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. *

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.

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.

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.

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.

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.

Extracting informative representations from videos is fundamental for effectively learning various downstream tasks. We present a novel approach for unsupervised learning of meaningful representations from videos, leveraging the concept of image spatial entropy (ISE) that quantifies the per-pixel information in an image. We argue that local entropy of pixel neighborhoods and their temporal evolution create valuable intrinsic supervisory signals for learning prominent features. Building on this idea, we abstract visual features into a concise representation of keypoints that act as dynamic information transmitters, and design a deep learning model that learns, purely unsupervised, spatially and temporally consistent representations directly from video frames. Two original information-theoretic losses, computed from local entropy, guide our model to discover consistent keypoint representations; a loss that maximizes the spatial information covered by the keypoints and a loss that optimizes the keypoints’ information transportation over time. We compare our keypoint representation to strong baselines for various downstream tasks, e.g., learning object dynamics. Our empirical results show superior performance for our information-driven keypoints that resolve challenges like attendance to static and dynamic objects or objects abruptly entering and leaving the scene.

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.

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.

In this work, we address the difficulty of matching local features between images captured at distant points in time resulting in a global appearance change. Inspired by recent neural style transfer techniques, we propose to use an image transformation network to translate night images into day-like appearance, with the objective of better matching performance. We extend traditional style transfer, that optimize for content and style, with a keypoint matching loss function. The joint optimization of these losses allows our model to generate images that can significantly improve the performance of local feature matching, in a self-supervised way. As a result, our approach is flexible and does not require paired training data, which is difficult to obtain in practice. We show how our method can be used as an extension to a state-of-the-art differentiable feature extractor to improve its performance in challenging scenarios. This is demonstrated in our evaluation on day-night image matching and visual localization tasks with night-rain image queries.

While there are many deep learning based approaches for single image compression, the field of end-to-end learned video coding has remained much less explored. Therefore, in this work we present an inter-frame compression approach for neural video coding that can seamlessly build up on different existing neural image codecs. Our end-to-end solution performs temporal prediction by optical flow based motion compensation in pixel space. The key insight is that we can increase both decoding efficiency and reconstruction quality by encoding the required information into a latent representation that directly decodes into motion and blending coefficients. In order to account for remaining prediction errors, residual information between the original image and the interpolated frame is needed. We propose to compute residuals directly in latent space instead of in pixel space as this allows to reuse the same image compression network for both key frames and intermediate frames. Our extended evaluation on different datasets and resolutions shows that the rate-distortion performance of our approach is competitive with existing state-of-the-art codecs.

Most approaches for video frame interpolation require accurate dense correspondences to synthesize an in-between frame. Therefore, they do not perform well in challenging scenarios with e.g. lighting changes or motion blur. Recent deep learning approaches that rely on kernels to represent motion can only alleviate these problems to some extent. In those cases, methods that use a per-pixel phase-based motion representation have been shown to work well. However, they are only applicable for a limited amount of motion. We propose a new approach, PhaseNet, that is designed to robustly handle challenging scenarios while also coping with larger motion. Our approach consists of a neural network decoder that directly estimates the phase decomposition of the intermediate frame. We show that this is superior to the hand-crafted heuristics previously used in phase-based methods and also compares favorably to recent deep learning based approaches for video frame interpolation on challenging datasets.

Standard approaches to computing interpolated (in-between) frames in a video sequence require accurate pixel correspondences between images e.g. using optical flow. We present an efficient alternative by leveraging recent developments in phase-based methods that represent motion in the phase shift of individual pixels. This concept allows in-between images to be generated by simple per-pixel phase modification, without the need for any form of explicit correspondence estimation. Up until now, such methods have been limited in the range of motion that can be interpolated, which fundamentally restricts their usefulness. In order to reduce these limitations, we introduce a novel, bounded phase shift correction method that combines phase information across the levels of a multi-scale pyramid. Additionally, we propose extensions for phase-based image synthesis that yield smoother transitions between the interpolated images. Our approach avoids expensive global optimization typical of optical flow methods, and is both simple to implement and easy to parallelize. This allows us to interpolate frames at a fraction of the computational cost of traditional optical flow-based solutions, while achieving similar quality and in some cases even superior results. Our method fails gracefully in difficult interpolation settings, e.g., significant appearance changes, where flow-based methods often introduce serious visual artifacts. Due to its efficiency, our method is especially well suited for frame interpolation and retiming of high resolution, high frame rate video.