Vision foundation models (VFMs) trained on large-scale image datasets provide high-quality features that have significantly advanced$2 D$visual recognition. However, their potential in 3D scene segmentation remains largely untapped, despite the common availability of$2 D$images alongside 3D point cloud datasets. While significant research has been dedicated to 2D-3D fusion, recent state-of-the-art 3D methods predominantly focus on 3D data, leaving the integration of VFMs into 3D models underexplored. In this work, we challenge this trend by introducing DITR, a generally applicable approach that extracts$2 D$foundation model features, projects them to 3D, and finally injects them into a 3D point cloud segmentation model. DITR achieves state-of-the-art results on both indoor and outdoor 3D semantic segmentation benchmarks. To enable the use of VFMs even when images are unavailable during inference, we additionally propose to pretrain 3D models by distilling 2D foundation models. By initializing the 3D backbone with knowledge distilled from 2D VFMs, we create a strong basis for downstream 3D segmentation tasks, ultimately boosting performance across various datasets.

Despite the advances in automated medical image segmentation, AI models still underperform in various clinical settings, posing challenges for integration into real-world workflows. In this pre-registered prospective multicenter evaluation, we analyzed 20 state-of-the-art mandibular segmentation models across 19,218 segmentations of 1,000 clinically resampled CT/CBCT scans. Our results suggest that for a given model, segmentation accuracy can vary by up to 25% in Dice score as socio-technical factors such as voxel size, bone orientation, and patient conditions (e.g., osteosynthesis or pathology) shift from favorable to adverse. Higher sharpness, isotropic smaller voxels, and neutral orientation significantly improved results, while metallic osteosynthesis and anatomical complexity led to significant degradation. Our findings challenge the common view of AI models as “plug-and-play” tools and suggest evidence-based optimization recommendations for both clinicians and developers. This will in turn boost the integration of AI segmentation tools in routine healthcare.

Vision Transformers (ViTs) have shown remarkable performance and scalability across various computer vision tasks. To apply single-scale ViTs to image segmentation, existing methods adopt a convolutional adapter to generate multi-scale features, a pixel decoder to fuse these features, and a Transformer decoder that uses the fused features to make predictions. In this paper, we show that the inductive biases introduced by these task-specific components can instead be learned by the ViT itself, given sufficiently large models and extensive pre-training. Based on these findings, we introduce the Encoder-Only Mask Transformer (EoMT), which repurposes the plain ViT architecture to conduct image segmentation. With large-scale models and pre-training, EoMT obtains a segmentation accuracy similar to state-of-the-art models that use task-specific components. At the same time, EoMT is significantly faster than these methods due to its architectural simplicity, e.g., up to 4 × faster with ViT-L. Across a range of model sizes, EoMT demonstrates an optimal balance between segmentation accuracy and prediction speed, suggesting that compute resources are better spent on scaling the ViT itself rather than adding architectural complexity. Code: https://www.tue-mps.org/eomt/.

Safe navigation of self-driving cars and robots requires a precise understanding of their environment. Training data for perception systems cannot cover the wide variety of objects that may appear during deployment. Thus, reliable identification of unknown objects, such as wild animals and untypical obstacles, is critical due to their potential to cause serious accidents. Significant progress in semantic segmentation of anomalies has been facilitated by the availability of out-of-distribution (OOD) benchmarks. However, a comprehensive understanding of scene dynamics requires the segmentation of individual objects, and thus the segmentation of instances is essential. Development in this area has been lagging, largely due to the lack of dedicated benchmarks. The situation is similar in object detection. While there is interest in detecting and potentially tracking every anomalous object, the availability of dedicated benchmarks is clearly limited. To address this gap, this work extends some commonly used anomaly segmentation benchmarks to include the instance segmentation and object detection tasks. Our evaluation of anomaly instance segmentation and object detection methods shows that both of these challenges remain unsolved problems. We provide a competition and benchmark website under https://vision.rwth-aachen.de/oodis.

Recent work showed that large diffusion models can be reused as highly precise monocular depth estimators by casting depth estimation as an image-conditional image generation task. While the proposed model achieved state-of-the-art results, high computational demands due to multi-step inference limited its use in many scenarios. In this paper, we show that the perceived inefficiency was caused by a flaw in the inference pipeline that has so far gone unnoticed. The fixed model performs comparably to the best previously reported configuration while being more than 200x faster. To optimize for downstream task performance, we perform end-to-end fine-tuning on top of the single-step model with task-specific losses and get a deterministic model that outperforms all other diffusion-based depth and normal estimation models on common zero-shot benchmarks. We surprisingly find that this fine-tuning protocol also works directly on Stable Diffusion and achieves comparable performance to current state-of-the-art diffusion-based depth and normal estimation models, calling into question some of the conclusions drawn from prior works.

The general domain of video segmentation is currently fragmented into different tasks spanning multiple benchmarks. Despite rapid progress in the state-of-the-art, current methods are overwhelmingly task-specific and cannot conceptually generalize to other tasks. Inspired by recent approaches with multi-task capability, we propose TarViS: a novel, unified network architecture that can be applied to any task that requires segmenting a set of arbitrarily defined ‘targets' in video. Our approach is flexible with respect to how tasks define these targets, since it models the latter as abstract ‘queries' which are then used to predict pixel-precise target masks. A single TarViS model can be trained jointly on a collection of datasets spanning different tasks, and can hot-swap between tasks during inference without any task-specific retraining. To demonstrate its effectiveness, we apply TarViS to four different tasks, namely Video Instance Segmentation (VIS), Video Panoptic Segmentation (VPS), Video Object Segmentation (VOS) and Point Exemplar-guided Tracking (PET). Our unified, jointly trained model achieves state-of-the-art performance on 5/7 benchmarks spanning these four tasks, and competitive performance on the remaining two. Code and model weights are available at: https://github.com/Ali2500/TarViS

Most state-of-the-art instance segmentation methods rely on large amounts of pixel-precise ground-truth annotations for training, which are expensive to create. Interactive segmentation networks help generate such annotations based on an image and the corresponding user interactions such as clicks. Existing methods for this task can only process a single instance at a time and each user interaction requires a full forward pass through the entire deep network. We introduce a more efficient approach, called DynaMITe, in which we represent user interactions as spatio-temporal queries to a Transformer decoder with a potential to segment multiple object instances in a single iteration. Our architecture also alleviates any need to re-compute image features during refinement, and requires fewer interactions for segmenting multiple instances in a single image when compared to other methods. DynaMITe achieves state-of-the-art results on multiple existing interactive segmentation benchmarks, and also on the new multi-instance benchmark that we propose in this paper.

Modern 3D semantic instance segmentation approaches predominantly rely on specialized voting mechanisms followed by carefully designed geometric clustering techniques. Building on the successes of recent Transformer-based methods for object detection and image segmentation, we propose the first Transformer-based approach for 3D semantic instance segmentation. We show that we can leverage generic Transformer building blocks to directly predict instance masks from 3D point clouds. In our model - called Mask3D - each object instance is represented as an instance query. Using Transformer decoders, the instance queries are learned by iteratively attending to point cloud features at multiple scales. Combined with point features, the instance queries directly yield all instance masks in parallel. Mask3D has several advantages over current state-of-the-art approaches, since it neither relies on (1) voting schemes which require hand-selected geometric properties (such as centers) nor (2) geometric grouping mechanisms requiring manually-tuned hyper-parameters (e.g. radii) and (3) enables a loss that directly optimizes instance masks. Mask3D sets a new state-of-the-art on ScanNet test (+6.2mAP), S3DIS 6-fold (+10.1 mAP), STPLS3D (+11.2 mAP) and ScanNet200 test (+12.4 mAP).

Deep learning-based 3D human pose estimation performs best when trained on large amounts of labeled data, making combined learning from many datasets an important research direction. One obstacle to this endeavor are the different skeleton formats provided by different datasets, i.e., they do not label the same set of anatomical landmarks. There is little prior research on how to best supervise one model with such discrepant labels. We show that simply using separate output heads for different skeletons results in inconsistent depth estimates and insufficient information sharing across skeletons. As a remedy, we propose a novel affine-combining autoencoder (ACAE) method to perform dimensionality reduction on the number of landmarks. The discovered latent 3D points capture the redundancy among skeletons, enabling enhanced information sharing when used for consistency regularization. Our approach scales to an extreme multi-dataset regime, where we use 28 3D human pose datasets to supervise one model, which outperforms prior work on a range of benchmarks, including the challenging 3D Poses in the Wild (3DPW) dataset. Our code and models are available for research purposes.1

Existing state-of-the-art methods for Video Object Segmentation (VOS) learn low-level pixel-to-pixel correspondences between frames to propagate object masks across video. This requires a large amount of densely annotated video data, which is costly to annotate, and largely redundant since frames within a video are highly correlated. In light of this, we propose HODOR: a novel method that tackles VOS by effectively leveraging annotated static images for understanding object appearance and scene context. We encode object instances and scene information from an image frame into robust high-level descriptors which can then be used to re-segment those objects in different frames. As a result, HODOR achieves state-of-the-art performance on the DAVIS and YouTube-VOS benchmarks compared to existing methods trained without video annotations. With out any architectural modification, HODOR can also learn from video context around single annotated video frames by utilizing cyclic consistency, whereas other methods rely on dense, temporally consistent annotations. Source code: https://github.com/Ali2500/HODOR.

Person detection is a crucial task for mobile robots navigating in human-populated environments. LiDAR sensors are promising for this task, thanks to their accurate depth measurements and large field of view. Two types of LiDAR sensors exist: the 2D LiDAR sensors, which scan a single plane, and the 3D LiDAR sensors, which scan multiple planes, thus forming a volume. How do they compare for the task of person detection? To answer this, we conduct a series of exper-iments, using the public, large-scale JackRabbot dataset and the state-of-the-art 2D and 3D LiDAR-based person detectors (DR-SPAAM and CenterPoint respectively). Our experiments include multiple aspects, ranging from the basic performance and speed comparison, to more detailed analysis on localization accuracy and robustness against distance and scene clutter. The insights from these experiments highlight the strengths and weaknesses of 2D and 3D LiDAR sensors as sources for person detection, and are especially valuable for designing mobile robots that will operate in close proximity to surrounding humans (e.g. service or social robot).

Deep learning is the essential building block of state-of-the-art person detectors in 2D range data. However, only a few annotated datasets are available for training and testing these deep networks, potentially limiting their performance when deployed in new environments or with different LiDAR models. We propose a method, which uses bounding boxes from an image-based detector (e.g. Faster R-CNN) on a calibrated camera to automatically generate training labels (called pseudo-labels) for 2D LiDAR-based person detectors. Through experiments on the JackRabbot dataset with two detector models, DROW3 and DR-SPAAM, we show that self-supervised detectors, trained or fine-tuned with pseudolabels, outperform detectors trained only on a different dataset. Combined with robust training techniques, the self-supervised detectors reach a performance close to the ones trained using manual annotations of the target dataset. Our method is an effective way to improve person detectors during deployment without any additional labeling effort, and we release our source code to support relevant robotic applications.

Detecting persons using a 2D LiDAR is a challenging task due to the low information content of 2D range data. To alleviate the problem caused by the sparsity of the LiDAR points, current state-of-the-art methods fuse multiple previous scans and perform detection using the combined scans. The downside of such a backward looking fusion is that all the scans need to be aligned explicitly, and the necessary alignment operation makes the whole pipeline more expensive - often too expensive for real-world applications. In this paper, we propose a person detection network which uses an alternative strategy to combine scans obtained at different times. Our method, Distance Robust SPatial Attention and Auto-regressive Model (DR-SPAAM), follows a forward looking paradigm. It keeps the intermediate features from the backbone network as a template and recurrently updates the template when a new scan becomes available. The updated feature template is in turn used for detecting persons currently in the scene. On the DROW dataset, our method outperforms the existing state-of-the-art, while being approximately four times faster, running at 87.2 FPS on a laptop with a dedicated GPU and at 22.6 FPS on an NVIDIA Jetson AGX embedded GPU. We release our code in PyTorch and a ROS node including pre-trained models.

In this work, we tackle the problem of instance segmentation, the task of simultaneously solving object detection and semantic segmentation. Towards this goal, we present a model, called MaskLab, which produces three outputs: box detection, semantic segmentation, and direction prediction. Building on top of the Faster-RCNN object detector, the predicted boxes provide accurate localization of object instances. Within each region of interest, MaskLab performs foreground/background segmentation by combining semantic and direction prediction. Semantic segmentation assists the model in distinguishing between objects of different semantic classes including background, while the direction prediction, estimating each pixel's direction towards its corresponding center, allows separating instances of the same semantic class. Moreover, we explore the effect of incorporating recent successful methods from both segmentation and detection (i.e. atrous convolution and hypercolumn). Our proposed model is evaluated on the COCO instance segmentation benchmark and shows comparable performance with other state-of-art models.

Semantic image segmentation is an essential component of modern autonomous driving systems, as an accurate understanding of the surrounding scene is crucial to navigation and action planning. Current state-of-the-art approaches in semantic image segmentation rely on pre-trained networks that were initially developed for classifying images as a whole. While these networks exhibit outstanding recognition performance (i.e., what is visible?), they lack localization accuracy (i.e., where precisely is something located?). Therefore, additional processing steps have to be performed in order to obtain pixel-accurate segmentation masks at the full image resolution. To alleviate this problem we propose a novel ResNet-like architecture that exhibits strong localization and recognition performance. We combine multi-scale context with pixel-level accuracy by using two processing streams within our network: One stream carries information at the full image resolution, enabling precise adherence to segment boundaries. The other stream undergoes a sequence of pooling operations to obtain robust features for recognition. The two streams are coupled at the full image resolution using residuals. Without additional processing steps and without pre-training, our approach achieves an intersection-over-union score of 71.8% on the Cityscapes dataset.

Most vision based systems for object tracking in urban environments focus on a limited number of important object categories such as cars or pedestrians, for which powerful detectors are available. However, practical driving scenarios contain many additional objects of interest, for which suitable detectors either do not yet exist or would be cumbersome to obtain. In this paper we propose a more general tracking approach which does not follow the often used tracking-by-detection principle. Instead, we investigate how far we can get by tracking unknown, generic objects in challenging street scenes. As such, we do not restrict ourselves to only tracking the most common categories, but are able to handle a large variety of static and moving objects. We evaluate our approach on the KITTI dataset and show competitive results for the annotated classes, even though we are not restricted to them.

Dense semantic segmentation of 3D point clouds is a challenging task. Many approaches deal with 2D semantic segmentation and can obtain impressive results. With the availability of cheap RGB-D sensors the field of indoor semantic segmentation has seen a lot of progress. Still it remains unclear how to deal with 3D semantic segmentation in the best way. We propose a novel 2D-3D label transfer based on Bayesian updates and dense pairwise 3D Conditional Random Fields. This approach allows us to use 2D semantic segmentations to create a consistent 3D semantic reconstruction of indoor scenes. To this end, we also propose a fast 2D semantic segmentation approach based on Randomized Decision Forests. Furthermore, we show that it is not needed to obtain a semantic segmentation for every frame in a sequence in order to create accurate semantic 3D reconstructions. We evaluate our approach on both NYU Depth datasets and show that we can obtain a significant speed-up compared to other methods.