arXiv subjects

Stefan Andreas Baumann

Publications and source records attributed to Stefan Andreas Baumann.

17 recordsLinked to original sources

arXiv2026

Improved Distributional Diffusion Models

Distributional Diffusion Models (DDMs) replace the standard mean-prediction denoiser with a \emph{distributional} denoiser trained via a scoring rule objective, learning a stochastic approximation to $p(x_1 \mid x_t)$ rather than its conditional mean. However, scaling DDMs to modern image-generation settings faces two obstacles: (i) multi-particle training incurs overhead that scales with the number of particles, (ii) DDMs use globally fixed scoring rule hyperparameters, forcing a single trade-off across sampling budgets. We mitigate these limitations by deferring particle expansion to late transformer layers, and the hyperparameter trade-off by introducing time-dependent scoring rule schedules informed by the dynamical regimes of~\citet{Biroli2024}. Combined with a DiT-based latent setup, these changes make DDM training practical on class-conditional ImageNet-$256^2$, achieving 4.48 FID at 4 steps and 2.38 at 50 steps with DiT-XL/2, from a single model trained from scratch in one stage, without a teacher, self-distillation or JVPs. The result is a stochastic few-step generator whose FID does not degrade as the sampling budget grows from 4 to 50 NFE, and the same recipe transfers to text-to-image generation. Code and pre-trained models available at https://github.com/CompVis/iDDM.

cs.CV↗

arXiv2026

What Moves? Localized Motion Representations for Compositional Scene Control

Real-world dynamics are inherently compositional: multiple entities move simultaneously within a shared scene, each exhibiting distinct motion patterns. Yet current motion representation models entangle the dynamics of different entities, without explicitly capturing localized motion for each individually. Crucially, motion is defined relative to a global reference frame, including camera motion and scene layout. However, localized embeddings are often computed from cropped images or obtained by masking features after encoding, discarding the context needed to interpret motion. To address this, we introduce a promptable localized motion representation that produces persistent embeddings for user-specified regions defined by spatial masks. Rather than cropping the input or masking features, our model processes the full video and conditions motion encoding directly on the queried region. This yields temporally consistent, region-addressable embeddings that isolate local dynamics while retaining the global context required for disambiguation. We demonstrate object-level motion transfer, enabling controlled composition of dynamic scenes. Beyond generative control, our embeddings support localized action classification in multi-actor videos. Across both tasks, our approach improves controllability and outperforms global representations localized through cropping or post-hoc masking.

cs.CV↗

arXiv2026

Logit Refiner: Improving Visual Autoregressive Models via Intra-Scale Dependency Modeling

Visual Autoregressive Models (VAR) generate images through next-scale prediction, producing all tokens within each scale in parallel. We show that this parallel decoding constitutes a mean-field-style approximation that discards spatial dependencies among same-scale tokens, causing locally incoherent samples regardless of backbone capacity -- a limitation of the decoding rule. Addressing this limitation, we introduce the Logit Refiner, a lightweight autoregressive module that restores intra-scale dependencies by sequentially sampling tokens conditioned on frozen backbone features. Adding only ~10% parameters and less than 5% of the base model's training compute, it plugs into any pretrained VAR checkpoint without retraining. Controlled ablations isolate joint intra-scale sampling -- rather than additional capacity or training -- as the critical ingredient. Across backbones from 310M to 2B parameters on class-conditional ImageNet 256x256, the refiner consistently improves generation quality, enabling a 1.1B-parameter model to surpass one twice its size. The approach further generalizes to text-to-image generation, confirming that the mean-field bottleneck persists across VAR variants and is effectively alleviated by our method. Project page: https://compvis.github.io/logit-refiner/

cs.CV↗

arXiv2026

Show Me Examples: Inferring Visual Concepts from Image Sets

Vision-language models (VLMs) can follow complex textual instructions, yet they struggle to reason from purely visual context. In particular, current models fail to infer shared concepts from sets of example images and apply them to new inputs. We introduce Visual Concept Inference from Sets (VICIS), a task that evaluates this capability. Given a small context set of images sharing a concept and a query image, the model must generate new images that preserve the context-defined concept while remaining consistent with the query. We show that state-of-the-art VLMs perform poorly on this task, often ignoring the visual context or defaulting to biased generations. To address this gap, we propose a training framework and architecture that learn to infer visual concepts from image sets and extract concept-specific embeddings from queries. Experiments on synthetic data and large-scale ImageNet/WordNet data show that our model generates more accurate and diverse outputs and generalizes to unseen concepts and modalities such as sketches.

cs.CV↗

arXiv2026

RayDer: Scalable Self-Supervised Novel View Synthesis from Real-World Video

Self-supervised novel view synthesis (NVS) remains challenging to scale, despite the abundance of video data, largely due to the brittleness of training on realistic videos and the hard-to-predict scaling behavior of multi-network system designs. We introduce RayDer, a unified, feed-forward transformer that consolidates camera estimation, scene reconstruction, and rendering into a single backbone, turning self-supervised NVS into a well-posed single-model scaling problem. A minimal dynamic state, treated as a nuisance factor, absorbs time-varying content and enables stable training on unconstrained real-world video. Importantly, RayDer keeps static-scene NVS as its target task: dynamic content is leveraged purely as scalable supervision, not reconstructed as in dynamic-scene (4D) NVS. Across multiple model sizes and orders of magnitude in data, RayDer exhibits clean power-law scaling with data and compute, and outperforms static-scene data mixtures. On a large number of benchmarks, RayDer achieves strong zero-shot open-set performance competitive with state-of-the-art supervised approaches. Project Page: https://compvis.github.io/rayder

cs.CV↗

arXiv2026

Guiding Token-Sparse Diffusion Models

Diffusion models deliver high quality in image synthesis but remain expensive during training and inference. Recent works have leveraged the inherent redundancy in visual content to make training more affordable by training only on a subset of visual information. While these methods were successful in providing cheaper and more effective training, sparsely trained diffusion models struggle in inference. This is due to their lacking response to Classifier-free Guidance (CFG) leading to underwhelming performance during inference. To overcome this, we propose Sparse Guidance (SG). Instead of using conditional dropout as a signal to guide diffusion models, SG uses token-level sparsity. As a result, SG preserves the high-variance of the conditional prediction better, achieving good quality and high variance outputs. Leveraging token-level sparsity at inference, SG improves fidelity at lower compute, achieving 1.58 FID on the commonly used ImageNet-256 benchmark with 25% fewer FLOPs, and yields up to 58% FLOP savings at matched baseline quality. To demonstrate the effectiveness of Sparse Guidance, we train a 2.5B text-to-image diffusion model using training time sparsity and leverage SG during inference. SG achieves improvements in composition and human preference score while increasing throughput at the same time.

cs.CV↗

arXiv2026

Learning Long-term Motion Embeddings for Efficient Kinematics Generation

Understanding and predicting motion is a fundamental component of visual intelligence. Although modern video models exhibit strong comprehension of scene dynamics, exploring multiple possible futures through full video synthesis remains prohibitively inefficient. We model scene dynamics orders of magnitude more efficiently by directly operating on a long-term motion embedding that is learned from large-scale trajectories obtained from tracker models. This enables efficient generation of long, realistic motions that fulfill goals specified via text prompts or spatial pokes. To achieve this, we first learn a highly compressed motion embedding with a temporal compression factor of 64x. In this space, we train a conditional flow-matching model to generate motion latents conditioned on task descriptions. The resulting motion distributions outperform those of both state-of-the-art video models and specialized task-specific approaches.

cs.CV↗

arXiv2026

Envisioning the Future, One Step at a Time

Accurately anticipating how complex, diverse scenes will evolve requires models that represent uncertainty, simulate along extended interaction chains, and efficiently explore many plausible futures. Yet most existing approaches rely on dense video or latent-space prediction, expending substantial capacity on dense appearance rather than on the underlying sparse trajectories of points in the scene. This makes large-scale exploration of future hypotheses costly and limits performance when long-horizon, multi-modal motion is essential. We address this by formulating the prediction of open-set future scene dynamics as step-wise inference over sparse point trajectories. Our autoregressive diffusion model advances these trajectories through short, locally predictable transitions, explicitly modeling the growth of uncertainty over time. This dynamics-centric representation enables fast rollout of thousands of diverse futures from a single image, optionally guided by initial constraints on motion, while maintaining physical plausibility and long-range coherence. We further introduce OWM, a benchmark for open-set motion prediction based on diverse in-the-wild videos, to evaluate accuracy and variability of predicted trajectory distributions under real-world uncertainty. Our method matches or surpasses dense simulators in predictive accuracy while achieving orders-of-magnitude higher sampling speed, making open-set future prediction both scalable and practical. Project page: http://compvis.github.io/myriad.

cs.CV↗

arXiv2026

Scalable High-Resolution Pixel-Space Image Synthesis with Hourglass Diffusion Transformers

We present the Hourglass Diffusion Transformer (HDiT), an image generative model that exhibits linear scaling with pixel count, supporting training at high-resolution (e.g. $1024 \times 1024$) directly in pixel-space. Building on the Transformer architecture, which is known to scale to billions of parameters, it bridges the gap between the efficiency of convolutional U-Nets and the scalability of Transformers. HDiT trains successfully without typical high-resolution training techniques such as multiscale architectures, latent autoencoders or self-conditioning. We demonstrate that HDiT performs competitively with existing models on ImageNet $256^2$, and sets a new state-of-the-art for diffusion models on FFHQ-$1024^2$.

cs.CV↗

arXiv2025

CleanDIFT: Diffusion Features without Noise

Internal features from large-scale pre-trained diffusion models have recently been established as powerful semantic descriptors for a wide range of downstream tasks. Works that use these features generally need to add noise to images before passing them through the model to obtain the semantic features, as the models do not offer the most useful features when given images with little to no noise. We show that this noise has a critical impact on the usefulness of these features that cannot be remedied by ensembling with different random noises. We address this issue by introducing a lightweight, unsupervised fine-tuning method that enables diffusion backbones to provide high-quality, noise-free semantic features. We show that these features readily outperform previous diffusion features by a wide margin in a wide variety of extraction setups and downstream tasks, offering better performance than even ensemble-based methods at a fraction of the cost.

cs.CV↗

arXiv2025

DisMo: Disentangled Motion Representations for Open-World Motion Transfer

Recent advances in text-to-video (T2V) and image-to-video (I2V) models, have enabled the creation of visually compelling and dynamic videos from simple textual descriptions or initial frames. However, these models often fail to provide an explicit representation of motion separate from content, limiting their applicability for content creators. To address this gap, we propose DisMo, a novel paradigm for learning abstract motion representations directly from raw video data via an image-space reconstruction objective. Our representation is generic and independent of static information such as appearance, object identity, or pose. This enables open-world motion transfer, allowing motion to be transferred across semantically unrelated entities without requiring object correspondences, even between vastly different categories. Unlike prior methods, which trade off motion fidelity and prompt adherence, are overfitting to source structure or drifting from the described action, our approach disentangles motion semantics from appearance, enabling accurate transfer and faithful conditioning. Furthermore, our motion representation can be combined with any existing video generator via lightweight adapters, allowing us to effortlessly benefit from future advancements in video models. We demonstrate the effectiveness of our method through a diverse set of motion transfer tasks. Finally, we show that the learned representations are well-suited for downstream motion understanding tasks, consistently outperforming state-of-the-art video representation models such as V-JEPA in zero-shot action classification on benchmarks including Something-Something v2 and Jester. Project page: https://compvis.github.io/DisMo

cs.CV↗

arXiv2025

What If : Understanding Motion Through Sparse Interactions

Understanding the dynamics of a physical scene involves reasoning about the diverse ways it can potentially change, especially as a result of local interactions. We present the Flow Poke Transformer (FPT), a novel framework for directly predicting the distribution of local motion, conditioned on sparse interactions termed "pokes". Unlike traditional methods that typically only enable dense sampling of a single realization of scene dynamics, FPT provides an interpretable directly accessible representation of multi-modal scene motion, its dependency on physical interactions and the inherent uncertainties of scene dynamics. We also evaluate our model on several downstream tasks to enable comparisons with prior methods and highlight the flexibility of our approach. On dense face motion generation, our generic pre-trained model surpasses specialized baselines. FPT can be fine-tuned in strongly out-of-distribution tasks such as synthetic datasets to enable significant improvements over in-domain methods in articulated object motion estimation. Additionally, predicting explicit motion distributions directly enables our method to achieve competitive performance on tasks like moving part segmentation from pokes which further demonstrates the versatility of our FPT. Code and models are publicly available at https://compvis.github.io/flow-poke-transformer.

cs.CV↗

arXiv2025

TREAD: Token Routing for Efficient Architecture-agnostic Diffusion Training

Diffusion models have emerged as the mainstream approach for visual generation. However, these models typically suffer from sample inefficiency and high training costs. Consequently, methods for efficient finetuning, inference and personalization were quickly adopted by the community. However, training these models in the first place remains very costly. While several recent approaches - including masking, distillation, and architectural modifications - have been proposed to improve training efficiency, each of these methods comes with a tradeoff: they achieve enhanced performance at the expense of increased computational cost or vice versa. In contrast, this work aims to improve training efficiency as well as generative performance at the same time through routes that act as a transport mechanism for randomly selected tokens from early layers to deeper layers of the model. Our method is not limited to the common transformer-based model - it can also be applied to state-space models and achieves this without architectural modifications or additional parameters. Finally, we show that TREAD reduces computational cost and simultaneously boosts model performance on the standard ImageNet-256 benchmark in class-conditional synthesis. Both of these benefits multiply to a convergence speedup of 14x at 400K training iterations compared to DiT and 37x compared to the best benchmark performance of DiT at 7M training iterations. Furthermore, we achieve a competitive FID of 2.09 in a guided and 3.93 in an unguided setting, which improves upon the DiT, without architectural changes.

cs.CV↗

arXiv2025

Continuous, Subject-Specific Attribute Control in T2I Models by Identifying Semantic Directions

Recent advances in text-to-image (T2I) diffusion models have significantly improved the quality of generated images. However, providing efficient control over individual subjects, particularly the attributes characterizing them, remains a key challenge. While existing methods have introduced mechanisms to modulate attribute expression, they typically provide either detailed, object-specific localization of such a modification or full-scale fine-grained, nuanced control of attributes. No current approach offers both simultaneously, resulting in a gap when trying to achieve precise continuous and subject-specific attribute modulation in image generation. In this work, we demonstrate that token-level directions exist within commonly used CLIP text embeddings that enable fine-grained, subject-specific control of high-level attributes in T2I models. We introduce two methods to identify these directions: a simple, optimization-free technique and a learning-based approach that utilizes the T2I model to characterize semantic concepts more specifically. Our methods allow the augmentation of the prompt text input, enabling fine-grained control over multiple attributes of individual subjects simultaneously, without requiring any modifications to the diffusion model itself. This approach offers a unified solution that fills the gap between global and localized control, providing competitive flexibility and precision in text-guided image generation. Project page: https://compvis.github.io/attribute-control. Code is available at https://github.com/CompVis/attribute-control.

cs.CV↗

arXiv2024

DepthFM: Fast Monocular Depth Estimation with Flow Matching

Current discriminative depth estimation methods often produce blurry artifacts, while generative approaches suffer from slow sampling due to curvatures in the noise-to-depth transport. Our method addresses these challenges by framing depth estimation as a direct transport between image and depth distributions. We are the first to explore flow matching in this field, and we demonstrate that its interpolation trajectories enhance both training and sampling efficiency while preserving high performance. While generative models typically require extensive training data, we mitigate this dependency by integrating external knowledge from a pre-trained image diffusion model, enabling effective transfer even across differing objectives. To further boost our model performance, we employ synthetic data and utilize image-depth pairs generated by a discriminative model on an in-the-wild image dataset. As a generative model, our model can reliably estimate depth confidence, which provides an additional advantage. Our approach achieves competitive zero-shot performance on standard benchmarks of complex natural scenes while improving sampling efficiency and only requiring minimal synthetic data for training.

cs.CV↗

arXiv2024

ZigMa: A DiT-style Zigzag Mamba Diffusion Model

The diffusion model has long been plagued by scalability and quadratic complexity issues, especially within transformer-based structures. In this study, we aim to leverage the long sequence modeling capability of a State-Space Model called Mamba to extend its applicability to visual data generation. Firstly, we identify a critical oversight in most current Mamba-based vision methods, namely the lack of consideration for spatial continuity in the scan scheme of Mamba. Secondly, building upon this insight, we introduce a simple, plug-and-play, zero-parameter method named Zigzag Mamba, which outperforms Mamba-based baselines and demonstrates improved speed and memory utilization compared to transformer-based baselines. Lastly, we integrate Zigzag Mamba with the Stochastic Interpolant framework to investigate the scalability of the model on large-resolution visual datasets, such as FacesHQ $1024\times 1024$ and UCF101, MultiModal-CelebA-HQ, and MS COCO $256\times 256$ . Code will be released at https://taohu.me/zigma/

cs.CV↗

arXiv2024

CTRLorALTer: Conditional LoRAdapter for Efficient 0-Shot Control & Altering of T2I Models

Text-to-image generative models have become a prominent and powerful tool that excels at generating high-resolution realistic images. However, guiding the generative process of these models to consider detailed forms of conditioning reflecting style and/or structure information remains an open problem. In this paper, we present LoRAdapter, an approach that unifies both style and structure conditioning under the same formulation using a novel conditional LoRA block that enables zero-shot control. LoRAdapter is an efficient, powerful, and architecture-agnostic approach to condition text-to-image diffusion models, which enables fine-grained control conditioning during generation and outperforms recent state-of-the-art approaches.

cs.CV↗