Lukas Neumann

Simon Rabanser, Lukas Neumann, Markus Haltmeier

The development of accurate and efficient image reconstruction algorithms is a central aspect of quantitative photoacoustic tomography (QPAT). In this paper, we address this issues for multi-source QPAT using the radiative transfer equation (RTE) as accurate model for light transport. The tissue parameters are jointly reconstructed from the acoustical data measured for each of the applied sources. We develop stochastic proximal gradient methods for multi-source QPAT, which are more efficient than standard proximal gradient methods in which a single iterative update has complexity proportional to the number applies sources. Additionally, we introduce a completely new formulation of QPAT as multilinear (MULL) inverse problem which avoids explicitly solving the RTE. The MULL formulation of QPAT is again addressed with stochastic proximal gradient methods. Numerical results for both approaches are presented. Besides the introduction of stochastic proximal gradient algorithms to QPAT, we consider the new MULL formulation of QPAT as main contribution of this paper.

Lukas Picek, Lukas Neumann, Jiri Matas

We propose a method that robustly exploits background and foreground in visual identification of individual animals. Experiments show that their automatic separation, made easy with methods like Segment Anything, together with independent foreground and background-related modeling, improves results. The two predictions are combined in a principled way, thanks to novel Per-Instance Temperature Scaling that helps the classifier to deal with appearance ambiguities in training and to produce calibrated outputs in the inference phase. For identity prediction from the background, we propose novel spatial and temporal models. On two problems, the relative error w.r.t. the baseline was reduced by 22.3% and 8.8%, respectively. For cases where objects appear in new locations, an example of background drift, accuracy doubles.

Markus Haltmeier, Lukas Neumann, Nadja Gruber et al.

Machine learning has achieved impressive performance in tomographic reconstruction, but supervised training requires paired measurements and ground-truth images that are often unavailable. This has motivated self-supervised approaches, which have primarily addressed denoising and, more recently, linear inverse problems. We address nonlinear inverse problems and introduce SPLIT (Self-supervised Partitioning for Learned Inversion in Nonlinear Tomography), a self-supervised machine-learning framework for reconstructing images from nonlinear, incomplete, and noisy projection data without any samples of ground-truth images. SPLIT enforces cross-partition consistency and measurement-domain fidelity while exploiting complementary information across multiple partitions. Our main theoretical result shows that, under mild conditions, the proposed self-supervised objective is equivalent to its supervised counterpart in expectation. We regularize training with an automatic stopping rule that halts optimization when a no-reference image-quality surrogate saturates. As a concrete application, we derive SPLIT variants for multispectral computed tomography. Experiments on sparse-view acquisitions demonstrate high reconstruction quality and robustness to noise, surpassing classical iterative reconstruction and recent self-supervised baselines.

Jan Skvrna, Lukas Neumann

Inferring object 3D position and orientation from a single RGB camera is a foundational task in computer vision with many important applications. Traditionally, 3D object detection methods are trained in a fully-supervised setup, requiring LiDAR and vast amounts of human annotations, which are laborious, costly, and do not scale well with the ever-increasing amounts of data being captured. We present a novel method to train a 3D object detector from a single RGB camera without domain-specific human annotations, making orders of magnitude more data available for training. The method uses newly proposed Local Object Motion Model to disentangle object movement source between subsequent frames, is approximately 700 times faster than previous work and compensates camera focal length differences to aggregate multiple datasets. The method is evaluated on three public datasets, where despite using no human labels, it outperforms prior work by a significant margin. It also shows its versatility as a pre-training tool for fully-supervised training and shows that combining pseudo-labels from multiple datasets can achieve comparable accuracy to using human labels from a single dataset. The source code and model are available at https://github.com/jskvrna/MonoSOWA.

Ondrej Tybl, Lukas Neumann

Knowledge Distillation (KD) transfers knowledge from a large teacher model to a smaller student by aligning their predictive distributions. However, conventional KD formulations - typically based on Kullback-Leibler divergence - assume that the teacher provides reliable soft targets. In practice, teacher predictions are often noisy or overconfident, and existing correction-based approaches rely on ad-hoc heuristics and extensive hyper-parameter tuning, which hinders generalization. We introduce REDistill (Robust Estimator Distillation), a simple yet principled framework grounded in robust statistics. REDistill replaces the standard KD objective with a power divergence loss, a generalization of KL divergence that adaptively downweights unreliable teacher output while preserving informative logit relationships. This formulation provides a unified and interpretable treatment of teacher noise, requires only logits, integrates seamlessly into existing KD pipelines, and incurs negligible computational overhead. Extensive experiments on CIFAR-100 and ImageNet-1k demonstrate that REDistill consistently improves student accuracy in diverse teacher-student architectures. Remarkably, it achieves these gains without model-specific hyper-parameter tuning, underscoring its robustness and strong generalization to unseen teacher-student pairs.

Jan Skvrna, Lukas Neumann

This paper presents the winning solution for the S23DR Challenge 2025, which involves predicting a house's 3D roof wireframe from a sparse point cloud and semantic segmentations. Our method operates directly in 3D, first identifying vertex candidates from the COLMAP point cloud using Gestalt segmentations. We then employ two PointNet-like models: one to refine and classify these candidates by analyzing local cubic patches, and a second to predict edges by processing the cylindrical regions connecting vertex pairs. This two-stage, 3D deep learning approach achieved a winning Hybrid Structure Score (HSS) of 0.43 on the private leaderboard.

Robert McCraith, Lukas Neumann, Andrea Vedaldi

Vision is one of the primary sensing modalities in autonomous driving. In this paper we look at the problem of estimating the velocity of road vehicles from a camera mounted on a moving car. Contrary to prior methods that train end-to-end deep networks that estimate the vehicles' velocity from the video pixels, we propose a two-step approach where first an off-the-shelf tracker is used to extract vehicle bounding boxes and then a small neural network is used to regress the vehicle velocity from the tracked bounding boxes. Surprisingly, we find that this still achieves state-of-the-art estimation performance with the significant benefit of separating perception from dynamics estimation via a clean, interpretable and verifiable interface which allows us distill the statistics which are crucial for velocity estimation. We show that the latter can be used to easily generate synthetic training data in the space of bounding boxes and use this to improve the performance of our method further.

Robert McCraith, Eldar Insafutdinov, Lukas Neumann et al.

We present a system for automatic converting of 2D mask object predictions and raw LiDAR point clouds into full 3D bounding boxes of objects. Because the LiDAR point clouds are partial, directly fitting bounding boxes to the point clouds is meaningless. Instead, we suggest that obtaining good results requires sharing information between \emph{all} objects in the dataset jointly, over multiple frames. We then make three improvements to the baseline. First, we address ambiguities in predicting the object rotations via direct optimization in this space while still backpropagating rotation prediction through the model. Second, we explicitly model outliers and task the network with learning their typical patterns, thus better discounting them. Third, we enforce temporal consistency when video data is available. With these contributions, our method significantly outperforms previous work despite the fact that those methods use significantly more complex pipelines, 3D models and additional human-annotated external sources of prior information.

Robert McCraith, Lukas Neumann, Andrea Vedaldi

In the recent years, many methods demonstrated the ability of neural networks to learn depth and pose changes in a sequence of images, using only self-supervision as the training signal. Whilst the networks achieve good performance, the often over-looked detail is that due to the inherent ambiguity of monocular vision they predict depth up to an unknown scaling factor. The scaling factor is then typically obtained from the LiDAR ground truth at test time, which severely limits practical applications of these methods. In this paper, we show that incorporating prior information about the camera configuration and the environment, we can remove the scale ambiguity and predict depth directly, still using the self-supervised formulation and not relying on any additional sensors.

Robert McCraith, Lukas Neumann, Andrew Zisserman et al.

Recent advances in self-supervised learning havedemonstrated that it is possible to learn accurate monoculardepth reconstruction from raw video data, without using any 3Dground truth for supervision. However, in robotics applications,multiple views of a scene may or may not be available, depend-ing on the actions of the robot, switching between monocularand multi-view reconstruction. To address this mixed setting,we proposed a new approach that extends any off-the-shelfself-supervised monocular depth reconstruction system to usemore than one image at test time. Our method builds on astandard prior learned to perform monocular reconstruction,but uses self-supervision at test time to further improve thereconstruction accuracy when multiple images are available.When used to update the correct components of the model, thisapproach is highly-effective. On the standard KITTI bench-mark, our self-supervised method consistently outperformsall the previous methods with an average 25% reduction inabsolute error for the three common setups (monocular, stereoand monocular+stereo), and comes very close in accuracy whencompared to the fully-supervised state-of-the-art methods.

Andreas Veit, Tomas Matera, Lukas Neumann et al.

This paper describes the COCO-Text dataset. In recent years large-scale datasets like SUN and Imagenet drove the advancement of scene understanding and object recognition. The goal of COCO-Text is to advance state-of-the-art in text detection and recognition in natural images. The dataset is based on the MS COCO dataset, which contains images of complex everyday scenes. The images were not collected with text in mind and thus contain a broad variety of text instances. To reflect the diversity of text in natural scenes, we annotate text with (a) location in terms of a bounding box, (b) fine-grained classification into machine printed text and handwritten text, (c) classification into legible and illegible text, (d) script of the text and (e) transcriptions of legible text. The dataset contains over 173k text annotations in over 63k images. We provide a statistical analysis of the accuracy of our annotations. In addition, we present an analysis of three leading state-of-the-art photo Optical Character Recognition (OCR) approaches on our dataset. While scene text detection and recognition enjoys strong advances in recent years, we identify significant shortcomings motivating future work.