Jan Eric Lenssen

Wolfgang Boettcher, Lukas Hoyer, Ozan Unal et al.

In this work, we introduce Scribbles for All, a label and training data generation algorithm for semantic segmentation trained on scribble labels. Training or fine-tuning semantic segmentation models with weak supervision has become an important topic recently and was subject to significant advances in model quality. In this setting, scribbles are a promising label type to achieve high quality segmentation results while requiring a much lower annotation effort than usual pixel-wise dense semantic segmentation annotations. The main limitation of scribbles as source for weak supervision is the lack of challenging datasets for scribble segmentation, which hinders the development of novel methods and conclusive evaluations. To overcome this limitation, Scribbles for All provides scribble labels for several popular segmentation datasets and provides an algorithm to automatically generate scribble labels for any dataset with dense annotations, paving the way for new insights and model advancements in the field of weakly supervised segmentation. In addition to providing datasets and algorithm, we evaluate state-of-the-art segmentation models on our datasets and show that models trained with our synthetic labels perform competitively with respect to models trained on manual labels. Thus, our datasets enable state-of-the-art research into methods for scribble-labeled semantic segmentation. The datasets, scribble generation algorithm, and baselines are publicly available at https://github.com/wbkit/Scribbles4All

Garvita Tiwari, Dimitrije Antic, Jan Eric Lenssen et al.

We present Pose-NDF, a continuous model for plausible human poses based on neural distance fields (NDFs). Pose or motion priors are important for generating realistic new poses and for reconstructing accurate poses from noisy or partial observations. Pose-NDF learns a manifold of plausible poses as the zero level set of a neural implicit function, extending the idea of modeling implicit surfaces in 3D to the high-dimensional domain SO(3)^K, where a human pose is defined by a single data point, represented by K quaternions. The resulting high-dimensional implicit function can be differentiated with respect to the input poses and thus can be used to project arbitrary poses onto the manifold by using gradient descent on the set of 3-dimensional hyperspheres. In contrast to previous VAE-based human pose priors, which transform the pose space into a Gaussian distribution, we model the actual pose manifold, preserving the distances between poses. We demonstrate that PoseNDF outperforms existing state-of-the-art methods as a prior in various downstream tasks, ranging from denoising real-world human mocap data, pose recovery from occluded data to 3D pose reconstruction from images. Furthermore, we show that it can be used to generate more diverse poses by random sampling and projection than VAE-based methods.

Keyang Zhou, Bharat Lal Bhatnagar, Jan Eric Lenssen et al.

We present TOCH, a method for refining incorrect 3D hand-object interaction sequences using a data prior. Existing hand trackers, especially those that rely on very few cameras, often produce visually unrealistic results with hand-object intersection or missing contacts. Although correcting such errors requires reasoning about temporal aspects of interaction, most previous works focus on static grasps and contacts. The core of our method are TOCH fields, a novel spatio-temporal representation for modeling correspondences between hands and objects during interaction. TOCH fields are a point-wise, object-centric representation, which encode the hand position relative to the object. Leveraging this novel representation, we learn a latent manifold of plausible TOCH fields with a temporal denoising auto-encoder. Experiments demonstrate that TOCH outperforms state-of-the-art 3D hand-object interaction models, which are limited to static grasps and contacts. More importantly, our method produces smooth interactions even before and after contact. Using a single trained TOCH model, we quantitatively and qualitatively demonstrate its usefulness for correcting erroneous sequences from off-the-shelf RGB/RGB-D hand-object reconstruction methods and transferring grasps across objects.

Christopher Wewer, Eddy Ilg, Bernt Schiele et al.

Existing neural field representations for 3D object reconstruction either (1) utilize object-level representations, but suffer from low-quality details due to conditioning on a global latent code, or (2) are able to perfectly reconstruct the observations, but fail to utilize object-level prior knowledge to infer unobserved regions. We present SimNP, a method to learn category-level self-similarities, which combines the advantages of both worlds by connecting neural point radiance fields with a category-level self-similarity representation. Our contribution is two-fold. (1) We design the first neural point representation on a category level by utilizing the concept of coherent point clouds. The resulting neural point radiance fields store a high level of detail for locally supported object regions. (2) We learn how information is shared between neural points in an unconstrained and unsupervised fashion, which allows to derive unobserved regions of an object during the reconstruction process from given observations. We show that SimNP is able to outperform previous methods in reconstructing symmetric unseen object regions, surpassing methods that build upon category-level or pixel-aligned radiance fields, while providing semantic correspondences between instances

Nazir Nayal, Christopher Wewer, Jan Eric Lenssen

Diffusion models and their variations, such as rectified flows, generate diverse and high-quality images, but they are still hindered by slow iterative sampling caused by the highly curved generative paths they learn. An important cause of high curvature, as shown by previous work, is independence between the source distribution (standard Gaussian) and the data distribution. In this work, we tackle this limitation by two complementary contributions. First, we attempt to break away from the standard Gaussian assumption by introducing $κ\texttt{-FC}$, a general formulation that conditions the source distribution on an arbitrary signal $κ$ that aligns it better with the data distribution. Then, we present MixFlow, a simple but effective training strategy that reduces the generative path curvatures and considerably improves sampling efficiency. MixFlow trains a flow model on linear mixtures of a fixed unconditional distribution and a $κ\texttt{-FC}$-based distribution. This simple mixture improves the alignment between the source and data, provides better generation quality with less required sampling steps, and accelerates the training convergence considerably. On average, our training procedure improves the generation quality by 12\% in FID compared to standard rectified flow and 7\% compared to previous baselines under a fixed sampling budget. Code available at: $\href{https://github.com/NazirNayal8/MixFlow}{https://github.com/NazirNayal8/MixFlow}$

Xianghui Xie, Jan Eric Lenssen, Gerard Pons-Moll

Tracking human object interaction from videos is important to understand human behavior from the rapidly growing stream of video data. Previous video-based methods require predefined object templates while single-image-based methods are template-free but lack temporal consistency. In this paper, we present a method to track human object interaction without any object shape templates. We decompose the 4D tracking problem into per-frame pose tracking and canonical shape optimization. We first apply a single-view reconstruction method to obtain temporally-inconsistent per-frame interaction reconstructions. Then, for the human, we propose an efficient autoencoder to predict SMPL vertices directly from the per-frame reconstructions, introducing temporally consistent correspondence. For the object, we introduce a pose estimator that leverages temporal information to predict smooth object rotations under occlusions. To train our model, we propose a method to generate synthetic interaction videos and synthesize in total 10 hour videos of 8.5k sequences with full 3D ground truth. Experiments on BEHAVE and InterCap show that our method significantly outperforms previous template-based video tracking and single-frame reconstruction methods. Our proposed synthetic video dataset also allows training video-based methods that generalize to real-world videos. Our code and dataset will be publicly released.

Yuanwen Yue, Anurag Das, Francis Engelmann et al.

Current visual foundation models are trained purely on unstructured 2D data, limiting their understanding of 3D structure of objects and scenes. In this work, we show that fine-tuning on 3D-aware data improves the quality of emerging semantic features. We design a method to lift semantic 2D features into an efficient 3D Gaussian representation, which allows us to re-render them for arbitrary views. Using the rendered 3D-aware features, we design a fine-tuning strategy to transfer such 3D awareness into a 2D foundation model. We demonstrate that models fine-tuned in that way produce features that readily improve downstream task performance in semantic segmentation and depth estimation through simple linear probing. Notably, though fined-tuned on a single indoor dataset, the improvement is transferable to a variety of indoor datasets and out-of-domain datasets. We hope our study encourages the community to consider injecting 3D awareness when training 2D foundation models. Project page: https://ywyue.github.io/FiT3D.

Kevin Raj, Christopher Wewer, Raza Yunus et al.

We introduce Spurfies, a novel method for sparse-view surface reconstruction that disentangles appearance and geometry information to utilize local geometry priors trained on synthetic data. Recent research heavily focuses on 3D reconstruction using dense multi-view setups, typically requiring hundreds of images. However, these methods often struggle with few-view scenarios. Existing sparse-view reconstruction techniques often rely on multi-view stereo networks that need to learn joint priors for geometry and appearance from a large amount of data. In contrast, we introduce a neural point representation that disentangles geometry and appearance to train a local geometry prior using a subset of the synthetic ShapeNet dataset only. During inference, we utilize this surface prior as additional constraint for surface and appearance reconstruction from sparse input views via differentiable volume rendering, restricting the space of possible solutions. We validate the effectiveness of our method on the DTU dataset and demonstrate that it outperforms previous state of the art by 35% in surface quality while achieving competitive novel view synthesis quality. Moreover, in contrast to previous works, our method can be applied to larger, unbounded scenes, such as Mip-NeRF 360.

Valter Hudovernik, Federico López, Vid Kocijan et al.

We introduce KumoRFM-2, the next iteration of a pre-trained foundation model for relational data. KumoRFM-2 supports in-context learning as well as fine-tuning and is applicable to a wide range of predictive tasks. In contrast to tabular foundation models, KumoRFM-2 natively operates on relational data, processing one or more connected tables simultaneously without manual table flattening or target variable generation, all while preserving temporal consistency. KumoRFM-2 leverages a large corpus of synthetic and real-world data to pre-train across four axes: the row and column dimensions at the individual table level, and the foreign key and cross-sample dimensions at the database level. In contrast to its predecessor, KumoRFM-2 injects task information as early as possible, enabling sharper selection of task-relevant columns and improved robustness to noisy data. Through extensive experiments on 41 challenging benchmarks and analysis around expressivity and sensitivity, we demonstrate that KumoRFM-2 outperforms supervised and foundational approaches by up to 8%, while maintaining strong performance under extreme settings of cold start and noisy data. To our knowledge, this is the first time a few-shot foundation model has been shown to surpass supervised approaches on common benchmark tasks, with performance further improving upon fine-tuning. Finally, while KumoRFM-1 was limited to small-scale in-memory datasets, KumoRFM-2 scales to billion-scale relational datasets.

Eric Nazarenus, Chuqiao Li, Yannan He et al.

We present ActionPlan, a unified motion diffusion framework that bridges real-time streaming with high-quality offline generation within a single model. The core idea is to introduce a per-frame action plan: the model predicts frame-level text latents that act as dense semantic anchors throughout denoising, and uses them to denoise the full motion sequence with combined semantic and motion cues. To support this structured workflow, we design latent-specific diffusion steps, allowing each motion latent to be denoised independently and sampled in flexible orders at inference. As a result, ActionPlan can run in a history-conditioned, future-aware mode for real-time streaming, while also supporting high-quality offline generation. The same mechanism further enables zero-shot motion editing and in-betweening without additional models. Experiments demonstrate that our real-time streaming is 5.25x faster while also achieving 18% motion quality improvement over the best previous method in terms of FID.

Yannan He, Garvita Tiwari, Xiaohan Zhang et al.

We introduce MoLingo, a text-to-motion (T2M) model that generates realistic, lifelike human motion by denoising in a continuous latent space. Recent works perform latent space diffusion, either on the whole latent at once or auto-regressively over multiple latents. In this paper, we study how to make diffusion on continuous motion latents work best. We focus on two questions: (1) how to build a semantically aligned latent space so diffusion becomes more effective, and (2) how to best inject text conditioning so the motion follows the description closely. We propose a semantic-aligned motion encoder trained with frame-level text labels so that latents with similar text meaning stay close, which makes the latent space more diffusion-friendly. We also compare single-token conditioning with a multi-token cross-attention scheme and find that cross-attention gives better motion realism and text-motion alignment. With semantically aligned latents, auto-regressive generation, and cross-attention text conditioning, our model sets a new state of the art in human motion generation on standard metrics and in a user study. We will release our code and models for further research and downstream usage.

Vid Kocijan, Jinu Sunil, Jan Eric Lenssen et al.

The purpose of predictive modeling on relational data is to predict future or missing values in a relational database, for example, future purchases of a user, risk of readmission of the patient, or the likelihood that a financial transaction is fraudulent. Typically powered by machine learning methods, predictive models are used in recommendations, financial fraud detection, supply chain optimization, and other systems, providing billions of predictions every day. However, training a machine learning model requires manual work to extract the required training examples - prediction entities and target labels - from the database, which is slow, laborious, and prone to mistakes. Here, we present the Predictive Query Language (PQL), a SQL-inspired declarative language for defining predictive tasks on relational databases. PQL allows specifying a predictive task in a single declarative query, enabling the automatic computation training labels for a large variety of machine learning tasks, such as regression, classification, time-series forecasting, and recommender systems. PQL is already successfully integrated and used in a collection of use cases as part of a predictive AI platform. The versatility of the language can be demonstrated through its many ongoing use cases, including financial fraud, item recommendations, and workload prediction. We demonstrate its versatile design through two implementations; one for small-scale, low-latency use and one that can handle large-scale databases.

Xinya Chen, Christopher Wewer, Jiahao Xie et al.

We present SemanticNVS, a camera-conditioned multi-view diffusion model for novel view synthesis (NVS), which improves generation quality and consistency by integrating pre-trained semantic feature extractors. Existing NVS methods perform well for views near the input view, however, they tend to generate semantically implausible and distorted images under long-range camera motion, revealing severe degradation. We speculate that this degradation is due to current models failing to fully understand their conditioning or intermediate generated scene content. Here, we propose to integrate pre-trained semantic feature extractors to incorporate stronger scene semantics as conditioning to achieve high-quality generation even at distant viewpoints. We investigate two different strategies, (1) warped semantic features and (2) an alternating scheme of understanding and generation at each denoising step. Experimental results on multiple datasets demonstrate the clear qualitative and quantitative (4.69%-15.26% in FID) improvement over state-of-the-art alternatives.

Zhipeng Du, Duolikun Danier, Jan Eric Lenssen et al.

In this paper, we focus on online zero-shot monocular 3D instance segmentation, a novel practical setting where existing approaches fail to perform because they rely on posed RGB-D sequences. To overcome this limitation, we leverage CUT3R, a recent Reconstructive Foundation Model (RFM), to provide reliable geometric priors from a single RGB stream. We propose MoonSeg3R, which introduces three key components: (1) a self-supervised query refinement module with spatial-semantic distillation that transforms segmentation masks from 2D visual foundation models (VFMs) into discriminative 3D queries; (2) a 3D query index memory that provides temporal consistency by retrieving contextual queries; and (3) a state-distribution token from CUT3R that acts as a mask identity descriptor to strengthen cross-frame fusion. Experiments on ScanNet200 and SceneNN show that MoonSeg3R is the first method to enable online monocular 3D segmentation and achieves performance competitive with state-of-the-art RGB-D-based systems. Code and models will be released.

Xianghui Xie, Bharat Lal Bhatnagar, Jan Eric Lenssen et al.

Reconstructing human-object interaction in 3D from a single RGB image is a challenging task and existing data driven methods do not generalize beyond the objects present in the carefully curated 3D interaction datasets. Capturing large-scale real data to learn strong interaction and 3D shape priors is very expensive due to the combinatorial nature of human-object interactions. In this paper, we propose ProciGen (Procedural interaction Generation), a method to procedurally generate datasets with both, plausible interaction and diverse object variation. We generate 1M+ human-object interaction pairs in 3D and leverage this large-scale data to train our HDM (Hierarchical Diffusion Model), a novel method to reconstruct interacting human and unseen objects, without any templates. Our HDM is an image-conditioned diffusion model that learns both realistic interaction and highly accurate human and object shapes. Experiments show that our HDM trained with ProciGen significantly outperforms prior methods that requires template meshes and that our dataset allows training methods with strong generalization ability to unseen object instances. Our code and data are released.

Xinting Hu, Haoran Wang, Jan Eric Lenssen et al.

We introduce PersonaHOI, a training- and tuning-free framework that fuses a general StableDiffusion model with a personalized face diffusion (PFD) model to generate identity-consistent human-object interaction (HOI) images. While existing PFD models have advanced significantly, they often overemphasize facial features at the expense of full-body coherence, PersonaHOI introduces an additional StableDiffusion (SD) branch guided by HOI-oriented text inputs. By incorporating cross-attention constraints in the PFD branch and spatial merging at both latent and residual levels, PersonaHOI preserves personalized facial details while ensuring interactive non-facial regions. Experiments, validated by a novel interaction alignment metric, demonstrate the superior realism and scalability of PersonaHOI, establishing a new standard for practical personalized face with HOI generation. Our code will be available at https://github.com/JoyHuYY1412/PersonaHOI

Haiyang Wang, Yue Fan, Muhammad Ferjad Naeem et al. · pku

Transformers have become the predominant architecture in foundation models due to their excellent performance across various domains. However, the substantial cost of scaling these models remains a significant concern. This problem arises primarily from their dependence on a fixed number of parameters within linear projections. When architectural modifications (e.g., channel dimensions) are introduced, the entire model typically requires retraining from scratch. As model sizes continue growing, this strategy results in increasingly high computational costs and becomes unsustainable. To overcome this problem, we introduce TokenFormer, a natively scalable architecture that leverages the attention mechanism not only for computations among input tokens but also for interactions between tokens and model parameters, thereby enhancing architectural flexibility. By treating model parameters as tokens, we replace all the linear projections in Transformers with our token-parameter attention layer, where input tokens act as queries and model parameters as keys and values. This reformulation allows for progressive and efficient scaling without necessitating retraining from scratch. Our model scales from 124M to 1.4B parameters by incrementally adding new key-value parameter pairs, achieving performance comparable to Transformers trained from scratch while greatly reducing training costs. Code and models are available at https://github.com/Haiyang-W/TokenFormer.

Julius Erbach, Dominik Narnhofer, Andreas Dombos et al.

Flow-based latent generative models such as Stable Diffusion 3 are able to generate images with remarkable quality, even enabling photorealistic text-to-image generation. Their impressive performance suggests that these models should also constitute powerful priors for inverse imaging problems, but that approach has not yet led to comparable fidelity. There are several key obstacles: (i) the data likelihood term is usually intractable; (ii) learned generative models cannot be directly conditioned on the distorted observations, leading to conflicting objectives between data likelihood and prior; and (iii) the reconstructions can deviate from the observed data. We present FLAIR, a novel, training-free variational framework that leverages flow-based generative models as prior for inverse problems. To that end, we introduce a variational objective for flow matching that is agnostic to the type of degradation, and combine it with deterministic trajectory adjustments to guide the prior towards regions which are more likely under the posterior. To enforce exact consistency with the observed data, we decouple the optimization of the data fidelity and regularization terms. Moreover, we introduce a time-dependent calibration scheme in which the strength of the regularization is modulated according to off-line accuracy estimates. Results on standard imaging benchmarks demonstrate that FLAIR consistently outperforms existing diffusion- and flow-based methods in terms of reconstruction quality and sample diversity. Our code is available at https://inverseflair.github.io/.

Xianghui Xie, Chuhang Zou, Meher Gitika Karumuri et al.

We propose MVGBench, a comprehensive benchmark for multi-view image generation models (MVGs) that evaluates 3D consistency in geometry and texture, image quality, and semantics (using vision language models). Recently, MVGs have been the main driving force in 3D object creation. However, existing metrics compare generated images against ground truth target views, which is not suitable for generative tasks where multiple solutions exist while differing from ground truth. Furthermore, different MVGs are trained on different view angles, synthetic data and specific lightings -- robustness to these factors and generalization to real data are rarely evaluated thoroughly. Without a rigorous evaluation protocol, it is also unclear what design choices contribute to the progress of MVGs. MVGBench evaluates three different aspects: best setup performance, generalization to real data and robustness. Instead of comparing against ground truth, we introduce a novel 3D self-consistency metric which compares 3D reconstructions from disjoint generated multi-views. We systematically compare 12 existing MVGs on 4 different curated real and synthetic datasets. With our analysis, we identify important limitations of existing methods specially in terms of robustness and generalization, and we find the most critical design choices. Using the discovered best practices, we propose ViFiGen, a method that outperforms all evaluated MVGs on 3D consistency. Our code, model, and benchmark suite will be publicly released.

Christopher Wewer, Kevin Raj, Eddy Ilg et al.

We present latentSplat, a method to predict semantic Gaussians in a 3D latent space that can be splatted and decoded by a light-weight generative 2D architecture. Existing methods for generalizable 3D reconstruction either do not scale to large scenes and resolutions, or are limited to interpolation of close input views. latentSplat combines the strengths of regression-based and generative approaches while being trained purely on readily available real video data. The core of our method are variational 3D Gaussians, a representation that efficiently encodes varying uncertainty within a latent space consisting of 3D feature Gaussians. From these Gaussians, specific instances can be sampled and rendered via efficient splatting and a fast, generative decoder. We show that latentSplat outperforms previous works in reconstruction quality and generalization, while being fast and scalable to high-resolution data.

Matthias Fey, Weihua Hu, Kexin Huang et al.

Much of the world's most valued data is stored in relational databases and data warehouses, where the data is organized into many tables connected by primary-foreign key relations. However, building machine learning models using this data is both challenging and time consuming. The core problem is that no machine learning method is capable of learning on multiple tables interconnected by primary-foreign key relations. Current methods can only learn from a single table, so the data must first be manually joined and aggregated into a single training table, the process known as feature engineering. Feature engineering is slow, error prone and leads to suboptimal models. Here we introduce an end-to-end deep representation learning approach to directly learn on data laid out across multiple tables. We name our approach Relational Deep Learning (RDL). The core idea is to view relational databases as a temporal, heterogeneous graph, with a node for each row in each table, and edges specified by primary-foreign key links. Message Passing Graph Neural Networks can then automatically learn across the graph to extract representations that leverage all input data, without any manual feature engineering. Relational Deep Learning leads to more accurate models that can be built much faster. To facilitate research in this area, we develop RelBench, a set of benchmark datasets and an implementation of Relational Deep Learning. The data covers a wide spectrum, from discussions on Stack Exchange to book reviews on the Amazon Product Catalog. Overall, we define a new research area that generalizes graph machine learning and broadens its applicability to a wide set of AI use cases.

Jialin Chen, Jan Eric Lenssen, Aosong Feng et al.

Time series forecasting has attracted significant attention in recent decades. Previous studies have demonstrated that the Channel-Independent (CI) strategy improves forecasting performance by treating different channels individually, while it leads to poor generalization on unseen instances and ignores potentially necessary interactions between channels. Conversely, the Channel-Dependent (CD) strategy mixes all channels with even irrelevant and indiscriminate information, which, however, results in oversmoothing issues and limits forecasting accuracy. There is a lack of channel strategy that effectively balances individual channel treatment for improved forecasting performance without overlooking essential interactions between channels. Motivated by our observation of a correlation between the time series model's performance boost against channel mixing and the intrinsic similarity on a pair of channels, we developed a novel and adaptable Channel Clustering Module (CCM). CCM dynamically groups channels characterized by intrinsic similarities and leverages cluster information instead of individual channel identities, combining the best of CD and CI worlds. Extensive experiments on real-world datasets demonstrate that CCM can (1) boost the performance of CI and CD models by an average margin of 2.4% and 7.2% on long-term and short-term forecasting, respectively; (2) enable zero-shot forecasting with mainstream time series forecasting models; (3) uncover intrinsic time series patterns among channels and improve interpretability of complex time series models.

Raza Yunus, Jan Eric Lenssen, Michael Niemeyer et al.

Reconstructing models of the real world, including 3D geometry, appearance, and motion of real scenes, is essential for computer graphics and computer vision. It enables the synthesizing of photorealistic novel views, useful for the movie industry and AR/VR applications. It also facilitates the content creation necessary in computer games and AR/VR by avoiding laborious manual design processes. Further, such models are fundamental for intelligent computing systems that need to interpret real-world scenes and actions to act and interact safely with the human world. Notably, the world surrounding us is dynamic, and reconstructing models of dynamic, non-rigidly moving scenes is a severely underconstrained and challenging problem. This state-of-the-art report (STAR) offers the reader a comprehensive summary of state-of-the-art techniques with monocular and multi-view inputs such as data from RGB and RGB-D sensors, among others, conveying an understanding of different approaches, their potential applications, and promising further research directions. The report covers 3D reconstruction of general non-rigid scenes and further addresses the techniques for scene decomposition, editing and controlling, and generalizable and generative modeling. More specifically, we first review the common and fundamental concepts necessary to understand and navigate the field and then discuss the state-of-the-art techniques by reviewing recent approaches that use traditional and machine-learning-based neural representations, including a discussion on the newly enabled applications. The STAR is concluded with a discussion of the remaining limitations and open challenges.

Mohammad Asim, Christopher Wewer, Thomas Wimmer et al. · eth-zurich

We introduce MEt3R, a metric for multi-view consistency in generated images. Large-scale generative models for multi-view image generation are rapidly advancing the field of 3D inference from sparse observations. However, due to the nature of generative modeling, traditional reconstruction metrics are not suitable to measure the quality of generated outputs and metrics that are independent of the sampling procedure are desperately needed. In this work, we specifically address the aspect of consistency between generated multi-view images, which can be evaluated independently of the specific scene. Our approach uses DUSt3R to obtain dense 3D reconstructions from image pairs in a feed-forward manner, which are used to warp image contents from one view into the other. Then, feature maps of these images are compared to obtain a similarity score that is invariant to view-dependent effects. Using MEt3R, we evaluate the consistency of a large set of previous methods for novel view and video generation, including our open, multi-view latent diffusion model.

Keyang Zhou, Bharat Lal Bhatnagar, Jan Eric Lenssen et al.

Generating realistic hand motion sequences in interaction with objects has gained increasing attention with the growing interest in digital humans. Prior work has illustrated the effectiveness of employing occupancy-based or distance-based virtual sensors to extract hand-object interaction features. Nonetheless, these methods show limited generalizability across object categories, shapes and sizes. We hypothesize that this is due to two reasons: 1) the limited expressiveness of employed virtual sensors, and 2) scarcity of available training data. To tackle this challenge, we introduce a novel joint-centered sensor designed to reason about local object geometry near potential interaction regions. The sensor queries for object surface points in the neighbourhood of each hand joint. As an important step towards mitigating the learning complexity, we transform the points from global frame to hand template frame and use a shared module to process sensor features of each individual joint. This is followed by a spatio-temporal transformer network aimed at capturing correlation among the joints in different dimensions. Moreover, we devise simple heuristic rules to augment the limited training sequences with vast static hand grasping samples. This leads to a broader spectrum of grasping types observed during training, in turn enhancing our model's generalization capability. We evaluate on two public datasets, GRAB and InterCap, where our method shows superiority over baselines both quantitatively and perceptually.

Yannan He, Garvita Tiwari, Tolga Birdal et al.

Faithfully modeling the space of articulations is a crucial task that allows recovery and generation of realistic poses, and remains a notorious challenge. To this end, we introduce Neural Riemannian Distance Fields (NRDFs), data-driven priors modeling the space of plausible articulations, represented as the zero-level-set of a neural field in a high-dimensional product-quaternion space. To train NRDFs only on positive examples, we introduce a new sampling algorithm, ensuring that the geodesic distances follow a desired distribution, yielding a principled distance field learning paradigm. We then devise a projection algorithm to map any random pose onto the level-set by an adaptive-step Riemannian optimizer, adhering to the product manifold of joint rotations at all times. NRDFs can compute the Riemannian gradient via backpropagation and by mathematical analogy, are related to Riemannian flow matching, a recent generative model. We conduct a comprehensive evaluation of NRDF against other pose priors in various downstream tasks, i.e., pose generation, image-based pose estimation, and solving inverse kinematics, highlighting NRDF's superior performance. Besides humans, NRDF's versatility extends to hand and animal poses, as it can effectively represent any articulation.

Matthias Fey, Jinu Sunil, Akihiro Nitta et al.

PyG (PyTorch Geometric) has evolved significantly since its initial release, establishing itself as a leading framework for Graph Neural Networks. In this paper, we present Pyg 2.0 (and its subsequent minor versions), a comprehensive update that introduces substantial improvements in scalability and real-world application capabilities. We detail the framework's enhanced architecture, including support for heterogeneous and temporal graphs, scalable feature/graph stores, and various optimizations, enabling researchers and practitioners to tackle large-scale graph learning problems efficiently. Over the recent years, PyG has been supporting graph learning in a large variety of application areas, which we will summarize, while providing a deep dive into the important areas of relational deep learning and large language modeling.

Yiwen Yuan, Zecheng Zhang, Xinwei He et al.

Recommendation systems predominantly utilize two-tower architectures, which evaluate user-item rankings through the inner product of their respective embeddings. However, one key limitation of two-tower models is that they learn a pair-agnostic representation of users and items. In contrast, pair-wise representations either scale poorly due to their quadratic complexity or are too restrictive on the candidate pairs to rank. To address these issues, we introduce Context-based Graph Neural Networks (ContextGNNs), a novel deep learning architecture for link prediction in recommendation systems. The method employs a pair-wise representation technique for familiar items situated within a user's local subgraph, while leveraging two-tower representations to facilitate the recommendation of exploratory items. A final network then predicts how to fuse both pair-wise and two-tower recommendations into a single ranking of items. We demonstrate that ContextGNN is able to adapt to different data characteristics and outperforms existing methods, both traditional and GNN-based, on a diverse set of practical recommendation tasks, improving performance by 20% on average.

Philipp Schröppel, Christopher Wewer, Jan Eric Lenssen et al.

Controllable generation of 3D assets is important for many practical applications like content creation in movies, games and engineering, as well as in AR/VR. Recently, diffusion models have shown remarkable results in generation quality of 3D objects. However, none of the existing models enable disentangled generation to control the shape and appearance separately. For the first time, we present a suitable representation for 3D diffusion models to enable such disentanglement by introducing a hybrid point cloud and neural radiance field approach. We model a diffusion process over point positions jointly with a high-dimensional feature space for a local density and radiance decoder. While the point positions represent the coarse shape of the object, the point features allow modeling the geometry and appearance details. This disentanglement enables us to sample both independently and therefore to control both separately. Our approach sets a new state of the art in generation compared to previous disentanglement-capable methods by reduced FID scores of 30-90% and is on-par with other non disentanglement-capable state-of-the art methods.

Christopher Wewer, Bart Pogodzinski, Bernt Schiele et al.

We introduce Spatial Reasoning Models (SRMs), a framework to perform reasoning over sets of continuous variables via denoising generative models. SRMs infer continuous representations on a set of unobserved variables, given observations on observed variables. Current generative models on spatial domains, such as diffusion and flow matching models, often collapse to hallucination in case of complex distributions. To measure this, we introduce a set of benchmark tasks that test the quality of complex reasoning in generative models and can quantify hallucination. The SRM framework allows to report key findings about importance of sequentialization in generation, the associated order, as well as the sampling strategies during training. It demonstrates, for the first time, that order of generation can successfully be predicted by the denoising network itself. Using these findings, we can increase the accuracy of specific reasoning tasks from <1% to >50%. Our project website provides additional videos, code, and the benchmark datasets: https://geometric-rl.mpi-inf.mpg.de/srm

Thomas Wimmer, Prune Truong, Marie-Julie Rakotosaona et al. · eth-zurich

We introduce AnyUp, a method for feature upsampling that can be applied to any vision feature at any resolution, without encoder-specific training. Existing learning-based upsamplers for features like DINO or CLIP need to be re-trained for every feature extractor and thus do not generalize to different feature types at inference time. In this work, we propose an inference-time feature-agnostic upsampling architecture to alleviate this limitation and improve upsampling quality. In our experiments, AnyUp sets a new state of the art for upsampled features, generalizes to different feature types, and preserves feature semantics while being efficient and easy to apply to a wide range of downstream tasks.

Mohammad Asim, Christopher Wewer, Jan Eric Lenssen

We present SceneTok, a novel tokenizer for encoding view sets of scenes into a compressed and diffusable set of unstructured tokens. Existing approaches for 3D scene representation and generation commonly use 3D data structures or view-aligned fields. In contrast, we introduce the first method that encodes scene information into a small set of permutation-invariant tokens that is disentangled from the spatial grid. The scene tokens are predicted by a multi-view tokenizer given many context views and rendered into novel views by employing a light-weight rectified flow decoder. We show that the compression is 1-3 orders of magnitude stronger than for other representations while still reaching state-of-the-art reconstruction quality. Further, our representation can be rendered from novel trajectories, including ones deviating from the input trajectory, and we show that the decoder gracefully handles uncertainty. Finally, the highly-compressed set of unstructured latent scene tokens enables simple and efficient scene generation in 5 seconds, achieving a much better quality-speed trade-off than previous paradigms.

Bart Pogodzinski, Christopher Wewer, Bernt Schiele et al.

We present Spatial Reasoners, a software framework to perform spatial reasoning over continuous variables with generative denoising models. Denoising generative models have become the de-facto standard for image generation, due to their effectiveness in sampling from complex, high-dimensional distributions. Recently, they have started being explored in the context of reasoning over multiple continuous variables. Providing infrastructure for generative reasoning with such models requires a high effort, due to a wide range of different denoising formulations, samplers, and inference strategies. Our presented framework aims to facilitate research in this area, providing easy-to-use interfaces to control variable mapping from arbitrary data domains, generative model paradigms, and inference strategies. Spatial Reasoners are openly available at https://spatialreasoners.github.io/

Cameron Braunstein, Hevra Petekkaya, Jan Eric Lenssen et al.

We introduce SLayR, Scene Layout Generation with Rectified flow, a novel transformer-based model for text-to-layout generation which can then be paired with existing layout-to-image models to produce images. SLayR addresses a domain in which current text-to-image pipelines struggle: generating scene layouts that are of significant variety and plausibility, when the given prompt is ambiguous and does not provide constraints on the scene. SLayR surpasses existing baselines including LLMs in unconstrained generation, and can generate layouts from an open caption set. To accurately evaluate the layout generation, we introduce a new benchmark suite, including numerical metrics and a carefully designed repeatable human-evaluation procedure that assesses the plausibility and variety of generated images. We show that our method sets a new state of the art for achieving both at the same time, while being at least 3x times smaller in the number of parameters.

Jonas Ernst, Wolfgang Boettcher, Lukas Hoyer et al.

We present Rewis3d, a framework that leverages recent advances in feed-forward 3D reconstruction to significantly improve weakly supervised semantic segmentation on 2D images. Obtaining dense, pixel-level annotations remains a costly bottleneck for training segmentation models. Alleviating this issue, sparse annotations offer an efficient weakly-supervised alternative. However, they still incur a performance gap. To address this, we introduce a novel approach that leverages 3D scene reconstruction as an auxiliary supervisory signal. Our key insight is that 3D geometric structure recovered from 2D videos provides strong cues that can propagate sparse annotations across entire scenes. Specifically, a dual student-teacher architecture enforces semantic consistency between 2D images and reconstructed 3D point clouds, using state-of-the-art feed-forward reconstruction to generate reliable geometric supervision. Extensive experiments demonstrate that Rewis3d achieves state-of-the-art performance in sparse supervision, outperforming existing approaches by 2-7% without requiring additional labels or inference overhead.

Anna Kukleva, Enis Simsar, Alessio Tonioni et al.

Most existing approaches to referring segmentation achieve strong performance only through fine-tuning or by composing multiple pre-trained models, often at the cost of additional training and architectural modifications. Meanwhile, large-scale generative diffusion models encode rich semantic information, making them attractive as general-purpose feature extractors. In this work, we introduce a new method that directly exploits features, attention scores, from diffusion transformers for downstream tasks, requiring neither architectural modifications nor additional training. To systematically evaluate these features, we extend benchmarks with vision-language grounding tasks spanning both images and videos. Our key insight is that stop words act as attention magnets: they accumulate surplus attention and can be filtered to reduce noise. Moreover, we identify global attention sinks (GAS) emerging in deeper layers and show that they can be safely suppressed or redirected onto auxiliary tokens, leading to sharper and more accurate grounding maps. We further propose an attention redistribution strategy, where appended stop words partition background activations into smaller clusters, yielding sharper and more localized heatmaps. Building on these findings, we develop RefAM, a simple training-free grounding framework that combines cross-attention maps, GAS handling, and redistribution. Across zero-shot referring image and video segmentation benchmarks, our approach consistently outperforms prior methods, establishing a new state of the art without fine-tuning or additional components.

Rohan Chabra, Jan Eric Lenssen, Eddy Ilg et al.

Efficiently reconstructing complex and intricate surfaces at scale is a long-standing goal in machine perception. To address this problem we introduce Deep Local Shapes (DeepLS), a deep shape representation that enables encoding and reconstruction of high-quality 3D shapes without prohibitive memory requirements. DeepLS replaces the dense volumetric signed distance function (SDF) representation used in traditional surface reconstruction systems with a set of locally learned continuous SDFs defined by a neural network, inspired by recent work such as DeepSDF. Unlike DeepSDF, which represents an object-level SDF with a neural network and a single latent code, we store a grid of independent latent codes, each responsible for storing information about surfaces in a small local neighborhood. This decomposition of scenes into local shapes simplifies the prior distribution that the network must learn, and also enables efficient inference. We demonstrate the effectiveness and generalization power of DeepLS by showing object shape encoding and reconstructions of full scenes, where DeepLS delivers high compression, accuracy, and local shape completion.

Yongheng Zhao, Tolga Birdal, Jan Eric Lenssen et al.

We present a 3D capsule module for processing point clouds that is equivariant to 3D rotations and translations, as well as invariant to permutations of the input points. The operator receives a sparse set of local reference frames, computed from an input point cloud and establishes end-to-end transformation equivariance through a novel dynamic routing procedure on quaternions. Further, we theoretically connect dynamic routing between capsules to the well-known Weiszfeld algorithm, a scheme for solving \emph{iterative re-weighted least squares} (IRLS) problems with provable convergence properties. It is shown that such group dynamic routing can be interpreted as robust IRLS rotation averaging on capsule votes, where information is routed based on the final inlier scores. Based on our operator, we build a capsule network that disentangles geometry from pose, paving the way for more informative descriptors and a structured latent space. Our architecture allows joint object classification and orientation estimation without explicit supervision of rotations. We validate our algorithm empirically on common benchmark datasets.

Jan Eric Lenssen, Christian Osendorfer, Jonathan Masci

This paper presents an end-to-end differentiable algorithm for robust and detail-preserving surface normal estimation on unstructured point-clouds. We utilize graph neural networks to iteratively parameterize an adaptive anisotropic kernel that produces point weights for weighted least-squares plane fitting in local neighborhoods. The approach retains the interpretability and efficiency of traditional sequential plane fitting while benefiting from adaptation to data set statistics through deep learning. This results in a state-of-the-art surface normal estimator that is robust to noise, outliers and point density variation, preserves sharp features through anisotropic kernels and equivariance through a local quaternion-based spatial transformer. Contrary to previous deep learning methods, the proposed approach does not require any hand-crafted features or preprocessing. It improves on the state-of-the-art results while being more than two orders of magnitude faster and more parameter efficient.

Matthias Fey, Jan Eric Lenssen

We introduce PyTorch Geometric, a library for deep learning on irregularly structured input data such as graphs, point clouds and manifolds, built upon PyTorch. In addition to general graph data structures and processing methods, it contains a variety of recently published methods from the domains of relational learning and 3D data processing. PyTorch Geometric achieves high data throughput by leveraging sparse GPU acceleration, by providing dedicated CUDA kernels and by introducing efficient mini-batch handling for input examples of different size. In this work, we present the library in detail and perform a comprehensive comparative study of the implemented methods in homogeneous evaluation scenarios.

Christopher Morris, Martin Ritzert, Matthias Fey et al.

In recent years, graph neural networks (GNNs) have emerged as a powerful neural architecture to learn vector representations of nodes and graphs in a supervised, end-to-end fashion. Up to now, GNNs have only been evaluated empirically -- showing promising results. The following work investigates GNNs from a theoretical point of view and relates them to the $1$-dimensional Weisfeiler-Leman graph isomorphism heuristic ($1$-WL). We show that GNNs have the same expressiveness as the $1$-WL in terms of distinguishing non-isomorphic (sub-)graphs. Hence, both algorithms also have the same shortcomings. Based on this, we propose a generalization of GNNs, so-called $k$-dimensional GNNs ($k$-GNNs), which can take higher-order graph structures at multiple scales into account. These higher-order structures play an essential role in the characterization of social networks and molecule graphs. Our experimental evaluation confirms our theoretical findings as well as confirms that higher-order information is useful in the task of graph classification and regression.

Jan Eric Lenssen, Matthias Fey, Pascal Libuschewski

We present group equivariant capsule networks, a framework to introduce guaranteed equivariance and invariance properties to the capsule network idea. Our work can be divided into two contributions. First, we present a generic routing by agreement algorithm defined on elements of a group and prove that equivariance of output pose vectors, as well as invariance of output activations, hold under certain conditions. Second, we connect the resulting equivariant capsule networks with work from the field of group convolutional networks. Through this connection, we provide intuitions of how both methods relate and are able to combine the strengths of both approaches in one deep neural network architecture. The resulting framework allows sparse evaluation of the group convolution operator, provides control over specific equivariance and invariance properties, and can use routing by agreement instead of pooling operations. In addition, it is able to provide interpretable and equivariant representation vectors as output capsules, which disentangle evidence of object existence from its pose.

Matthias Fey, Jan Eric Lenssen, Frank Weichert et al.

We present Spline-based Convolutional Neural Networks (SplineCNNs), a variant of deep neural networks for irregular structured and geometric input, e.g., graphs or meshes. Our main contribution is a novel convolution operator based on B-splines, that makes the computation time independent from the kernel size due to the local support property of the B-spline basis functions. As a result, we obtain a generalization of the traditional CNN convolution operator by using continuous kernel functions parametrized by a fixed number of trainable weights. In contrast to related approaches that filter in the spectral domain, the proposed method aggregates features purely in the spatial domain. In addition, SplineCNN allows entire end-to-end training of deep architectures, using only the geometric structure as input, instead of handcrafted feature descriptors. For validation, we apply our method on tasks from the fields of image graph classification, shape correspondence and graph node classification, and show that it outperforms or pars state-of-the-art approaches while being significantly faster and having favorable properties like domain-independence.