Vasileios Belagiannis

Yuyuan Liu, Choubo Ding, Yu Tian et al.

Semantic segmentation models classify pixels into a set of known (``in-distribution'') visual classes. When deployed in an open world, the reliability of these models depends on their ability not only to classify in-distribution pixels but also to detect out-of-distribution (OoD) pixels. Historically, the poor OoD detection performance of these models has motivated the design of methods based on model re-training using synthetic training images that include OoD visual objects. Although successful, these re-trained methods have two issues: 1) their in-distribution segmentation accuracy may drop during re-training, and 2) their OoD detection accuracy does not generalise well to new contexts (e.g., country surroundings) outside the training set (e.g., city surroundings). In this paper, we mitigate these issues with: (i) a new residual pattern learning (RPL) module that assists the segmentation model to detect OoD pixels without affecting the inlier segmentation performance; and (ii) a novel context-robust contrastive learning (CoroCL) that enforces RPL to robustly detect OoD pixels among various contexts. Our approach improves by around 10\% FPR and 7\% AuPRC the previous state-of-the-art in Fishyscapes, Segment-Me-If-You-Can, and RoadAnomaly datasets. Our code is available at: https://github.com/yyliu01/RPL.

Arij Bouazizi, Adrian Holzbock, Ulrich Kressel et al.

In this work, we present MotionMixer, an efficient 3D human body pose forecasting model based solely on multi-layer perceptrons (MLPs). MotionMixer learns the spatial-temporal 3D body pose dependencies by sequentially mixing both modalities. Given a stacked sequence of 3D body poses, a spatial-MLP extracts fine grained spatial dependencies of the body joints. The interaction of the body joints over time is then modelled by a temporal MLP. The spatial-temporal mixed features are finally aggregated and decoded to obtain the future motion. To calibrate the influence of each time step in the pose sequence, we make use of squeeze-and-excitation (SE) blocks. We evaluate our approach on Human3.6M, AMASS, and 3DPW datasets using the standard evaluation protocols. For all evaluations, we demonstrate state-of-the-art performance, while having a model with a smaller number of parameters. Our code is available at: https://github.com/MotionMLP/MotionMixer

Yuyuan Liu, Yu Tian, Chong Wang et al.

3D medical image segmentation methods have been successful, but their dependence on large amounts of voxel-level annotated data is a disadvantage that needs to be addressed given the high cost to obtain such annotation. Semi-supervised learning (SSL) solve this issue by training models with a large unlabelled and a small labelled dataset. The most successful SSL approaches are based on consistency learning that minimises the distance between model responses obtained from perturbed views of the unlabelled data. These perturbations usually keep the spatial input context between views fairly consistent, which may cause the model to learn segmentation patterns from the spatial input contexts instead of the segmented objects. In this paper, we introduce the Translation Consistent Co-training (TraCoCo) which is a consistency learning SSL method that perturbs the input data views by varying their spatial input context, allowing the model to learn segmentation patterns from visual objects. Furthermore, we propose the replacement of the commonly used mean squared error (MSE) semi-supervised loss by a new Cross-model confident Binary Cross entropy (CBC) loss, which improves training convergence and keeps the robustness to co-training pseudo-labelling mistakes. We also extend CutMix augmentation to 3D SSL to further improve generalisation. Our TraCoCo shows state-of-the-art results for the Left Atrium (LA) and Brain Tumor Segmentation (BRaTS19) datasets with different backbones. Our code is available at https://github.com/yyliu01/TraCoCo.

Julia Hornauer, Vasileios Belagiannis

In monocular depth estimation, disturbances in the image context, like moving objects or reflecting materials, can easily lead to erroneous predictions. For that reason, uncertainty estimates for each pixel are necessary, in particular for safety-critical applications such as automated driving. We propose a post hoc uncertainty estimation approach for an already trained and thus fixed depth estimation model, represented by a deep neural network. The uncertainty is estimated with the gradients which are extracted with an auxiliary loss function. To avoid relying on ground-truth information for the loss definition, we present an auxiliary loss function based on the correspondence of the depth prediction for an image and its horizontally flipped counterpart. Our approach achieves state-of-the-art uncertainty estimation results on the KITTI and NYU Depth V2 benchmarks without the need to retrain the neural network. Models and code are publicly available at https://github.com/jhornauer/GrUMoDepth.

Azhar Hussian, Marina Ritthaler, André Kaup et al.

We present an image-conditioned point cloud completion approach that treats images as the primary geometric source rather than a secondary guide. To this end, we introduce an Image-to-Point (I2P) module that can reconstruct complete point clouds directly from a single RGB image, with no need for 3D inputs. Additionally, we introduce a transformer-based Point-to-Point (P2P) refinement module that uses self- and cross-attention between point tokens and image features to iteratively refine the coarse I2P output. The I2P module enables the image encoder to learn rich geometric representations, while the P2P module progressively recovers fine-grained details. Unlike existing multimodal methods that rely on auxiliary losses or fusion modules, our explicit I2P task provides a strong, geometry-aware prior based on images alone. Extensive experiments on ShapeNet-ViPC demonstrate state-of-the-art completion performance with a 12.3% relative Chamfer Distance improvement over prior methods. Code is available at: https://github.com/AzharSindhi/I2PRef.git

Amir El-Ghoussani, André Kaup, Nassir Navab et al.

We propose a monocular depth estimation method based on visual autoregressive (VAR) priors, offering an alternative to diffusion-based approaches. Our method adapts a large-scale text-to-image VAR model and introduces a scale-wise conditional upsampling mechanism with classifier-free guidance. Our approach performs inference in ten fixed autoregressive stages, requiring only 74K synthetic samples for fine-tuning, and achieves competitive results. We report state-of-the-art performance in indoor benchmarks under constrained training conditions, and strong performance when applied to outdoor datasets. This work establishes autoregressive priors as a complementary family of geometry-aware generative models for depth estimation, highlighting advantages in data scalability, and adaptability to 3D vision tasks. Code available at "https://github.com/AmirMaEl/VAR-Depth".

Adrian Holzbock, Achyut Hegde, Klaus Dietmayer et al.

Model compression techniques reduce the computational load and memory consumption of deep neural networks. After the compression operation, e.g. parameter pruning, the model is normally fine-tuned on the original training dataset to recover from the performance drop caused by compression. However, the training data is not always available due to privacy issues or other factors. In this work, we present a data-free fine-tuning approach for pruning the backbone of deep neural networks. In particular, the pruned network backbone is trained with synthetically generated images, and our proposed intermediate supervision to mimic the unpruned backbone's output feature map. Afterwards, the pruned backbone can be combined with the original network head to make predictions. We generate synthetic images by back-propagating gradients to noise images while relying on L1-pruning for the backbone pruning. In our experiments, we show that our approach is task-independent due to pruning only the backbone. By evaluating our approach on 2D human pose estimation, object detection, and image classification, we demonstrate promising performance compared to the unpruned model. Our code is available at https://github.com/holzbock/dfbf.

Yuyuan Liu, Yuanhong Chen, Hu Wang et al.

The costly and time-consuming annotation process to produce large training sets for modelling semantic LiDAR segmentation methods has motivated the development of semi-supervised learning (SSL) methods. However, such SSL approaches often concentrate on employing consistency learning only for individual LiDAR representations. This narrow focus results in limited perturbations that generally fail to enable effective consistency learning. Additionally, these SSL approaches employ contrastive learning based on the sampling from a limited set of positive and negative embedding samples. This paper introduces a novel semi-supervised LiDAR semantic segmentation framework called ItTakesTwo (IT2). IT2 is designed to ensure consistent predictions from peer LiDAR representations, thereby improving the perturbation effectiveness in consistency learning. Furthermore, our contrastive learning employs informative samples drawn from a distribution of positive and negative embeddings learned from the entire training set. Results on public benchmarks show that our approach achieves remarkable improvements over the previous state-of-the-art (SOTA) methods in the field. The code is available at: https://github.com/yyliu01/IT2.

Adrian Holzbock, Alexander Tsaregorodtsev, Youssef Dawoud et al.

Gesture recognition is essential for the interaction of autonomous vehicles with humans. While the current approaches focus on combining several modalities like image features, keypoints and bone vectors, we present neural network architecture that delivers state-of-the-art results only with body skeleton input data. We propose the spatio-temporal multilayer perceptron for gesture recognition in the context of autonomous vehicles. Given 3D body poses over time, we define temporal and spatial mixing operations to extract features in both domains. Additionally, the importance of each time step is re-weighted with Squeeze-and-Excitation layers. An extensive evaluation of the TCG and Drive&Act datasets is provided to showcase the promising performance of our approach. Furthermore, we deploy our model to our autonomous vehicle to show its real-time capability and stable execution.

Julia Hornauer, Vasileios Belagiannis

Our work investigates out-of-distribution (OOD) detection as a neural network output explanation problem. We learn a heatmap representation for detecting OOD images while visualizing in- and out-of-distribution image regions at the same time. Given a trained and fixed classifier, we train a decoder neural network to produce heatmaps with zero response for in-distribution samples and high response heatmaps for OOD samples, based on the classifier features and the class prediction. Our main innovation lies in the heatmap definition for an OOD sample, as the normalized difference from the closest in-distribution sample. The heatmap serves as a margin to distinguish between in- and out-of-distribution samples. Our approach generates the heatmaps not only for OOD detection, but also to indicate in- and out-of-distribution regions of the input image. In our evaluations, our approach mostly outperforms the prior work on fixed classifiers, trained on CIFAR-10, CIFAR-100 and Tiny ImageNet. The code is publicly available at: https://github.com/jhornauer/heatmap_ood.

Youssef Dawoud, Arij Bouazizi, Katharina Ernst et al.

In microscopy image cell segmentation, it is common to train a deep neural network on source data, containing different types of microscopy images, and then fine-tune it using a support set comprising a few randomly selected and annotated training target images. In this paper, we argue that the random selection of unlabelled training target images to be annotated and included in the support set may not enable an effective fine-tuning process, so we propose a new approach to optimise this image selection process. Our approach involves a new scoring function to find informative unlabelled target images. In particular, we propose to measure the consistency in the model predictions on target images against specific data augmentations. However, we observe that the model trained with source datasets does not reliably evaluate consistency on target images. To alleviate this problem, we propose novel self-supervised pretext tasks to compute the scores of unlabelled target images. Finally, the top few images with the least consistency scores are added to the support set for oracle (i.e., expert) annotation and later used to fine-tune the model to the target images. In our evaluations that involve the segmentation of five different types of cell images, we demonstrate promising results on several target test sets compared to the random selection approach as well as other selection approaches, such as Shannon's entropy and Monte-Carlo dropout.

Michael Rudolph, Youssef Dawoud, Ronja Güldenring et al.

We present a lightweight encoder-decoder architecture for monocular depth estimation, specifically designed for embedded platforms. Our main contribution is the Guided Upsampling Block (GUB) for building the decoder of our model. Motivated by the concept of guided image filtering, GUB relies on the image to guide the decoder on upsampling the feature representation and the depth map reconstruction, achieving high resolution results with fine-grained details. Based on multiple GUBs, our model outperforms the related methods on the NYU Depth V2 dataset in terms of accuracy while delivering up to 35.1 fps on the NVIDIA Jetson Nano and up to 144.5 fps on the NVIDIA Xavier NX. Similarly, on the KITTI dataset, inference is possible with up to 23.7 fps on the Jetson Nano and 102.9 fps on the Xavier NX. Our code and models are made publicly available.

Julian Schmidt, Pascal Huissel, Julian Wiederer et al.

It is desirable to predict the behavior of traffic participants conditioned on different planned trajectories of the autonomous vehicle. This allows the downstream planner to estimate the impact of its decisions. Recent approaches for conditional behavior prediction rely on a regression decoder, meaning that coordinates or polynomial coefficients are regressed. In this work we revisit set-based trajectory prediction, where the probability of each trajectory in a predefined trajectory set is determined by a classification model, and first-time employ it to the task of conditional behavior prediction. We propose RESET, which combines a new metric-driven algorithm for trajectory set generation with a graph-based encoder. For unconditional prediction, RESET achieves comparable performance to a regression-based approach. Due to the nature of set-based approaches, it has the advantageous property of being able to predict a flexible number of trajectories without influencing runtime or complexity. For conditional prediction, RESET achieves reasonable results with late fusion of the planned trajectory, which was not observed for regression-based approaches before. This means that RESET is computationally lightweight to combine with a planner that proposes multiple future plans of the autonomous vehicle, as large parts of the forward pass can be reused.

Youssef Dawoud, Gustavo Carneiro, Vasileios Belagiannis

Generalisation of deep neural networks becomes vulnerable when distribution shifts are encountered between train (source) and test (target) domain data. Few-shot domain adaptation mitigates this issue by adapting deep neural networks pre-trained on the source domain to the target domain using a randomly selected and annotated support set from the target domain. This paper argues that randomly selecting the support set can be further improved for effectively adapting the pre-trained source models to the target domain. Alternatively, we propose SelectNAdapt, an algorithm to curate the selection of the target domain samples, which are then annotated and included in the support set. In particular, for the K-shot adaptation problem, we first leverage self-supervision to learn features of the target domain data. Then, we propose a per-class clustering scheme of the learned target domain features and select K representative target samples using a distance-based scoring function. Finally, we bring our selection setup towards a practical ground by relying on pseudo-labels for clustering semantically similar target domain samples. Our experiments show promising results on three few-shot domain adaptation benchmarks for image recognition compared to related approaches and the standard random selection.

Amir El-Ghoussani, Marc Hölle, Gustavo Carneiro et al.

We address the problem of prompt-guided image editing in visual autoregressive models. Given a source image and a target text prompt, we aim to modify the source image according to the target prompt, while preserving all regions which are unrelated to the requested edit. To this end, we present Masked Logit Nudging, which uses the source image token maps to introduce a guidance step that aligns the model's predictions under the target prompt with these source token maps. Specifically, we convert the fixed source encodings into logits using the VAR encoding, nudging the model's predicted logits towards the targets along a semantic trajectory defined by the source-target prompts. Edits are applied only within spatial masks obtained through a dedicated masking scheme that leverages cross-attention differences between the source and edited prompts. Then, we introduce a refinement to correct quantization errors and improve reconstruction quality. Our approach achieves the best image editing performance on the PIE benchmark at 512px and 1024px resolutions. Beyond editing, our method delivers faithful reconstructions and outperforms previous methods on COCO at 512px and OpenImages at 1024px. Overall, our method outperforms VAR-related approaches and achieves comparable or even better performance than diffusion models, while being much faster. Code is available at 'https://github.com/AmirMaEl/MLN'.

Youssef Dawoud, Katharina Ernst, Gustavo Carneiro et al.

Deep neural networks currently deliver promising results for microscopy image cell segmentation, but they require large-scale labelled databases, which is a costly and time-consuming process. In this work, we relax the labelling requirement by combining self-supervised with semi-supervised learning. We propose the prediction of edge-based maps for self-supervising the training of the unlabelled images, which is combined with the supervised training of a small number of labelled images for learning the segmentation task. In our experiments, we evaluate on a few-shot microscopy image cell segmentation benchmark and show that only a small number of annotated images, e.g. 10% of the original training set, is enough for our approach to reach similar performance as with the fully annotated databases on 1- to 10-shots. Our code and trained models is made publicly available

Alexander Tsaregorodtsev, Johannes Müller, Jan Strohbeck et al.

Monocular camera sensors are vital to intelligent vehicle operation and automated driving assistance and are also heavily employed in traffic control infrastructure. Calibrating the monocular camera, though, is time-consuming and often requires significant manual intervention. In this work, we present an extrinsic camera calibration approach that automatizes the parameter estimation by utilizing semantic segmentation information from images and point clouds. Our approach relies on a coarse initial measurement of the camera pose and builds on lidar sensors mounted on a vehicle with high-precision localization to capture a point cloud of the camera environment. Afterward, a mapping between the camera and world coordinate spaces is obtained by performing a lidar-to-camera registration of the semantically segmented sensor data. We evaluate our method on simulated and real-world data to demonstrate low error measurements in the calibration results. Our approach is suitable for infrastructure sensors as well as vehicle sensors, while it does not require motion of the camera platform.

Yuan Lin, Murong Xu, Marc Hölle et al.

Widely adopted medical image segmentation methods, although efficient, are primarily deterministic and remain poorly amenable to natural language prompts. Thus, they lack the capability to estimate multiple proposals, human interaction, and cross-modality adaptation. Recently, text-to-image diffusion models have shown potential to bridge the gap. However, training them from scratch requires a large dataset-a limitation for medical image segmentation. Furthermore, they are often limited to binary segmentation and cannot be conditioned on a natural language prompt. To this end, we propose a novel framework called ProGiDiff that leverages existing image generation models for medical image segmentation purposes. Specifically, we propose a ControlNet-style conditioning mechanism with a custom encoder, suitable for image conditioning, to steer a pre-trained diffusion model to output segmentation masks. It naturally extends to a multi-class setting simply by prompting the target organ. Our experiment on organ segmentation from CT images demonstrates strong performance compared to previous methods and could greatly benefit from an expert-in-the-loop setting to leverage multiple proposals. Importantly, we demonstrate that the learned conditioning mechanism can be easily transferred through low-rank, few-shot adaptation to segment MR images.

Julia Hornauer, Adrian Holzbock, Vasileios Belagiannis

In monocular depth estimation, uncertainty estimation approaches mainly target the data uncertainty introduced by image noise. In contrast to prior work, we address the uncertainty due to lack of knowledge, which is relevant for the detection of data not represented by the training distribution, the so-called out-of-distribution (OOD) data. Motivated by anomaly detection, we propose to detect OOD images from an encoder-decoder depth estimation model based on the reconstruction error. Given the features extracted with the fixed depth encoder, we train an image decoder for image reconstruction using only in-distribution data. Consequently, OOD images result in a high reconstruction error, which we use to distinguish between in- and out-of-distribution samples. We built our experiments on the standard NYU Depth V2 and KITTI benchmarks as in-distribution data. Our post hoc method performs astonishingly well on different models and outperforms existing uncertainty estimation approaches without modifying the trained encoder-decoder depth estimation model.

Michele De Vita, Julian Wiederer, Vasileios Belagiannis

Trajectory prediction models often fail in real-world automated driving due to distributional shifts between training and test conditions. Such distributional shifts, whether behavioural or environmental, pose a critical risk by causing the model to make incorrect forecasts in unfamiliar situations. We propose a self-supervised method that trains a decoder in a post-hoc fashion on the self-supervised task of forecasting the second half of observed trajectories from the first half. The L2 norm of the gradient of this forecasting loss with respect to the decoder's final layer defines a score to identify distribution shifts. Our approach, first, does not affect the trajectory prediction model, ensuring no interference with original prediction performance and second, demonstrates substantial improvements on distribution shift detection for trajectory prediction on the Shifts and Argoverse datasets. Moreover, we show that this method can also be used to early detect collisions of a deep Q-Network motion planner in the Highway simulator. Source code is available at https://github.com/Michedev/forecasting-the-past.

Adrian Holzbock, Nicolai Kern, Christian Waldschmidt et al.

We present a joint camera and radar approach to enable autonomous vehicles to understand and react to human gestures in everyday traffic. Initially, we process the radar data with a PointNet followed by a spatio-temporal multilayer perceptron (stMLP). Independently, the human body pose is extracted from the camera frame and processed with a separate stMLP network. We propose a fusion neural network for both modalities, including an auxiliary loss for each modality. In our experiments with a collected dataset, we show the advantages of gesture recognition with two modalities. Motivated by adverse weather conditions, we also demonstrate promising performance when one of the sensors lacks functionality.

Alexander Tsaregorodtsev, Adrian Holzbock, Jan Strohbeck et al.

Connected and cooperative driving requires precise calibration of the roadside infrastructure for having a reliable perception system. To solve this requirement in an automated manner, we present a robust extrinsic calibration method for automated geo-referenced camera calibration. Our method requires a calibration vehicle equipped with a combined GNSS/RTK receiver and an inertial measurement unit (IMU) for self-localization. In order to remove any requirements for the target's appearance and the local traffic conditions, we propose a novel approach using hypothesis filtering. Our method does not require any human interaction with the information recorded by both the infrastructure and the vehicle. Furthermore, we do not limit road access for other road users during calibration. We demonstrate the feasibility and accuracy of our approach by evaluating our approach on synthetic datasets as well as a real-world connected intersection, and deploying the calibration on real infrastructure. Our source code is publicly available.

Yutong Yang, Katarina Popović, Julian Wiederer et al.

Detection Transformer (DETR) and its variants show strong performance on object detection, a key task for autonomous systems. However, a critical limitation of these models is that their confidence scores only reflect semantic uncertainty, failing to capture the equally important spatial uncertainty. This results in an incomplete assessment of the detection reliability. On the other hand, Deep Ensembles can tackle this by providing high-quality spatial uncertainty estimates. However, their immense memory consumption makes them impractical for real-world applications. A cheaper alternative, Monte Carlo (MC) Dropout, suffers from high latency due to the need of multiple forward passes during inference to estimate uncertainty. To address these limitations, we introduce GroupEnsemble, an efficient and effective uncertainty estimation method for DETR-like models. GroupEnsemble simultaneously predicts multiple individual detection sets by feeding additional diverse groups of object queries to the transformer decoder during inference. Each query group is transformed by the shared decoder in isolation and predicts a complete detection set for the same input. An attention mask is applied to the decoder to prevent inter-group query interactions, ensuring each group detects independently to achieve reliable ensemble-based uncertainty estimation. By leveraging the decoder's inherent parallelism, GroupEnsemble efficiently estimates uncertainty in a single forward pass without sequential repetition. We validated our method under autonomous driving scenes and common daily scenes using the Cityscapes and COCO datasets, respectively. The results show that a hybrid approach combining MC-Dropout and GroupEnsemble outperforms Deep Ensembles on several metrics at a fraction of the cost. The code is available at https://github.com/yutongy98/GroupEnsemble.

Alexander Tsaregorodtsev, Michael Buchholz, Vasileios Belagiannis

While automotive radar sensors are widely adopted and have been used for automatic cruise control and collision avoidance tasks, their application outside of vehicles is still limited. As they have the ability to resolve multiple targets in 3D space, radars can also be used for improving environment perception. This application, however, requires a precise calibration, which is usually a time-consuming and labor-intensive task. We, therefore, present an approach for automated and geo-referenced extrinsic calibration of automotive radar sensors that is based on a novel hypothesis filtering scheme. Our method does not require external modifications of a vehicle and instead uses the location data obtained from automated vehicles. This location data is then combined with filtered sensor data to create calibration hypotheses. Subsequent filtering and optimization recovers the correct calibration. Our evaluation on data from a real testing site shows that our method can correctly calibrate infrastructure sensors in an automated manner, thus enabling cooperative driving scenarios.

Adrian Holzbock, Alexander Tsaregorodtsev, Vasileios Belagiannis

Besides interacting correctly with other vehicles, automated vehicles should also be able to react in a safe manner to vulnerable road users like pedestrians or cyclists. For a safe interaction between pedestrians and automated vehicles, the vehicle must be able to interpret the pedestrian's behavior. Common environment models do not contain information like body poses used to understand the pedestrian's intent. In this work, we propose an environment model that includes the position of the pedestrians as well as their pose information. We only use images from a monocular camera and the vehicle's localization data as input to our pedestrian environment model. We extract the skeletal information with a neural network human pose estimator from the image. Furthermore, we track the skeletons with a simple tracking algorithm based on the Hungarian algorithm and an ego-motion compensation. To obtain the 3D information of the position, we aggregate the data from consecutive frames in conjunction with the vehicle position. We demonstrate our pedestrian environment model on data generated with the CARLA simulator and the nuScenes dataset. Overall, we reach a relative position error of around 16% on both datasets.

Johanna S. Fröhlich, Bastian Heinlein, Jan U. Claar et al.

Explainable artificial intelligence has emerged as a promising field of research to address reliability concerns in artificial intelligence. Despite significant progress in explainable artificial intelligence, few methods provide a systematic way to visualize and understand how classes are confused and how their relationships evolve as training progresses. In this work, we present GRAPHIC, an architecture-agnostic approach that analyzes neural networks on a class level. It leverages confusion matrices derived from intermediate layers using linear classifiers. We interpret these as adjacency matrices of directed graphs, allowing tools from network science to visualize and quantify learning dynamics across training epochs and intermediate layers. GRAPHIC provides insights into linear class separability, dataset issues, and architectural behavior, revealing, for example, similarities between flatfish and man and labeling ambiguities validated in a human study. In summary, by uncovering real confusions, GRAPHIC offers new perspectives on how neural networks learn. The code is available at https://github.com/Johanna-S-Froehlich/GRAPHIC.

Osman Ülger, Julian Wiederer, Mohsen Ghafoorian et al.

Graph neural networks have shown to learn effective node representations, enabling node-, link-, and graph-level inference. Conventional graph networks assume static relations between nodes, while relations between entities in a video often evolve over time, with nodes entering and exiting dynamically. In such temporally-dynamic graphs, a core problem is inferring the future state of spatio-temporal edges, which can constitute multiple types of relations. To address this problem, we propose MTD-GNN, a graph network for predicting temporally-dynamic edges for multiple types of relations. We propose a factorized spatio-temporal graph attention layer to learn dynamic node representations and present a multi-task edge prediction loss that models multiple relations simultaneously. The proposed architecture operates on top of scene graphs that we obtain from videos through object detection and spatio-temporal linking. Experimental evaluations on ActionGenome and CLEVRER show that modeling multiple relations in our temporally-dynamic graph network can be mutually beneficial, outperforming existing static and spatio-temporal graph neural networks, as well as state-of-the-art predicate classification methods.

Mirlan Karimov, Teodora Spasojevic, Markus Braun et al.

Controllable video generation has emerged as a versatile tool for autonomous driving, enabling realistic synthesis of traffic scenarios. However, existing methods depend on control signals at inference time to guide the generative model towards temporally consistent generation of dynamic objects, limiting their utility as scalable and generalizable data engines. In this work, we propose Localized Semantic Alignment (LSA), a simple yet effective framework for fine-tuning pre-trained video generation models. LSA enhances temporal consistency by aligning semantic features between ground-truth and generated video clips. Specifically, we compare the output of an off-the-shelf feature extraction model between the ground-truth and generated video clips localized around dynamic objects inducing a semantic feature consistency loss. We fine-tune the base model by combining this loss with the standard diffusion loss. The model fine-tuned for a single epoch with our novel loss outperforms the baselines in common video generation evaluation metrics. To further test the temporal consistency in generated videos we adapt two additional metrics from object detection task, namely mAP and mIoU. Extensive experiments on nuScenes and KITTI datasets show the effectiveness of our approach in enhancing temporal consistency in video generation without the need for external control signals during inference and any computational overheads.

Julia Hornauer, Amir El-Ghoussani, Vasileios Belagiannis

Monocular depth estimation, similar to other image-based tasks, is prone to erroneous predictions due to ambiguities in the image, for example, caused by dynamic objects or shadows. For this reason, pixel-wise uncertainty assessment is required for safety-critical applications to highlight the areas where the prediction is unreliable. We address this in a post hoc manner and introduce gradient-based uncertainty estimation for already trained depth estimation models. To extract gradients without depending on the ground truth depth, we introduce an auxiliary loss function based on the consistency of the predicted depth and a reference depth. The reference depth, which acts as pseudo ground truth, is in fact generated using a simple image or feature augmentation, making our approach simple and effective. To obtain the final uncertainty score, the derivatives w.r.t. the feature maps from single or multiple layers are calculated using back-propagation. We demonstrate that our gradient-based approach is effective in determining the uncertainty without re-training using the two standard depth estimation benchmarks KITTI and NYU. In particular, for models trained with monocular sequences and therefore most prone to uncertainty, our method outperforms related approaches. In addition, we publicly provide our code and models: https://github.com/jhornauer/GrUMoDepth

Patrick Schmidt, Vasileios Belagiannis, Lazaros Nalpantidis

Autonomous robotic systems applied to new domains require an abundance of expensive, pixel-level dense labels to train robust semantic segmentation models under full supervision. This study proposes a model-agnostic Depth Edge Alignment Loss to improve Weakly Supervised Semantic Segmentation models across different datasets. The methodology generates pixel-level semantic labels from image-level supervision, avoiding expensive annotation processes. While weak supervision is widely explored in traditional computer vision, our approach adds supervision with pixel-level depth information, a modality commonly available in robotic systems. We demonstrate how our approach improves segmentation performance across datasets and models, but can also be combined with other losses for even better performance, with improvements up to +5.439, +1.274 and +16.416 points in mean Intersection over Union on the PASCAL VOC / MS COCO validation, and the HOPE static onboarding split, respectively. Our code will be made publicly available.

Rohan Asthana, Joschua Conrad, Maurits Ortmanns et al.

Zero-shot Neural Architecture Search (NAS) typically optimises the architecture search process by exploiting the network or gradient properties at initialisation through zero-cost proxies. The existing proxies often rely on labelled data, which is usually unavailable in real-world settings. Furthermore, the majority of the current methods focus either on optimising the convergence and generalisation attributes or solely on the expressivity of the network architectures. To address both limitations, we first demonstrate how channel collinearity affects the convergence and generalisation properties of a neural network. Then, by incorporating the convergence, generalisation and expressivity in one approach, we propose a zero-cost proxy that omits the requirement of labelled data for its computation. In particular, we leverage the Singular Value Decomposition (SVD) of the neural network layer features and the extrinsic curvature of the network output to design our proxy. %As a result, the proposed proxy is formulated as the simplified harmonic mean of the logarithms of two key components: the sum of the inverse of the feature condition number and the extrinsic curvature of the network output. Our approach enables accurate prediction of network performance on test data using only a single label-free data sample. Our extensive evaluation includes a total of six experiments, including the Convolutional Neural Network (CNN) search space, i.e. DARTS and the Transformer search space, i.e. AutoFormer. The proposed proxy demonstrates a superior performance on multiple correlation benchmarks, including NAS-Bench-101, NAS-Bench-201, and TransNAS-Bench-101-micro; as well as on the NAS task within the DARTS and the AutoFormer search space, all while being notably efficient. The code is available at https://github.com/rohanasthana/Dextr.

Marc Hölle, Walter Kellermann, Vasileios Belagiannis

Semantic segmentation models trained on known object classes often fail in real-world autonomous driving scenarios by confidently misclassifying unknown objects. While pixel-wise out-of-distribution detection can identify unknown objects, existing methods struggle in complex scenes where rare object classes are often confused with truly unknown objects. We introduce an uncertainty-aware likelihood ratio estimation method that addresses these limitations. Our approach uses an evidential classifier within a likelihood ratio test to distinguish between known and unknown pixel features from a semantic segmentation model, while explicitly accounting for uncertainty. Instead of producing point estimates, our method outputs probability distributions that capture uncertainty from both rare training examples and imperfect synthetic outliers. We show that by incorporating uncertainty in this way, outlier exposure can be leveraged more effectively. Evaluated on five standard benchmark datasets, our method achieves the lowest average false positive rate (2.5%) among state-of-the-art while maintaining high average precision (90.91%) and incurring only negligible computational overhead. Code is available at https://github.com/glasbruch/ULRE.

Yuyuan Liu, Yu Tian, Yuanhong Chen et al.

Consistency learning using input image, feature, or network perturbations has shown remarkable results in semi-supervised semantic segmentation, but this approach can be seriously affected by inaccurate predictions of unlabelled training images. There are two consequences of these inaccurate predictions: 1) the training based on the "strict" cross-entropy (CE) loss can easily overfit prediction mistakes, leading to confirmation bias; and 2) the perturbations applied to these inaccurate predictions will use potentially erroneous predictions as training signals, degrading consistency learning. In this paper, we address the prediction accuracy problem of consistency learning methods with novel extensions of the mean-teacher (MT) model, which include a new auxiliary teacher, and the replacement of MT's mean square error (MSE) by a stricter confidence-weighted cross-entropy (Conf-CE) loss. The accurate prediction by this model allows us to use a challenging combination of network, input data and feature perturbations to improve the consistency learning generalisation, where the feature perturbations consist of a new adversarial perturbation. Results on public benchmarks show that our approach achieves remarkable improvements over the previous SOTA methods in the field. Our code is available at https://github.com/yyliu01/PS-MT.

Filipe R. Cordeiro, Vasileios Belagiannis, Ian Reid et al.

The most competitive noisy label learning methods rely on an unsupervised classification of clean and noisy samples, where samples classified as noisy are re-labelled and "MixMatched" with the clean samples. These methods have two issues in large noise rate problems: 1) the noisy set is more likely to contain hard samples that are in-correctly re-labelled, and 2) the number of samples produced by MixMatch tends to be reduced because it is constrained by the small clean set size. In this paper, we introduce the learning algorithm PropMix to handle the issues above. PropMix filters out hard noisy samples, with the goal of increasing the likelihood of correctly re-labelling the easy noisy samples. Also, PropMix places clean and re-labelled easy noisy samples in a training set that is augmented with MixUp, removing the clean set size constraint and including a large proportion of correctly re-labelled easy noisy samples. We also include self-supervised pre-training to improve robustness to high noisy label scenarios. Our experiments show that PropMix has state-of-the-art (SOTA) results on CIFAR-10/-100(with symmetric, asymmetric and semantic label noise), Red Mini-ImageNet (from the Controlled Noisy Web Labels), Clothing1M and WebVision. In severe label noise bench-marks, our results are substantially better than other methods. The code is available athttps://github.com/filipe-research/PropMix.

Julian Wiederer, Arij Bouazizi, Marco Troina et al.

Human drivers can recognise fast abnormal driving situations to avoid accidents. Similar to humans, automated vehicles are supposed to perform anomaly detection. In this work, we propose the spatio-temporal graph auto-encoder for learning normal driving behaviours. Our innovation is the ability to jointly learn multiple trajectories of a dynamic number of agents. To perform anomaly detection, we first estimate a density function of the learned trajectory feature representation and then detect anomalies in low-density regions. Due to the lack of multi-agent trajectory datasets for anomaly detection in automated driving, we introduce our dataset using a driving simulator for normal and abnormal manoeuvres. Our evaluations show that our approach learns the relation between different agents and delivers promising results compared to the related works. The code, simulation and the dataset are publicly available on https://github.com/againerju/maad_highway.

Arij Bouazizi, Ulrich Kressel, Vasileios Belagiannis

We present a simple, yet effective, approach for self-supervised 3D human pose estimation. Unlike the prior work, we explore the temporal information next to the multi-view self-supervision. During training, we rely on triangulating 2D body pose estimates of a multiple-view camera system. A temporal convolutional neural network is trained with the generated 3D ground-truth and the geometric multi-view consistency loss, imposing geometrical constraints on the predicted 3D body skeleton. During inference, our model receives a sequence of 2D body pose estimates from a single-view to predict the 3D body pose for each of them. An extensive evaluation shows that our method achieves state-of-the-art performance in the Human3.6M and MPI-INF-3DHP benchmarks. Our code and models are publicly available at \url{https://github.com/vru2020/TM_HPE/}.

Filipe R. Cordeiro, Ragav Sachdeva, Vasileios Belagiannis et al.

Deep neural network models are robust to a limited amount of label noise, but their ability to memorise noisy labels in high noise rate problems is still an open issue. The most competitive noisy-label learning algorithms rely on a 2-stage process comprising an unsupervised learning to classify training samples as clean or noisy, followed by a semi-supervised learning that minimises the empirical vicinal risk (EVR) using a labelled set formed by samples classified as clean, and an unlabelled set with samples classified as noisy. In this paper, we hypothesise that the generalisation of such 2-stage noisy-label learning methods depends on the precision of the unsupervised classifier and the size of the training set to minimise the EVR. We empirically validate these two hypotheses and propose the new 2-stage noisy-label training algorithm LongReMix. We test LongReMix on the noisy-label benchmarks CIFAR-10, CIFAR-100, WebVision, Clothing1M, and Food101-N. The results show that our LongReMix generalises better than competing approaches, particularly in high label noise problems. Furthermore, our approach achieves state-of-the-art performance in most datasets. The code is available at https://github.com/filipe-research/LongReMix.

Fengbei Liu, Yuanhong Chen, Yu Tian et al.

Real-world large-scale medical image analysis (MIA) datasets have three challenges: 1) they contain noisy-labelled samples that affect training convergence and generalisation, 2) they usually have an imbalanced distribution of samples per class, and 3) they normally comprise a multi-label problem, where samples can have multiple diagnoses. Current approaches are commonly trained to solve a subset of those problems, but we are unaware of methods that address the three problems simultaneously. In this paper, we propose a new training module called Non-Volatile Unbiased Memory (NVUM), which non-volatility stores running average of model logits for a new regularization loss on noisy multi-label problem. We further unbias the classification prediction in NVUM update for imbalanced learning problem. We run extensive experiments to evaluate NVUM on new benchmarks proposed by this paper, where training is performed on noisy multi-label imbalanced chest X-ray (CXR) training sets, formed by Chest-Xray14 and CheXpert, and the testing is performed on the clean multi-label CXR datasets OpenI and PadChest. Our method outperforms previous state-of-the-art CXR classifiers and previous methods that can deal with noisy labels on all evaluations. Our code is available at https://github.com/FBLADL/NVUM.

Ragav Sachdeva, Filipe R. Cordeiro, Vasileios Belagiannis et al.

The efficacy of deep learning depends on large-scale data sets that have been carefully curated with reliable data acquisition and annotation processes. However, acquiring such large-scale data sets with precise annotations is very expensive and time-consuming, and the cheap alternatives often yield data sets that have noisy labels. The field has addressed this problem by focusing on training models under two types of label noise: 1) closed-set noise, where some training samples are incorrectly annotated to a training label other than their known true class; and 2) open-set noise, where the training set includes samples that possess a true class that is (strictly) not contained in the set of known training labels. In this work, we study a new variant of the noisy label problem that combines the open-set and closed-set noisy labels, and introduce a benchmark evaluation to assess the performance of training algorithms under this setup. We argue that such problem is more general and better reflects the noisy label scenarios in practice. Furthermore, we propose a novel algorithm, called EvidentialMix, that addresses this problem and compare its performance with the state-of-the-art methods for both closed-set and open-set noise on the proposed benchmark. Our results show that our method produces superior classification results and better feature representations than previous state-of-the-art methods. The code is available at https://github.com/ragavsachdeva/EvidentialMix.

Nico Engel, Vasileios Belagiannis, Klaus Dietmayer

In this work, we present Point Transformer, a deep neural network that operates directly on unordered and unstructured point sets. We design Point Transformer to extract local and global features and relate both representations by introducing the local-global attention mechanism, which aims to capture spatial point relations and shape information. For that purpose, we propose SortNet, as part of the Point Transformer, which induces input permutation invariance by selecting points based on a learned score. The output of Point Transformer is a sorted and permutation invariant feature list that can directly be incorporated into common computer vision applications. We evaluate our approach on standard classification and part segmentation benchmarks to demonstrate competitive results compared to the prior work. Code is publicly available at: https://github.com/engelnico/point-transformer

Rohan Asthana, Vasileios Belagiannis

Diffusion-based image generative models produce high-fidelity images through iterative denoising but remain vulnerable to memorization, where they unintentionally reproduce exact copies or parts of training images. Recent memorization detection methods are primarily based on the norm of score difference as indicators of memorization. We prove that such norm-based metrics are mainly effective under the assumption of isotropic log-probability distributions, which generally holds at high or medium noise levels. In contrast, analyzing the anisotropic regime reveals that memorized samples exhibit strong angular alignment between the guidance vector and unconditional scores in the low-noise setting. Through these insights, we develop a memorization detection metric by integrating isotropic norm and anisotropic alignment. Our detection metric can be computed directly on pure noise inputs via two conditional and unconditional forward passes, eliminating the need for costly denoising steps. Detection experiments on Stable Diffusion v1.4 and v2 show that our metric outperforms existing denoising-free detection methods while being at least approximately 5x faster than the previous best approach. Finally, we demonstrate the effectiveness of our approach by utilizing a mitigation strategy that adapts memorized prompts based on our developed metric.

Rohan Asthana, Joschua Conrad, Youssef Dawoud et al.

Neural architecture search automates the design of neural network architectures usually by exploring a large and thus complex architecture search space. To advance the architecture search, we present a graph diffusion-based NAS approach that uses discrete conditional graph diffusion processes to generate high-performing neural network architectures. We then propose a multi-conditioned classifier-free guidance approach applied to graph diffusion networks to jointly impose constraints such as high accuracy and low hardware latency. Unlike the related work, our method is completely differentiable and requires only a single model training. In our evaluations, we show promising results on six standard benchmarks, yielding novel and unique architectures at a fast speed, i.e. less than 0.2 seconds per architecture. Furthermore, we demonstrate the generalisability and efficiency of our method through experiments on ImageNet dataset.

Michele De Vita, Vasileios Belagiannis

Despite the remarkable progress in generative modelling, current diffusion models lack a quantitative approach to assess image quality. To address this limitation, we propose to estimate the pixel-wise aleatoric uncertainty during the sampling phase of diffusion models and utilise the uncertainty to improve the sample generation quality. The uncertainty is computed as the variance of the denoising scores with a perturbation scheme that is specifically designed for diffusion models. We then show that the aleatoric uncertainty estimates are related to the second-order derivative of the diffusion noise distribution. We evaluate our uncertainty estimation algorithm and the uncertainty-guided sampling on the ImageNet and CIFAR-10 datasets. In our comparisons with the related work, we demonstrate promising results in filtering out low quality samples. Furthermore, we show that our guided approach leads to better sample generation in terms of FID scores.

Yu Yao, Salil Bhatnagar, Markus Mazzola et al.

Rare, yet critical, scenarios pose a significant challenge in testing and evaluating autonomous driving planners. Relying solely on real-world driving scenes requires collecting massive datasets to capture these scenarios. While automatic generation of traffic scenarios appears promising, data-driven models require extensive training data and often lack fine-grained control over the output. Moreover, generating novel scenarios from scratch can introduce a distributional shift from the original training scenes which undermines the validity of evaluations especially for learning-based planners. To sidestep this, recent work proposes to generate challenging scenarios by augmenting original scenarios from the test set. However, this involves the manual augmentation of scenarios by domain experts. An approach that is unable to meet the demands for scale in the evaluation of self-driving systems. Therefore, this paper introduces a novel LLM-agent based framework for augmenting real-world traffic scenarios using natural language descriptions, addressing the limitations of existing methods. A key innovation is the use of an agentic design, enabling fine-grained control over the output and maintaining high performance even with smaller, cost-effective LLMs. Extensive human expert evaluation demonstrates our framework's ability to accurately adhere to user intent, generating high quality augmented scenarios comparable to those created manually.

Marcel Schreiber, Vasileios Belagiannis, Claudius Gläser et al.

A common approach for modeling the environment of an autonomous vehicle are dynamic occupancy grid maps, in which the surrounding is divided into cells, each containing the occupancy and velocity state of its location. Despite the advantage of modeling arbitrary shaped objects, the used algorithms rely on hand-designed inverse sensor models and semantic information is missing. Therefore, we introduce a multi-task recurrent neural network to predict grid maps providing occupancies, velocity estimates, semantic information and the driveable area. During training, our network architecture, which is a combination of convolutional and recurrent layers, processes sequences of raw lidar data, that is represented as bird's eye view images with several height channels. The multi-task network is trained in an end-to-end fashion to predict occupancy grid maps without the usual preprocessing steps consisting of removing ground points and applying an inverse sensor model. In our evaluations, we show that our learned inverse sensor model is able to overcome some limitations of a geometric inverse sensor model in terms of representing object shapes and modeling freespace. Moreover, we report a better runtime performance and more accurate semantic predictions for our end-to-end approach, compared to our network relying on measurement grid maps as input data.

Fengbei Liu, Yu Tian, Yuanhong Chen et al.

Effective semi-supervised learning (SSL) in medical image analysis (MIA) must address two challenges: 1) work effectively on both multi-class (e.g., lesion classification) and multi-label (e.g., multiple-disease diagnosis) problems, and 2) handle imbalanced learning (because of the high variance in disease prevalence). One strategy to explore in SSL MIA is based on the pseudo labelling strategy, but it has a few shortcomings. Pseudo-labelling has in general lower accuracy than consistency learning, it is not specifically designed for both multi-class and multi-label problems, and it can be challenged by imbalanced learning. In this paper, unlike traditional methods that select confident pseudo label by threshold, we propose a new SSL algorithm, called anti-curriculum pseudo-labelling (ACPL), which introduces novel techniques to select informative unlabelled samples, improving training balance and allowing the model to work for both multi-label and multi-class problems, and to estimate pseudo labels by an accurate ensemble of classifiers (improving pseudo label accuracy). We run extensive experiments to evaluate ACPL on two public medical image classification benchmarks: Chest X-Ray14 for thorax disease multi-label classification and ISIC2018 for skin lesion multi-class classification. Our method outperforms previous SOTA SSL methods on both datasets

Anastasia Makarevich, Azade Farshad, Vasileios Belagiannis et al.

The lack of sufficient annotated image data is a common issue in medical image segmentation. For some organs and densities, the annotation may be scarce, leading to poor model training convergence, while other organs have plenty of annotated data. In this work, we present MetaMedSeg, a gradient-based meta-learning algorithm that redefines the meta-learning task for the volumetric medical data with the goal to capture the variety between the slices. We also explore different weighting schemes for gradients aggregation, arguing that different tasks might have different complexity, and hence, contribute differently to the initialization. We propose an importance-aware weighting scheme to train our model. In the experiments, we present an evaluation of the medical decathlon dataset by extracting 2D slices from CT and MRI volumes of different organs and performing semantic segmentation. The results show that our proposed volumetric task definition leads to up to 30% improvement in terms of IoU compared to related baselines. The proposed update rule is also shown to improve the performance for complex scenarios where the data distribution of the target organ is very different from the source organs.

Arij Bouazizi, Julian Wiederer, Ulrich Kressel et al.

We present a self-supervised learning algorithm for 3D human pose estimation of a single person based on a multiple-view camera system and 2D body pose estimates for each view. To train our model, represented by a deep neural network, we propose a four-loss function learning algorithm, which does not require any 2D or 3D body pose ground-truth. The proposed loss functions make use of the multiple-view geometry to reconstruct 3D body pose estimates and impose body pose constraints across the camera views. Our approach utilizes all available camera views during training, while the inference is single-view. In our evaluations, we show promising performance on Human3.6M and HumanEva benchmarks, while we also present a generalization study on MPI-INF-3DHP dataset, as well as several ablation results. Overall, we outperform all self-supervised learning methods and reach comparable results to supervised and weakly-supervised learning approaches. Our code and models are publicly available

Julia Hornauer, Lazaros Nalpantidis, Vasileios Belagiannis

Real-world perception systems in many cases build on hardware with limited resources to adhere to cost and power limitations of their carrying system. Deploying deep neural networks on resource-constrained hardware became possible with model compression techniques, as well as efficient and hardware-aware architecture design. However, model adaptation is additionally required due to the diverse operation environments. In this work, we address the problem of training deep neural networks on resource-constrained hardware in the context of visual domain adaptation. We select the task of monocular depth estimation where our goal is to transform a pre-trained model to the target's domain data. While the source domain includes labels, we assume an unlabelled target domain, as it happens in real-world applications. Then, we present an adversarial learning approach that is adapted for training on the device with limited resources. Since visual domain adaptation, i.e. neural network training, has not been previously explored for resource-constrained hardware, we present the first feasibility study for image-based depth estimation. Our experiments show that visual domain adaptation is relevant only for efficient network architectures and training sets at the order of a few hundred samples. Models and code are publicly available.

Nico Engel, Vasileios Belagiannis, Klaus Dietmayer

We present a vehicle self-localization method using point-based deep neural networks. Our approach processes measurements and point features, i.e. landmarks, from a high-definition digital map to infer the vehicle's pose. To learn the best association and incorporate local information between the point sets, we propose an attention mechanism that matches the measurements to the corresponding landmarks. Finally, we use this representation for the point-cloud registration and the subsequent pose regression task. Furthermore, we introduce a training simulation framework that artificially generates measurements and landmarks to facilitate the deployment process and reduce the cost of creating extensive datasets from real-world data. We evaluate our method on our dataset, as well as an adapted version of the Kitti odometry dataset, where we achieve superior performance compared to related approaches; and additionally show dominant generalization capabilities.