol. 11 (10) , 11243-11250
Generation of Uncertainty-Aware High-Level Spatial Concepts in Factorized 3D Scene Graphs via Graph Neural Networks
- 1 October 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Robotics and Automation Letters
- Vol. 11 (10) , 11243-11250
Open AccessGeneration of Uncertainty-Aware High-Level Spatial Concepts in Factorized 3D Scene Graphs via Graph Neural Networks
- Top Cited Papers
- 18 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Robotics
- Vol. 42 (15523098) , 1405-1427
How NeRFs and 3-D Gaussian Splatting Are Reshaping SLAM: A Survey
Over the past two decades, research in the field of Simultaneous Localization and Mapping (SLAM) has undergone a significant evolution, highlighting its critical role in enabling autonomous exploration of unknown environments. This evo-lution ranges from hand-crafted methods, through the era of deep learning, to more recent developments focused on Neural Radiance Fields (NeRFs) and 3D Gaussian Splatting (3DGS) rep-resentations. Recognizing the growing body of research and the absence of a comprehensive survey on the topic, this paper aims to provide the first comprehensive overview of SLAM progress through the lens of the latest advancements in radiance fields. It sheds light on the background, evolutionary path, inherent strengths and limitations, and serves as a fundamental reference to highlight the dynamic progress and specific challenges.
- 18 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Robotics
- Vol. 42 (15523098) , 1405-1427
Top Cited PapersHow NeRFs and 3-D Gaussian Splatting Are Reshaping SLAM: A Survey
Over the past two decades, research in the field of Simultaneous Localization and Mapping (SLAM) has undergone a significant evolution, highlighting its critical role in enabling autonomous exploration of unknown environments. This evo-lution ranges from hand-crafted methods, through the era of deep learning, to more recent developments focused on Neural Radiance Fields (NeRFs) and 3D Gaussian Splatting (3DGS) rep-resentations. Recognizing the growing body of research and the absence of a comprehensive survey on the topic, this paper aims to provide the first comprehensive overview of SLAM progress through the lens of the latest advancements in radiance fields. It sheds light on the background, evolutionary path, inherent strengths and limitations, and serves as a fundamental reference to highlight the dynamic progress and specific challenges.
- 1 January 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Robotics and Automation Letters
- p. 1-8
Open-Vocabulary Online Semantic Mapping for SLAM
- 1 January 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Robotics and Automation Letters
- p. 1-8
Open-Vocabulary Online Semantic Mapping for SLAM
- 1 November 2024
- Published by Springer Nature
- p. 178-195
T-MAE : Temporal Masked Autoencoders for Point Cloud Representation Learning
- 1 November 2024
- Published by Springer Nature
- p. 178-195
T-MAE : Temporal Masked Autoencoders for Point Cloud Representation Learning
- Top Cited PapersOpen Access
- 10 November 2023
- Published by Elsevier in Medical Image Analysis
- Vol. 91, 103027
Automatic registration with continuous pose updates for marker-less surgical navigation in spine surgery
- 10 November 2023
- Published by Elsevier in Medical Image Analysis
- Vol. 91, 103027
Top Cited PapersOpen AccessAutomatic registration with continuous pose updates for marker-less surgical navigation in spine surgery
- Open Access
- 12 October 2023
- Published by MDPI AG in Applied Sciences
- Vol. 13 (20) , 11215
Cross-View Outdoor Localization in Augmented Reality by Fusing Map and Satellite Data
Visual positioning is the task of finding the location of a given image and is necessary for augmented reality applications. Traditional algorithms solve this problem by matching against premade 3D point clouds or panoramic images. Recently, more attention has been given to models that match the ground-level image with overhead imagery. In this paper, we introduce AlignNet, which builds upon previous work to bridge the gap between ground-level and top-level images. By making multiple key insights, we push the model results to achieve up to 4 times higher recall rates on a visual position dataset. We use a fusion of both satellite and map data from OpenStreetMap for this matching by extending the previously available satellite database with corresponding map data. The model pushes the input images through a two-branch U-Net and is able to make matches using a geometric projection module to map the top-level image to the ground-level domain at a given position. By calculating the difference between the projection and ground-level image in a differentiable fashion, we can use a Levenberg–Marquardt (LM) module to iteratively align the estimated position towards the ground-truth position. This sample-wise optimization strategy allows the model to align the position better than if the model has to obtain the location in a single step. We provide key insights into the model’s behavior, which allows us to increase the model’s ability to obtain competitive results on the KITTI cross-view dataset. We compare our obtained results with the state of the art and obtain new best results on 3 of the 9 categories we look at, which include a 57% likelihood of lateral localization within 1 m in a 40 m × 40 m area and a 93% azimuth localization within 3 when using a 20 rotation noise prior.- 12 October 2023
- Published by MDPI AG in Applied Sciences
- Vol. 13 (20) , 11215
Open AccessCross-View Outdoor Localization in Augmented Reality by Fusing Map and Satellite Data
Visual positioning is the task of finding the location of a given image and is necessary for augmented reality applications. Traditional algorithms solve this problem by matching against premade 3D point clouds or panoramic images. Recently, more attention has been given to models that match the ground-level image with overhead imagery. In this paper, we introduce AlignNet, which builds upon previous work to bridge the gap between ground-level and top-level images. By making multiple key insights, we push the model results to achieve up to 4 times higher recall rates on a visual position dataset. We use a fusion of both satellite and map data from OpenStreetMap for this matching by extending the previously available satellite database with corresponding map data. The model pushes the input images through a two-branch U-Net and is able to make matches using a geometric projection module to map the top-level image to the ground-level domain at a given position. By calculating the difference between the projection and ground-level image in a differentiable fashion, we can use a Levenberg–Marquardt (LM) module to iteratively align the estimated position towards the ground-truth position. This sample-wise optimization strategy allows the model to align the position better than if the model has to obtain the location in a single step. We provide key insights into the model’s behavior, which allows us to increase the model’s ability to obtain competitive results on the KITTI cross-view dataset. We compare our obtained results with the state of the art and obtain new best results on 3 of the 9 categories we look at, which include a 57% likelihood of lateral localization within 1 m in a 40 m × 40 m area and a 93% azimuth localization within 3 when using a 20 rotation noise prior. - Open Access
- 2 October 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1747-1756
APNet: Urban-level Scene Segmentation of Aerial Images and Point Clouds
In this paper, we focus on semantic segmentation method for point clouds of urban scenes. Our fundamental concept revolves around the collaborative utilization of diverse scene representations to benefit from different context information and network architectures. To this end, the proposed network architecture, called APNet, is split into two branches: a point cloud branch and an aerial image branch which input is generated from a point cloud. To leverage the different properties of each branch, we employ a geometry-aware fusion module that is learned to combine the results of each branch. Additional separate losses for each branch avoid that one branch dominates the results, ensure the best performance for each branch individually and explicitly define the input domain of the fusion network assuring it only performs data fusion. Our experiments demonstrate that the fusion output consistently outperforms the individual network branches and that APNet achieves state-of-the-art performance of 65.2 mIoU on the SensatUrban dataset. Upon acceptance, the source code will be made accessible.
- 2 October 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1747-1756
Open AccessAPNet: Urban-level Scene Segmentation of Aerial Images and Point Clouds
In this paper, we focus on semantic segmentation method for point clouds of urban scenes. Our fundamental concept revolves around the collaborative utilization of diverse scene representations to benefit from different context information and network architectures. To this end, the proposed network architecture, called APNet, is split into two branches: a point cloud branch and an aerial image branch which input is generated from a point cloud. To leverage the different properties of each branch, we employ a geometry-aware fusion module that is learned to combine the results of each branch. Additional separate losses for each branch avoid that one branch dominates the results, ensure the best performance for each branch individually and explicitly define the input domain of the fusion network assuring it only performs data fusion. Our experiments demonstrate that the fusion output consistently outperforms the individual network branches and that APNet achieves state-of-the-art performance of 65.2 mIoU on the SensatUrban dataset. Upon acceptance, the source code will be made accessible.
- 20 September 2022
- Published by Springer Nature
- p. 317-333
NeuralMeshing: Differentiable Meshing of Implicit Neural Representations
- 20 September 2022
- Published by Springer Nature
- p. 317-333
NeuralMeshing: Differentiable Meshing of Implicit Neural Representations
- Open Access
CompNVS: Novel View Synthesis with Scene Completion
We introduce a scalable framework for novel view synthesis from RGB-D images with largely incomplete scene coverage. While generative neural approaches have demonstrated spectacular results on 2D images, they have not yet achieved similar photorealistic results in combination with scene completion where a spatial 3D scene understanding is essential. To this end, we propose a generative pipeline performing on a sparse grid-based neural scene representation to complete unobserved scene parts via a learned distribution of scenes in a 2.5D-3D-2.5D manner. We process encoded image features in 3D space with a geometry completion network and a subsequent texture inpainting network to extrapolate the missing area. Photorealistic image sequences can be finally obtained via consistency-relevant differentiable rendering. Comprehensive experiments show that the graphical outputs of our method outperform the state of the art, especially within unobserved scene parts.
Open AccessCompNVS: Novel View Synthesis with Scene Completion
We introduce a scalable framework for novel view synthesis from RGB-D images with largely incomplete scene coverage. While generative neural approaches have demonstrated spectacular results on 2D images, they have not yet achieved similar photorealistic results in combination with scene completion where a spatial 3D scene understanding is essential. To this end, we propose a generative pipeline performing on a sparse grid-based neural scene representation to complete unobserved scene parts via a learned distribution of scenes in a 2.5D-3D-2.5D manner. We process encoded image features in 3D space with a geometry completion network and a subsequent texture inpainting network to extrapolate the missing area. Photorealistic image sequences can be finally obtained via consistency-relevant differentiable rendering. Comprehensive experiments show that the graphical outputs of our method outperform the state of the art, especially within unobserved scene parts.
- 28 December 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Robotics and Automation Letters
- Vol. 7 (2) , 1332-1339
A Real-Time Online Learning Framework for Joint 3D Reconstruction and Semantic Segmentation of Indoor Scenes
This letter presents a real-time online vision framework to jointly recover an indoor scene’s 3D structure and semantic label. Given noisy depth maps, a camera trajectory, and 2D semantic labels at train time, the proposed deep neural network based approach learns to fuse the depth over frames with suitable semantic labels in the scene space. Our approach exploits the joint volumetric representation of the depth and semantics in the scene feature space to solve this task. For a compelling online fusion of the semantic labels and geometry in real-time, we introduce an efficient vortex pooling block while dropping the use of routing network in online depth fusion to preserve high-frequency surface details. We show that the context information provided by the semantics of the scene helps the depth fusion network learn noise-resistant features. Not only that, it helps overcome the shortcomings of the current online depth fusion method in dealing with thin object structures, thickening artifacts, and false surfaces. Experimental evaluation on the Replica dataset shows that our approach can perform depth fusion at 37 and 10 frames per second with an average reconstruction F-score of 88% and 91%, respectively, depending on the depth map resolution. Moreover, our model shows an average IoU score of 0.515 on the ScanNet 3D semantic benchmark leaderboard. Code and example dataset information is available at https://github.com/suryanshkumar/online-joint-depthfusion-and-semantic .
- 28 December 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Robotics and Automation Letters
- Vol. 7 (2) , 1332-1339
A Real-Time Online Learning Framework for Joint 3D Reconstruction and Semantic Segmentation of Indoor Scenes
This letter presents a real-time online vision framework to jointly recover an indoor scene’s 3D structure and semantic label. Given noisy depth maps, a camera trajectory, and 2D semantic labels at train time, the proposed deep neural network based approach learns to fuse the depth over frames with suitable semantic labels in the scene space. Our approach exploits the joint volumetric representation of the depth and semantics in the scene feature space to solve this task. For a compelling online fusion of the semantic labels and geometry in real-time, we introduce an efficient vortex pooling block while dropping the use of routing network in online depth fusion to preserve high-frequency surface details. We show that the context information provided by the semantics of the scene helps the depth fusion network learn noise-resistant features. Not only that, it helps overcome the shortcomings of the current online depth fusion method in dealing with thin object structures, thickening artifacts, and false surfaces. Experimental evaluation on the Replica dataset shows that our approach can perform depth fusion at 37 and 10 frames per second with an average reconstruction F-score of 88% and 91%, respectively, depending on the depth map resolution. Moreover, our model shows an average IoU score of 0.515 on the ScanNet 3D semantic benchmark leaderboard. Code and example dataset information is available at https://github.com/suryanshkumar/online-joint-depthfusion-and-semantic .