Stefan Roth

Sherwin Bahmani, Oliver Hahn, Eduard Zamfir et al.

The lack of out-of-domain generalization is a critical weakness of deep networks for semantic segmentation. Previous studies relied on the assumption of a static model, i. e., once the training process is complete, model parameters remain fixed at test time. In this work, we challenge this premise with a self-adaptive approach for semantic segmentation that adjusts the inference process to each input sample. Self-adaptation operates on two levels. First, it fine-tunes the parameters of convolutional layers to the input image using consistency regularization. Second, in Batch Normalization layers, self-adaptation interpolates between the training and the reference distribution derived from a single test sample. Despite both techniques being well known in the literature, their combination sets new state-of-the-art accuracy on synthetic-to-real generalization benchmarks. Our empirical study suggests that self-adaptation may complement the established practice of model regularization at training time for improving deep network generalization to out-of-domain data. Our code and pre-trained models are available at https://github.com/visinf/self-adaptive.

Christoph Reich, Oliver Hahn, Nikita Araslanov et al.

Video panoptic segmentation (VPS) aims to jointly detect, segment, and track all objects while partitioning the video into semantically consistent regions. We introduce the task setting of unsupervised VPS, omitting any human supervision. Existing unsupervised scene understanding works mainly focused on image segmentation tasks; the video domain remains underexplored. We propose VideoCUPS, the first unsupervised VPS approach. VideoCUPS generates temporally consistent panoptic video pseudo-labels from scene-centric videos by exploiting unsupervised depth, motion, and visual cues. Training on these pseudo-labels using a novel Video DropLoss yields an accurate, unsupervised VPS model. To benchmark progress, we introduce a comprehensive evaluation protocol and four competitive baselines, extending state-of-the-art unsupervised panoptic image and instance video segmentation models to VPS. VideoCUPS outperforms all baselines and demonstrates strong label-efficient learning. With VideoCUPS, our evaluation protocol, and baselines, we provide a strong foundation for future research on unsupervised VPS.

Robin Hesse, Simone Schaub-Meyer, Stefan Roth

The field of explainable artificial intelligence (XAI) aims to uncover the inner workings of complex deep neural models. While being crucial for safety-critical domains, XAI inherently lacks ground-truth explanations, making its automatic evaluation an unsolved problem. We address this challenge by proposing a novel synthetic vision dataset, named FunnyBirds, and accompanying automatic evaluation protocols. Our dataset allows performing semantically meaningful image interventions, e.g., removing individual object parts, which has three important implications. First, it enables analyzing explanations on a part level, which is closer to human comprehension than existing methods that evaluate on a pixel level. Second, by comparing the model output for inputs with removed parts, we can estimate ground-truth part importances that should be reflected in the explanations. Third, by mapping individual explanations into a common space of part importances, we can analyze a variety of different explanation types in a single common framework. Using our tools, we report results for 24 different combinations of neural models and XAI methods, demonstrating the strengths and weaknesses of the assessed methods in a fully automatic and systematic manner.

Gopika Sudhakaran, Devendra Singh Dhami, Kristian Kersting et al.

Recent years have seen a growing interest in Scene Graph Generation (SGG), a comprehensive visual scene understanding task that aims to predict entity relationships using a relation encoder-decoder pipeline stacked on top of an object encoder-decoder backbone. Unfortunately, current SGG methods suffer from an information loss regarding the entities local-level cues during the relation encoding process. To mitigate this, we introduce the Vision rElation TransfOrmer (VETO), consisting of a novel local-level entity relation encoder. We further observe that many existing SGG methods claim to be unbiased, but are still biased towards either head or tail classes. To overcome this bias, we introduce a Mutually Exclusive ExperT (MEET) learning strategy that captures important relation features without bias towards head or tail classes. Experimental results on the VG and GQA datasets demonstrate that VETO + MEET boosts the predictive performance by up to 47 percentage over the state of the art while being 10 times smaller.

Claudia Cuttano, Gabriele Trivigno, Carlo Masone et al.

Recent advances in semantic correspondence rely on dual-encoder architectures, combining DINOv2 with diffusion backbones. While accurate, these billion-parameter models generalize poorly beyond training keypoints, revealing a gap between benchmark performance and real-world usability, where queried points rarely match those seen during training. Building upon DINOv2, we introduce MARCO, a unified model for generalizable correspondence driven by a novel training framework that enhances both fine-grained localization and semantic generalization. By coupling a coarse-to-fine objective that refines spatial precision with a self-distillation framework, which expands sparse supervision beyond annotated regions, our approach transforms a handful of keypoints into dense, semantically coherent correspondences. MARCO sets a new state of the art on SPair-71k, AP-10K, and PF-PASCAL, with gains that amplify at fine-grained localization thresholds (+8.9 PCK@0.01), strongest generalization to unseen keypoints (+5.1, SPair-U) and categories (+4.7, MP-100), while remaining 3x smaller and 10x faster than diffusion-based approaches. Code is available at https://github.com/visinf/MARCO .

Moritz Nottebaum, Stefan Roth, Simone Schaub-Meyer

Most deep learning methods for video frame interpolation consist of three main components: feature extraction, motion estimation, and image synthesis. Existing approaches are mainly distinguishable in terms of how these modules are designed. However, when interpolating high-resolution images, e.g. at 4K, the design choices for achieving high accuracy within reasonable memory requirements are limited. The feature extraction layers help to compress the input and extract relevant information for the latter stages, such as motion estimation. However, these layers are often costly in parameters, computation time, and memory. We show how ideas from dimensionality reduction combined with a lightweight optimization can be used to compress the input representation while keeping the extracted information suitable for frame interpolation. Further, we require neither a pretrained flow network nor a synthesis network, additionally reducing the number of trainable parameters and required memory. When evaluating on three 4K benchmarks, we achieve state-of-the-art image quality among the methods without pretrained flow while having the lowest network complexity and memory requirements overall.

Claudia Cuttano, Gabriele Trivigno, Christoph Reich et al.

In-context segmentation (ICS) aims to segment arbitrary concepts, e.g., objects, parts, or personalized instances, given one annotated visual examples. Existing work relies on (i) fine-tuning vision foundation models (VFMs), which improves in-domain results but harms generalization, or (ii) combines multiple frozen VFMs, which preserves generalization but yields architectural complexity and fixed segmentation granularities. We revisit ICS from a minimalist perspective and ask: Can a single self-supervised backbone support both semantic matching and segmentation, without any supervision or auxiliary models? We show that scaled-up dense self-supervised features from DINOv3 exhibit strong spatial structure and semantic correspondence. We introduce INSID3, a training-free approach that segments concepts at varying granularities only from frozen DINOv3 features, given an in-context example. INSID3 achieves state-of-the-art results across one-shot semantic, part, and personalized segmentation, outperforming previous work by +7.5 % mIoU, while using 3x fewer parameters and without any mask or category-level supervision. Code is available at https://github.com/visinf/INSID3 .

Franz Klein, Shweta Mahajan, Stefan Roth

Stylized image captioning as presented in prior work aims to generate captions that reflect characteristics beyond a factual description of the scene composition, such as sentiments. Such prior work relies on given sentiment identifiers, which are used to express a certain global style in the caption, e.g. positive or negative, however without taking into account the stylistic content of the visual scene. To address this shortcoming, we first analyze the limitations of current stylized captioning datasets and propose COCO attribute-based augmentations to obtain varied stylized captions from COCO annotations. Furthermore, we encode the stylized information in the latent space of a Variational Autoencoder; specifically, we leverage extracted image attributes to explicitly structure its sequential latent space according to different localized style characteristics. Our experiments on the Senticap and COCO datasets show the ability of our approach to generate accurate captions with diversity in styles that are grounded in the image.

Robin Hesse, Simone Schaub-Meyer, Janina Hesse et al.

Explainable artificial intelligence (XAI) aims to provide human-interpretable insights into the behavior of deep neural networks (DNNs), typically by estimating a simplified causal structure of the model. In existing work, this causal structure often includes relationships where the presence of a concept is associated with a strong activation of a neuron. For example, attribution methods primarily identify input pixels that contribute most to a prediction, and feature visualization methods reveal inputs that cause high activation of a target neuron - the former implicitly assuming that the relevant information resides in the input, and the latter that neurons encode the presence of concepts. However, a largely overlooked type of causal relationship is that of encoded absences, where the absence of a concept increases neural activation. In this work, we show that such missing but relevant concepts are common and that mainstream XAI methods struggle to reveal them when applied in their standard form. To address this, we propose two simple extensions to attribution and feature visualization techniques that uncover encoded absences. Across experiments, we show how mainstream XAI methods can be used to reveal and explain encoded absences, how ImageNet models exploit them, and that debiasing can be improved when considering them.

Krishnakant Singh, Simone Schaub-Meyer, Stefan Roth

The high-quality images yielded by generative adversarial networks (GANs) have motivated investigations into their application for image editing. However, GANs are often limited in the control they provide for performing specific edits. One of the principal challenges is the entangled latent space of GANs, which is not directly suitable for performing independent and detailed edits. Recent editing methods allow for either controlled style edits or controlled semantic edits. In addition, methods that use semantic masks to edit images have difficulty preserving the identity and are unable to perform controlled style edits. We propose a method to disentangle a GAN$\text{'}$s latent space into semantic and style spaces, enabling controlled semantic and style edits for face images independently within the same framework. To achieve this, we design an encoder-decoder based network architecture ($S^2$-Flow), which incorporates two proposed inductive biases. We show the suitability of $S^2$-Flow quantitatively and qualitatively by performing various semantic and style edits.

Krishnakant Singh, Simone Schaub-Meyer, Stefan Roth

Object-centric learning aims to decompose an input image into a set of meaningful object files (slots). These latent object representations enable a variety of downstream tasks. Yet, object-centric learning struggles on real-world datasets, which contain multiple objects of complex textures and shapes in natural everyday scenes. To address this, we introduce Guided Latent Slot Diffusion (GLASS), a novel slot-attention model that learns in the space of generated images and uses semantic and instance guidance modules to learn better slot embeddings for various downstream tasks. Our experiments show that GLASS surpasses state-of-the-art slot-attention methods by a wide margin on tasks such as (zero-shot) object discovery and conditional image generation for real-world scenes. Moreover, GLASS enables the first application of slot attention to the compositional generation of complex, realistic scenes.

Robin Hesse, Simone Schaub-Meyer, Stefan Roth

Attribution maps are one of the most established tools to explain the functioning of computer vision models. They assign importance scores to input features, indicating how relevant each feature is for the prediction of a deep neural network. While much research has gone into proposing new attribution methods, their proper evaluation remains a difficult challenge. In this work, we propose a novel evaluation protocol that overcomes two fundamental limitations of the widely used incremental-deletion protocol, i.e., the out-of-domain issue and lacking inter-model comparisons. This allows us to evaluate 23 attribution methods and how different design choices of popular vision backbones affect their attribution quality. We find that intrinsically explainable models outperform standard models and that raw attribution values exhibit a higher attribution quality than what is known from previous work. Further, we show consistent changes in the attribution quality when varying the network design, indicating that some standard design choices promote attribution quality.

Tobias Lingenberg, Markus Reuter, Gopika Sudhakaran et al.

Simple data augmentation techniques, such as rotations and flips, are widely used to enhance the generalization power of computer vision models. However, these techniques often fail to modify high-level semantic attributes of a class. To address this limitation, researchers have explored generative augmentation methods like the recently proposed DA-Fusion. Despite some progress, the variations are still largely limited to textural changes, thus falling short on aspects like varied viewpoints, environment, weather conditions, or even class-level semantic attributes (eg, variations in a dog's breed). To overcome this challenge, we propose DIAGen, building upon DA-Fusion. First, we apply Gaussian noise to the embeddings of an object learned with Textual Inversion to diversify generations using a pre-trained diffusion model's knowledge. Second, we exploit the general knowledge of a text-to-text generative model to guide the image generation of the diffusion model with varied class-specific prompts. Finally, we introduce a weighting mechanism to mitigate the impact of poorly generated samples. Experimental results across various datasets show that DIAGen not only enhances semantic diversity but also improves the performance of subsequent classifiers. The advantages of DIAGen over standard augmentations and the DA-Fusion baseline are particularly pronounced with out-of-distribution samples.

Oliver Hahn, Nikita Araslanov, Simone Schaub-Meyer et al.

Unsupervised semantic segmentation aims to automatically partition images into semantically meaningful regions by identifying global semantic categories within an image corpus without any form of annotation. Building upon recent advances in self-supervised representation learning, we focus on how to leverage these large pre-trained models for the downstream task of unsupervised segmentation. We present PriMaPs - Principal Mask Proposals - decomposing images into semantically meaningful masks based on their feature representation. This allows us to realize unsupervised semantic segmentation by fitting class prototypes to PriMaPs with a stochastic expectation-maximization algorithm, PriMaPs-EM. Despite its conceptual simplicity, PriMaPs-EM leads to competitive results across various pre-trained backbone models, including DINO and DINOv2, and across different datasets, such as Cityscapes, COCO-Stuff, and Potsdam-3. Importantly, PriMaPs-EM is able to boost results when applied orthogonally to current state-of-the-art unsupervised semantic segmentation pipelines. Code is available at https://github.com/visinf/primaps.

Sascha Rosbach, Stefan M. Leupold, Simon Großjohann et al.

Automated vehicles operating in urban environments have to reliably interact with other traffic participants. Planning algorithms often utilize separate prediction modules forecasting probabilistic, multi-modal, and interactive behaviors of objects. Designing prediction and planning as two separate modules introduces significant challenges, particularly due to the interdependence of these modules. This work proposes a deep learning methodology to combine prediction and planning. A conditional GAN with the U-Net architecture is trained to predict two high-resolution image sequences. The sequences represent explicit motion predictions, mainly used to train context understanding, and pixel state values suitable for planning encoding kinematic reachability, object dynamics, safety, and driving comfort. The model can be trained offline on target images rendered by a sampling-based model-predictive planner, leveraging real-world driving data. Our results demonstrate intuitive behavior in complex situations, such as lane changes amidst conflicting objectives.

Oliver Hahn, Christoph Reich, Nikita Araslanov et al.

Unsupervised panoptic segmentation aims to partition an image into semantically meaningful regions and distinct object instances without training on manually annotated data. In contrast to prior work on unsupervised panoptic scene understanding, we eliminate the need for object-centric training data, enabling the unsupervised understanding of complex scenes. To that end, we present the first unsupervised panoptic method that directly trains on scene-centric imagery. In particular, we propose an approach to obtain high-resolution panoptic pseudo labels on complex scene-centric data, combining visual representations, depth, and motion cues. Utilizing both pseudo-label training and a panoptic self-training strategy yields a novel approach that accurately predicts panoptic segmentation of complex scenes without requiring any human annotations. Our approach significantly improves panoptic quality, e.g., surpassing the recent state of the art in unsupervised panoptic segmentation on Cityscapes by 9.4% points in PQ.

Christoph Reich, Oliver Hahn, Daniel Cremers et al.

Resource-constrained hardware, such as edge devices or cell phones, often rely on cloud servers to provide the required computational resources for inference in deep vision models. However, transferring image and video data from an edge or mobile device to a cloud server requires coding to deal with network constraints. The use of standardized codecs, such as JPEG or H.264, is prevalent and required to ensure interoperability. This paper aims to examine the implications of employing standardized codecs within deep vision pipelines. We find that using JPEG and H.264 coding significantly deteriorates the accuracy across a broad range of vision tasks and models. For instance, strong compression rates reduce semantic segmentation accuracy by more than 80% in mIoU. In contrast to previous findings, our analysis extends beyond image and action classification to localization and dense prediction tasks, thus providing a more comprehensive perspective.

Aleksandar Jevtić, Christoph Reich, Felix Wimbauer et al.

Semantic scene completion (SSC) aims to infer both the 3D geometry and semantics of a scene from single images. In contrast to prior work on SSC that heavily relies on expensive ground-truth annotations, we approach SSC in an unsupervised setting. Our novel method, SceneDINO, adapts techniques from self-supervised representation learning and 2D unsupervised scene understanding to SSC. Our training exclusively utilizes multi-view consistency self-supervision without any form of semantic or geometric ground truth. Given a single input image, SceneDINO infers the 3D geometry and expressive 3D DINO features in a feed-forward manner. Through a novel 3D feature distillation approach, we obtain unsupervised 3D semantics. In both 3D and 2D unsupervised scene understanding, SceneDINO reaches state-of-the-art segmentation accuracy. Linear probing our 3D features matches the segmentation accuracy of a current supervised SSC approach. Additionally, we showcase the domain generalization and multi-view consistency of SceneDINO, taking the first steps towards a strong foundation for single image 3D scene understanding.

Robin Hesse, Jonas Fischer, Simone Schaub-Meyer et al.

Mechanistic interpretability is concerned with analyzing individual components in a (convolutional) neural network (CNN) and how they form larger circuits representing decision mechanisms. These investigations are challenging since CNNs frequently learn polysemantic channels that encode distinct concepts, making them hard to interpret. To address this, we propose an algorithm to disentangle a specific kind of polysemantic channel into multiple channels, each responding to a single concept. Our approach restructures weights in a CNN, utilizing that different concepts within the same channel exhibit distinct activation patterns in the previous layer. By disentangling these polysemantic features, we enhance the interpretability of CNNs, ultimately improving explanatory techniques such as feature visualizations.

Johannes Meier, Jonathan Michel, Oussema Dhaouadi et al.

Real-time monocular 3D object detection remains challenging due to severe depth ambiguity, viewpoint shifts, and the high computational cost of 3D reasoning. Existing approaches either rely on LiDAR or geometric priors to compensate for missing depth, or sacrifice efficiency to achieve competitive accuracy. We introduce LeAD-M3D, a monocular 3D detector that achieves state-of-the-art accuracy and real-time inference without extra modalities. Our method is powered by three key components. Asymmetric Augmentation Denoising Distillation (A2D2) transfers geometric knowledge from a clean-image teacher to a mixup-noised student via a quality- and importance-weighted depth-feature loss, enabling stronger depth reasoning without LiDAR supervision. 3D-aware Consistent Matching (CM3D) improves prediction-to-ground truth assignment by integrating 3D MGIoU into the matching score, yielding more stable and precise supervision. Finally, Confidence-Gated 3D Inference (CGI3D) accelerates detection by restricting expensive 3D regression to top-confidence regions. Together, these components set a new Pareto frontier for monocular 3D detection: LeAD-M3D achieves state-of-the-art accuracy on KITTI and Waymo, and the best reported car AP on Rope3D, while running up to 3.6x faster than prior high-accuracy methods. Our results demonstrate that high fidelity and real-time efficiency in monocular 3D detection are simultaneously attainable - without LiDAR, stereo, or geometric assumptions.

Simon Kiefhaber, Stefan Roth, Simone Schaub-Meyer

Cost volumes are used in every modern optical flow estimator, but due to their computational and space complexity, they are often a limiting factor regarding both processing speed and the resolution of input frames. Motivated by our empirical observation that cost volumes lose their importance once all other network parts of, e.g., a RAFT-based pipeline have been sufficiently trained, we introduce a training strategy that allows removing the cost volume from optical flow estimators throughout training. This leads to significantly improved inference speed and reduced memory requirements. Using our training strategy, we create three different models covering different compute budgets. Our most accurate model reaches state-of-the-art accuracy while being $1.2\times$ faster and having a $6\times$ lower memory footprint than comparable models; our fastest model is capable of processing Full HD frames at $20\,\mathrm{FPS}$ using only $500\,\mathrm{MB}$ of GPU memory.

Denis Zavadski, Damjan Kalšan, Tim Küchler et al.

Synthetic datasets are widely used for training urban scene recognition models, but even highly realistic renderings show a noticeable gap to real imagery. This gap is particularly pronounced when adapting to a specific target domain, such as Cityscapes, where differences in architecture, vegetation, object appearance, and camera characteristics limit downstream performance. Closing this gap with more detailed 3D modelling would require expensive asset and scene design, defeating the purpose of low-cost labelled data. To address this, we present a new framework that adapts an off-the-shelf diffusion model to a target domain using only imperfect pseudo-labels. Once trained, it generates high-fidelity, target-aligned images from semantic maps of any synthetic dataset, including low-effort sources created in hours rather than months. The method filters suboptimal generations, rectifies image-label misalignments, and standardises semantics across datasets, transforming weak synthetic data into competitive real-domain training sets. Experiments on five synthetic datasets and two real target datasets show segmentation gains of up to +8.0%pt. mIoU over state-of-the-art translation methods, making rapidly constructed synthetic datasets as effective as high-effort, time-intensive synthetic datasets requiring extensive manual design. This work highlights a valuable collaborative paradigm where fast semantic prototyping, combined with generative models, enables scalable, high-quality training data creation for urban scene understanding.

Xinrui Gong, Oliver Hahn, Christoph Reich et al.

Unsupervised multi-object discovery (MOD) aims to detect and localize distinct object instances in visual scenes without any form of human supervision. Recent approaches leverage object-centric learning (OCL) and motion cues from video to identify individual objects. However, these approaches use supervision to generate pseudo labels to train the OCL model. We address this limitation with MR-DINOSAUR -- Motion-Refined DINOSAUR -- a minimalistic unsupervised approach that extends the self-supervised pre-trained OCL model, DINOSAUR, to the task of unsupervised multi-object discovery. We generate high-quality unsupervised pseudo labels by retrieving video frames without camera motion for which we perform motion segmentation of unsupervised optical flow. We refine DINOSAUR's slot representations using these pseudo labels and train a slot deactivation module to assign slots to foreground and background. Despite its conceptual simplicity, MR-DINOSAUR achieves strong multi-object discovery results on the TRI-PD and KITTI datasets, outperforming the previous state of the art despite being fully unsupervised.

Barış Zöngür, Robin Hesse, Stefan Roth

To ensure the reliability of deep models in real-world applications, out-of-distribution (OOD) detection methods aim to distinguish samples close to the training distribution (in-distribution, ID) from those farther away (OOD). In this work, we propose a novel OOD detection method that utilizes singular value decomposition of the weight matrix of the classification head to decompose the model's activations into decisive and insignificant components, which contribute maximally, respectively minimally, to the final classifier output. We find that the subspace of insignificant components more effectively distinguishes ID from OOD data than raw activations in regimes of large distribution shifts (Far-OOD). This occurs because the classification objective leaves the insignificant subspace largely unaffected, yielding features that are ''untainted'' by the target classification task. Conversely, in regimes of smaller distribution shifts (Near-OOD), we find that activation shaping methods profit from only considering the decisive subspace, as the insignificant component can cause interference in the activation space. By combining two findings into a single approach, termed ActSub, we achieve state-of-the-art results in various standard OOD benchmarks.

Dustin Carrión-Ojeda, Stefan Roth, Simone Schaub-Meyer

Few-shot classification and segmentation (FS-CS) focuses on jointly performing multi-label classification and multi-class segmentation using few annotated examples. Although the current state of the art (SOTA) achieves high accuracy in both tasks, it struggles with small objects. To overcome this, we propose the Efficient Masked Attention Transformer (EMAT), which improves classification and segmentation accuracy, especially for small objects. EMAT introduces three modifications: a novel memory-efficient masked attention mechanism, a learnable downscaling strategy, and parameter-efficiency enhancements. EMAT outperforms all FS-CS methods on the PASCAL-5$^i$ and COCO-20$^i$ datasets, using at least four times fewer trainable parameters. Moreover, as the current FS-CS evaluation setting discards available annotations, despite their costly collection, we introduce two novel evaluation settings that consider these annotations to better reflect practical scenarios.

Gopika Sudhakaran, Hikaru Shindo, Patrick Schramowski et al.

Visual relation detection (VRD) is the task of identifying the relationships between objects in a scene. VRD models trained solely on relation detection data struggle to generalize beyond the relations on which they are trained. While prompt tuning has been used to adapt vision-language models (VLMs) for VRD, it uses handcrafted prompts and struggles with novel or complex relations. We argue that instruction tuning offers a more effective solution by fine-tuning VLMs on diverse instructional data. We thus introduce ART, an Adaptive Relation Tuning framework that adapts VLMs for VRD through instruction tuning and strategic instance selection. By converting VRD datasets into an instruction tuning format and employing an adaptive sampling algorithm, ART directs the VLM to focus on informative relations while maintaining generalizability. Specifically, we focus on the relation classification, where subject-object boxes are given and the model predicts the predicate between them. We tune on a held-in set and evaluate across multiple held-out datasets of varying complexity. Our approach strongly improves over its baselines and can infer unseen relation concepts, a capability absent in mainstream VRD methods. We demonstrate ART's practical value by using the predicted relations for segmenting complex scenes.

Jannik Endres, Oliver Hahn, Charles Corbière et al.

Omnidirectional depth perception is essential for mobile robotics applications that require scene understanding across a full 360° field of view. Camera-based setups offer a cost-effective option by using stereo depth estimation to generate dense, high-resolution depth maps without relying on expensive active sensing. However, existing omnidirectional stereo matching approaches achieve only limited depth accuracy across diverse environments, depth ranges, and lighting conditions, due to the scarcity of real-world data. We present DFI-OmniStereo, a novel omnidirectional stereo matching method that leverages a large-scale pre-trained foundation model for relative monocular depth estimation within an iterative optimization-based stereo matching architecture. We introduce a dedicated two-stage training strategy to utilize the relative monocular depth features for our omnidirectional stereo matching before scale-invariant fine-tuning. DFI-OmniStereo achieves state-of-the-art results on the real-world Helvipad dataset, reducing disparity MAE by approximately 16% compared to the previous best omnidirectional stereo method.

Robin Hesse, Doğukan Bağcı, Bernt Schiele et al.

Deep learning has become an essential part of computer vision, with deep neural networks (DNNs) excelling in predictive performance. However, they often fall short in other critical quality dimensions, such as robustness, calibration, or fairness. While existing studies have focused on a subset of these quality dimensions, none have explored a more general form of "well-behavedness" of DNNs. With this work, we address this gap by simultaneously studying nine different quality dimensions for image classification. Through a large-scale study, we provide a bird's-eye view by analyzing 326 backbone models and how different training paradigms and model architectures affect the quality dimensions. We reveal various new insights such that (i) vision-language models exhibit high fairness on ImageNet-1k classification and strong robustness against domain changes; (ii) self-supervised learning is an effective training paradigm to improve almost all considered quality dimensions; and (iii) the training dataset size is a major driver for most of the quality dimensions. We conclude our study by introducing the QUBA score (Quality Understanding Beyond Accuracy), a novel metric that ranks models across multiple dimensions of quality, enabling tailored recommendations based on specific user needs.

Jan-Martin O. Steitz, Stefan Roth

Adapters provide an efficient and lightweight mechanism for adapting trained transformer models to a variety of different tasks. However, they have often been found to be outperformed by other adaptation mechanisms, including low-rank adaptation. In this paper, we provide an in-depth study of adapters, their internal structure, as well as various implementation choices. We uncover pitfalls for using adapters and suggest a concrete, improved adapter architecture, called Adapter+, that not only outperforms previous adapter implementations but surpasses a number of other, more complex adaptation mechanisms in several challenging settings. Despite this, our suggested adapter is highly robust and, unlike previous work, requires little to no manual intervention when addressing a novel scenario. Adapter+ reaches state-of-the-art average accuracy on the VTAB benchmark, even without a per-task hyperparameter optimization.

Robin Hesse, Simone Schaub-Meyer, Stefan Roth

Many convolutional neural networks (CNNs) rely on progressive downsampling of their feature maps to increase the network's receptive field and decrease computational cost. However, this comes at the price of losing granularity in the feature maps, limiting the ability to correctly understand images or recover fine detail in dense prediction tasks. To address this, common practice is to replace the last few downsampling operations in a CNN with dilated convolutions, allowing to retain the feature map resolution without reducing the receptive field, albeit increasing the computational cost. This allows to trade off predictive performance against cost, depending on the output feature resolution. By either regularly downsampling or not downsampling the entire feature map, existing work implicitly treats all regions of the input image and subsequent feature maps as equally important, which generally does not hold. We propose an adaptive downsampling scheme that generalizes the above idea by allowing to process informative regions at a higher resolution than less informative ones. In a variety of experiments, we demonstrate the versatility of our adaptive downsampling strategy and empirically show that it improves the cost-accuracy trade-off of various established CNNs.

Paul Staat, Simon Mulzer, Stefan Roth et al.

Wireless radio channels are known to contain information about the surrounding propagation environment, which can be extracted using established wireless sensing methods. Thus, today's ubiquitous wireless devices are attractive targets for passive eavesdroppers to launch reconnaissance attacks. In particular, by overhearing standard communication signals, eavesdroppers obtain estimations of wireless channels which can give away sensitive information about indoor environments. For instance, by applying simple statistical methods, adversaries can infer human motion from wireless channel observations, allowing to remotely monitor premises of victims. In this work, building on the advent of intelligent reflecting surfaces (IRSs), we propose IRShield as a novel countermeasure against adversarial wireless sensing. IRShield is designed as a plug-and-play privacy-preserving extension to existing wireless networks. At the core of IRShield, we design an IRS configuration algorithm to obfuscate wireless channels. We validate the effectiveness with extensive experimental evaluations. In a state-of-the-art human motion detection attack using off-the-shelf Wi-Fi devices, IRShield lowered detection rates to 5% or less.

Robin Hesse, Simone Schaub-Meyer, Stefan Roth

Mitigating the dependence on spurious correlations present in the training dataset is a quickly emerging and important topic of deep learning. Recent approaches include priors on the feature attribution of a deep neural network (DNN) into the training process to reduce the dependence on unwanted features. However, until now one needed to trade off high-quality attributions, satisfying desirable axioms, against the time required to compute them. This in turn either led to long training times or ineffective attribution priors. In this work, we break this trade-off by considering a special class of efficiently axiomatically attributable DNNs for which an axiomatic feature attribution can be computed with only a single forward/backward pass. We formally prove that nonnegatively homogeneous DNNs, here termed $\mathcal{X}$-DNNs, are efficiently axiomatically attributable and show that they can be effortlessly constructed from a wide range of regular DNNs by simply removing the bias term of each layer. Various experiments demonstrate the advantages of $\mathcal{X}$-DNNs, beating state-of-the-art generic attribution methods on regular DNNs for training with attribution priors.

Nikita Araslanov, Simone Schaub-Meyer, Stefan Roth

We present a novel approach to unsupervised learning for video object segmentation (VOS). Unlike previous work, our formulation allows to learn dense feature representations directly in a fully convolutional regime. We rely on uniform grid sampling to extract a set of anchors and train our model to disambiguate between them on both inter- and intra-video levels. However, a naive scheme to train such a model results in a degenerate solution. We propose to prevent this with a simple regularisation scheme, accommodating the equivariance property of the segmentation task to similarity transformations. Our training objective admits efficient implementation and exhibits fast training convergence. On established VOS benchmarks, our approach exceeds the segmentation accuracy of previous work despite using significantly less training data and compute power.

Shweta Mahajan, Stefan Roth

Autoregressive models are a class of exact inference approaches with highly flexible functional forms, yielding state-of-the-art density estimates for natural images. Yet, the sequential ordering on the dimensions makes these models computationally expensive and limits their applicability to low-resolution imagery. In this work, we propose Pixel-Pyramids, a block-autoregressive approach employing a lossless pyramid decomposition with scale-specific representations to encode the joint distribution of image pixels. Crucially, it affords a sparser dependency structure compared to fully autoregressive approaches. Our PixelPyramids yield state-of-the-art results for density estimation on various image datasets, especially for high-resolution data. For CelebA-HQ 1024 x 1024, we observe that the density estimates (in terms of bits/dim) are improved to ~44% of the baseline despite sampling speeds superior even to easily parallelizable flow-based models.

Jonas Pfeiffer, Gregor Geigle, Aishwarya Kamath et al.

Recent advances in multimodal vision and language modeling have predominantly focused on the English language, mostly due to the lack of multilingual multimodal datasets to steer modeling efforts. In this work, we address this gap and provide xGQA, a new multilingual evaluation benchmark for the visual question answering task. We extend the established English GQA dataset to 7 typologically diverse languages, enabling us to detect and explore crucial challenges in cross-lingual visual question answering. We further propose new adapter-based approaches to adapt multimodal transformer-based models to become multilingual, and -- vice versa -- multilingual models to become multimodal. Our proposed methods outperform current state-of-the-art multilingual multimodal models (e.g., M3P) in zero-shot cross-lingual settings, but the accuracy remains low across the board; a performance drop of around 38 accuracy points in target languages showcases the difficulty of zero-shot cross-lingual transfer for this task. Our results suggest that simple cross-lingual transfer of multimodal models yields latent multilingual multimodal misalignment, calling for more sophisticated methods for vision and multilingual language modeling.

Jan-Martin O. Steitz, Jonas Pfeiffer, Iryna Gurevych et al.

Reasoning over multiple modalities, e.g. in Visual Question Answering (VQA), requires an alignment of semantic concepts across domains. Despite the widespread success of end-to-end learning, today's multimodal pipelines by and large leverage pre-extracted, fixed features from object detectors, typically Faster R-CNN, as representations of the visual world. The obvious downside is that the visual representation is not specifically tuned to the multimodal task at hand. At the same time, while transformer-based object detectors have gained popularity, they have not been employed in today's multimodal pipelines. We address both shortcomings with TxT, a transformer-based crossmodal pipeline that enables fine-tuning both language and visual components on the downstream task in a fully end-to-end manner. We overcome existing limitations of transformer-based detectors for multimodal reasoning regarding the integration of global context and their scalability. Our transformer-based multimodal model achieves considerable gains from end-to-end learning for multimodal question answering.

Junhwa Hur, Stefan Roth

Estimating 3D scene flow from a sequence of monocular images has been gaining increased attention due to the simple, economical capture setup. Owing to the severe ill-posedness of the problem, the accuracy of current methods has been limited, especially that of efficient, real-time approaches. In this paper, we introduce a multi-frame monocular scene flow network based on self-supervised learning, improving the accuracy over previous networks while retaining real-time efficiency. Based on an advanced two-frame baseline with a split-decoder design, we propose (i) a multi-frame model using a triple frame input and convolutional LSTM connections, (ii) an occlusion-aware census loss for better accuracy, and (iii) a gradient detaching strategy to improve training stability. On the KITTI dataset, we observe state-of-the-art accuracy among monocular scene flow methods based on self-supervised learning.

Nikita Araslanov, Stefan Roth

We propose an approach to domain adaptation for semantic segmentation that is both practical and highly accurate. In contrast to previous work, we abandon the use of computationally involved adversarial objectives, network ensembles and style transfer. Instead, we employ standard data augmentation techniques $-$ photometric noise, flipping and scaling $-$ and ensure consistency of the semantic predictions across these image transformations. We develop this principle in a lightweight self-supervised framework trained on co-evolving pseudo labels without the need for cumbersome extra training rounds. Simple in training from a practitioner's standpoint, our approach is remarkably effective. We achieve significant improvements of the state-of-the-art segmentation accuracy after adaptation, consistent both across different choices of the backbone architecture and adaptation scenarios.

Jiangxin Dong, Stefan Roth, Bernt Schiele

We present a simple and effective approach for non-blind image deblurring, combining classical techniques and deep learning. In contrast to existing methods that deblur the image directly in the standard image space, we propose to perform an explicit deconvolution process in a feature space by integrating a classical Wiener deconvolution framework with learned deep features. A multi-scale cascaded feature refinement module then predicts the deblurred image from the deconvolved deep features, progressively recovering detail and small-scale structures. The proposed model is trained in an end-to-end manner and evaluated on scenarios with simulated Gaussian noise, saturated pixels, or JPEG compression artifacts as well as real-world images. Moreover, we present detailed analyses of the benefit of the feature-based Wiener deconvolution and of the multi-scale cascaded feature refinement as well as the robustness of the proposed approach. Our extensive experimental results show that the proposed deep Wiener deconvolution network facilitates deblurred results with visibly fewer artifacts and quantitatively outperforms state-of-the-art non-blind image deblurring methods by a wide margin.

Jannik Schmitt, Stefan Roth

To adopt neural networks in safety critical domains, knowing whether we can trust their predictions is crucial. Bayesian neural networks (BNNs) provide uncertainty estimates by averaging predictions with respect to the posterior weight distribution. Variational inference methods for BNNs approximate the intractable weight posterior with a tractable distribution, yet mostly rely on sampling from the variational distribution during training and inference. Recent sampling-free approaches offer an alternative, but incur a significant parameter overhead. We here propose a more efficient parameterization of the posterior approximation for sampling-free variational inference that relies on the distribution induced by multiplicative Gaussian activation noise. This allows us to combine parameter efficiency with the benefits of sampling-free variational inference. Our approach yields competitive results for standard regression problems and scales well to large-scale image classification tasks including ImageNet.

Stefan Roth, Stefano Tomasin, Marco Maso et al.

In fifth-generation (5G) cellular networks, users feed back to the base station the index of the precoder (from a codebook) to be used for downlink transmission. The precoder is strongly related to the user channel and in turn to the user position within the cell. We propose a method by which an external attacker determines the user position by passively overhearing this unencrypted layer-2 feedback signal. The attacker first builds a map of fed back precoder indices in the cell. Then, by overhearing the precoder index fed back by the victim user, the attacker finds its position on the map. We focus on the type-I single-panel codebook, which today is the only mandatory solution in the 3GPP standard. We analyze the attack and assess the obtained localization accuracy against various parameters. We analyze the localization error of a simplified precoder feedback model and describe its asymptotic localization precision. We also propose a mitigation against our attack, wherein the user randomly selects the precoder among those providing the highest rate. Simulations confirm that the attack can achieve a high localization accuracy, which is significantly reduced when the mitigation solution is adopted, at the cost of a negligible rate degradation.

Shweta Mahajan, Stefan Roth

Diverse image captioning models aim to learn one-to-many mappings that are innate to cross-domain datasets, such as of images and texts. Current methods for this task are based on generative latent variable models, e.g. VAEs with structured latent spaces. Yet, the amount of multimodality captured by prior work is limited to that of the paired training data -- the true diversity of the underlying generative process is not fully captured. To address this limitation, we leverage the contextual descriptions in the dataset that explain similar contexts in different visual scenes. To this end, we introduce a novel factorization of the latent space, termed context-object split, to model diversity in contextual descriptions across images and texts within the dataset. Our framework not only enables diverse captioning through context-based pseudo supervision, but extends this to images with novel objects and without paired captions in the training data. We evaluate our COS-CVAE approach on the standard COCO dataset and on the held-out COCO dataset consisting of images with novel objects, showing significant gains in accuracy and diversity.

Patrick Dendorfer, Aljoša Ošep, Anton Milan et al.

Standardized benchmarks have been crucial in pushing the performance of computer vision algorithms, especially since the advent of deep learning. Although leaderboards should not be over-claimed, they often provide the most objective measure of performance and are therefore important guides for research. We present MOTChallenge, a benchmark for single-camera Multiple Object Tracking (MOT) launched in late 2014, to collect existing and new data, and create a framework for the standardized evaluation of multiple object tracking methods. The benchmark is focused on multiple people tracking, since pedestrians are by far the most studied object in the tracking community, with applications ranging from robot navigation to self-driving cars. This paper collects the first three releases of the benchmark: (i) MOT15, along with numerous state-of-the-art results that were submitted in the last years, (ii) MOT16, which contains new challenging videos, and (iii) MOT17, that extends MOT16 sequences with more precise labels and evaluates tracking performance on three different object detectors. The second and third release not only offers a significant increase in the number of labeled boxes but also provide labels for multiple object classes beside pedestrians, as well as the level of visibility for every single object of interest. We finally provide a categorization of state-of-the-art trackers and a broad error analysis. This will help newcomers understand the related work and research trends in the MOT community, and hopefully shed some light on potential future research directions.

Sascha Rosbach, Xing Li, Simon Großjohann et al.

General-purpose trajectory planning algorithms for automated driving utilize complex reward functions to perform a combined optimization of strategic, behavioral, and kinematic features. The specification and tuning of a single reward function is a tedious task and does not generalize over a large set of traffic situations. Deep learning approaches based on path integral inverse reinforcement learning have been successfully applied to predict local situation-dependent reward functions using features of a set of sampled driving policies. Sample-based trajectory planning algorithms are able to approximate a spatio-temporal subspace of feasible driving policies that can be used to encode the context of a situation. However, the interaction with dynamic objects requires an extended planning horizon, which depends on sequential context modeling. In this work, we are concerned with the sequential reward prediction over an extended time horizon. We present a neural network architecture that uses a policy attention mechanism to generate a low-dimensional context vector by concentrating on trajectories with a human-like driving style. Apart from this, we propose a temporal attention mechanism to identify context switches and allow for stable adaptation of rewards. We evaluate our results on complex simulated driving situations, including other moving vehicles. Our evaluation shows that our policy attention mechanism learns to focus on collision-free policies in the configuration space. Furthermore, the temporal attention mechanism learns persistent interaction with other vehicles over an extended planning horizon.

Wentong Liao, Xiang Chen, Jingfeng Yang et al.

State-of-the-art object detection approaches such as Fast/Faster R-CNN, SSD, or YOLO have difficulties detecting dense, small targets with arbitrary orientation in large aerial images. The main reason is that using interpolation to align RoI features can result in a lack of accuracy or even loss of location information. We present the Local-aware Region Convolutional Neural Network (LR-CNN), a novel two-stage approach for vehicle detection in aerial imagery. We enhance translation invariance to detect dense vehicles and address the boundary quantization issue amongst dense vehicles by aggregating the high-precision RoIs' features. Moreover, we resample high-level semantic pooled features, making them regain location information from the features of a shallower convolutional block. This strengthens the local feature invariance for the resampled features and enables detecting vehicles in an arbitrary orientation. The local feature invariance enhances the learning ability of the focal loss function, and the focal loss further helps to focus on the hard examples. Taken together, our method better addresses the challenges of aerial imagery. We evaluate our approach on several challenging datasets (VEDAI, DOTA), demonstrating a significant improvement over state-of-the-art methods. We demonstrate the good generalization ability of our approach on the DLR 3K dataset.

Nikita Araslanov, Stefan Roth

Recent years have seen a rapid growth in new approaches improving the accuracy of semantic segmentation in a weakly supervised setting, i.e. with only image-level labels available for training. However, this has come at the cost of increased model complexity and sophisticated multi-stage training procedures. This is in contrast to earlier work that used only a single stage $-$ training one segmentation network on image labels $-$ which was abandoned due to inferior segmentation accuracy. In this work, we first define three desirable properties of a weakly supervised method: local consistency, semantic fidelity, and completeness. Using these properties as guidelines, we then develop a segmentation-based network model and a self-supervised training scheme to train for semantic masks from image-level annotations in a single stage. We show that despite its simplicity, our method achieves results that are competitive with significantly more complex pipelines, substantially outperforming earlier single-stage methods.

Junhwa Hur, Stefan Roth

Scene flow estimation has been receiving increasing attention for 3D environment perception. Monocular scene flow estimation -- obtaining 3D structure and 3D motion from two temporally consecutive images -- is a highly ill-posed problem, and practical solutions are lacking to date. We propose a novel monocular scene flow method that yields competitive accuracy and real-time performance. By taking an inverse problem view, we design a single convolutional neural network (CNN) that successfully estimates depth and 3D motion simultaneously from a classical optical flow cost volume. We adopt self-supervised learning with 3D loss functions and occlusion reasoning to leverage unlabeled data. We validate our design choices, including the proxy loss and augmentation setup. Our model achieves state-of-the-art accuracy among unsupervised/self-supervised learning approaches to monocular scene flow, and yields competitive results for the optical flow and monocular depth estimation sub-tasks. Semi-supervised fine-tuning further improves the accuracy and yields promising results in real-time.

Shweta Mahajan, Apratim Bhattacharyya, Mario Fritz et al.

Flow-based generative models are an important class of exact inference models that admit efficient inference and sampling for image synthesis. Owing to the efficiency constraints on the design of the flow layers, e.g. split coupling flow layers in which approximately half the pixels do not undergo further transformations, they have limited expressiveness for modeling long-range data dependencies compared to autoregressive models that rely on conditional pixel-wise generation. In this work, we improve the representational power of flow-based models by introducing channel-wise dependencies in their latent space through multi-scale autoregressive priors (mAR). Our mAR prior for models with split coupling flow layers (mAR-SCF) can better capture dependencies in complex multimodal data. The resulting model achieves state-of-the-art density estimation results on MNIST, CIFAR-10, and ImageNet. Furthermore, we show that mAR-SCF allows for improved image generation quality, with gains in FID and Inception scores compared to state-of-the-art flow-based models.

Junhwa Hur, Stefan Roth

Akin to many subareas of computer vision, the recent advances in deep learning have also significantly influenced the literature on optical flow. Previously, the literature had been dominated by classical energy-based models, which formulate optical flow estimation as an energy minimization problem. However, as the practical benefits of Convolutional Neural Networks (CNNs) over conventional methods have become apparent in numerous areas of computer vision and beyond, they have also seen increased adoption in the context of motion estimation to the point where the current state of the art in terms of accuracy is set by CNN approaches. We first review this transition as well as the developments from early work to the current state of CNNs for optical flow estimation. Alongside, we discuss some of their technical details and compare them to recapitulate which technical contribution led to the most significant accuracy improvements. Then we provide an overview of the various optical flow approaches introduced in the deep learning age, including those based on alternative learning paradigms (e.g., unsupervised and semi-supervised methods) as well as the extension to the multi-frame case, which is able to yield further accuracy improvements.

Anne S. Wannenwetsch, Stefan Roth

Encoder-decoder networks have found widespread use in various dense prediction tasks. However, the strong reduction of spatial resolution in the encoder leads to a loss of location information as well as boundary artifacts. To address this, image-adaptive post-processing methods have shown beneficial by leveraging the high-resolution input image(s) as guidance data. We extend such approaches by considering an important orthogonal source of information: the network's confidence in its own predictions. We introduce probabilistic pixel-adaptive convolutions (PPACs), which not only depend on image guidance data for filtering, but also respect the reliability of per-pixel predictions. As such, PPACs allow for image-adaptive smoothing and simultaneously propagating pixels of high confidence into less reliable regions, while respecting object boundaries. We demonstrate their utility in refinement networks for optical flow and semantic segmentation, where PPACs lead to a clear reduction in boundary artifacts. Moreover, our proposed refinement step is able to substantially improve the accuracy on various widely used benchmarks.