Synthesizing natural human motion that adapts to complex environments while allowing creative control remains a fundamental challenge in motion synthesis. Existing models often fall short, either by assuming flat terrain or lacking the ability to control motion semantics through text. To address these limitations, we introduce SCENIC, a diffusion model designed to generate human motion that adapts to dynamic terrains within virtual scenes while enabling semantic control through natural language. The key technical challenge lies in simultaneously reasoning about complex scene geometry while maintaining text control. This requires understanding both high-level navigation goals and fine-grained environmental constraints. The model must ensure physical plausibility and precise navigation across varied terrain, while also preserving user-specified text control, such as “carefully stepping over obstacles” or “walking upstairs like a zombie.” Our solution introduces a hierarchical scene reasoning approach. At the core of our method is a novel hierarchical scene reasoning framework. It combines two key components: a motion-scene cross-attention block that aligns the human body's motion features with local scene geometry, enabling precise low-level interactions; and a target point canonicalization module that provides global goal conditioning by normalizing target scene coordinates for high-level guidance. To ensure plausibility and naturalness, we leverage a pre-trained motion diffusion prior and apply scene-constrained diffusion noise optimization during sampling, enabling long-horizon motion generation that respects both scene structure and semantic text input. Experiments demonstrate that our novel diffusion model generates arbitrarily long human motions that both adapt to complex scenes with varying terrain surfaces and respond to textual prompts. Additionally, we show SCENIC can generalize to four real-scene datasets.

Interactions between human and objects are influenced not only by the object's pose and shape, but also by physical attributes such as object mass and surface friction. They introduce important motion nuances that are essential for diversity and realism. Despite advancements in recent human-object interaction methods, this aspect has been overlooked. Generating nuanced human motion presents two challenges. First, it is non-trivial to learn from multi-modal human and object information derived from both the physical and non-physical attributes. Second, there exists no dataset capturing nuanced human interactions with objects of varying physical properties, hampering model development. This work addresses the gap by introducing the FORCE model, an approach for synthesizing diverse, nuanced human-object interactions by modeling physical attributes. Our key insight is that human motion is dictated by the interrelation between the force exerted by the human and the perceived resistance. Guided by a novel intuitive physics encoding, the model captures the interplay between human force and resistance. Experiments also demonstrate incorporating human force facilitates learning multi-class motion. Accompanying our model, we contribute a dataset, which features diverse, different-styled motion through interactions with varying resistances.

Hand-object interaction modeling from a single RGB image is a significantly challenging task. Previous works typically reconstruct hand-object interactions as texture-less meshes, ignoring photo-realistic image generation. In this work, we introduce the HO123, a novel method to synthesize novel-view hand-object interaction images from a single image. To this end, we first train a 2D diffusion prior. Given the camera pose in novel views, our approach transfers the camera information into explicit hand representations, including hand depth and skeleton images. We propose a global hand embedding to control the diffusion model based on these hand representations. We then learn a 3D Gaussian splatting for novel-view rendering using the diffusion prior. However, occluded objects present a persistent challenge. To address this issue, we further introduce local hand embedding, where a contact field is defined in the 3D Gaussian Splatting. We leverage contact information to guide the rendering in the contact field. Extensive experiments on the HO3D and DexYCB datasets demonstrate that our method significantly outperforms state-of-the-art novel-view synthesis for hand-object interactions.

We propose Covisibility Map-Based Gaussian Splatting (CoMapGS), designed to recover underrepresented sparse regions in sparse novel view synthesis. CoMapGS addresses both high- and low-uncertainty regions by constructing covisibility maps, enhancing initial point clouds, and applying uncertainty-aware weighted supervision using a proximity classifier. Our contributions are threefold: (1) CoMapGS reframes novel view synthesis by leveraging covisibility maps as a core component to address region-specific uncertainty; (2) Enhanced initial point clouds for both low- and high-uncertainty regions compensate for sparse COLMAPderived point clouds, improving reconstruction quality and benefiting few-shot 3DGS methods; (3) Adaptive supervision with covisibility-score-based weighting and proximity classification achieves consistent performance gains across scenes with varying sparsity scores derived from covisibility maps. Experimental results demonstrate that CoMapGS outperforms state-of-the-art methods on datasets including Mip-NeRF 360 and LLFF.1

Dynamic Novel View Synthesis aims to generate photorealistic views of moving subjects from arbitrary viewpoints. This task is particularly challenging when relying on monocular video, where disentangling structure from motion is ill-posed and supervision is scarce. We introduce Video Diffusion-Aware Reconstruction (ViDAR), a novel 4D reconstruction framework that leverages personalised diffusion models to synthesise a pseudo multi-view supervision signal for training a Gaussian splatting representation. By conditioning on scene-specific features, ViDAR recovers fine-grained appearance details while mitigating artefacts introduced by monocular ambiguity. To address the spatio-temporal inconsistency of diffusion-based supervision, we propose a diffusion-aware loss function and a camera pose optimisation strategy that aligns synthetic views with the underlying scene geometry. Experiments on DyCheck, a challenging benchmark with extreme viewpoint variation, show that ViDAR outperforms all state-of-the-art baselines in visual quality and geometric consistency. We further highlight ViDAR’s strong improvement over baselines on dynamic regions and provide a new benchmark to compare performance in reconstructing motion-rich parts of the scene.

Modeling hand-object interactions is a fundamentally challenging task in 3D computer vision. Despite remarkable progress that has been achieved in this field, existing methods still fail to synthesize the hand-object interaction photo-realistically, suffering from degraded rendering quality caused by the heavy mutual occlusions between the hand and the object, and inaccurate hand-object pose estimation. To tackle these challenges, we present a novel free-viewpoint rendering framework, Neural Contact Radiance Field (NCRF), to reconstruct hand-object interactions from a sparse set of videos. In particular, the proposed NCRF framework consists of two key components: (a) A contact optimization field that predicts an accurate contact field from 3D query points for achieving desirable contact between the hand and the object. (b) A hand-object neural radiance field to learn an implicit hand-object representation in a static canonical space, in concert with the specifically designed hand-object motion field to produce observation-to-canonical correspondences. We jointly learn these key components where they mutually help and regularize each other with visual and geometric constraints, producing a high-quality hand-object reconstruction that achieves photorealistic novel view synthesis. Extensive experiments on HO3D and DexYCB datasets show that our approach outperforms the current state-of-the-art in terms of both rendering quality and pose estimation accuracy.

This work addresses the problem of real-time rendering of photorealistic human body avatars learned from multi-view videos. While the classical approaches to model and render virtual humans generally use a textured mesh, recent research has developed neural body representations that achieve impressive visual quality. However, these models are difficult to render in real-time and their quality degrades when the character is animated with body poses different than the training observations. We propose an animatable human model based on 3D Gaussian Splatting, that has recently emerged as a very efficient alternative to neural radiance fields. The body is represented by a set of gaussian primitives in a canonical space which is deformed with a coarse to fine approach that combines forward skinning and local non-rigid refinement. We describe how to learn our Human Gaussian Splatting (HuGS) model in an end-to-end fashion from multi-view observations, and evaluate it against the state-of-the-art approaches for novel pose synthesis of clothed body. Our method achieves 1.5 dB PSNR improvement over the state-of-the-art on THuman4 dataset while being able to render in real-time (≈ 80 fps for$512 \times 512$resolution).

Traditionally, 3d indoor datasets have generally prioritized scale over ground-truth accuracy in order to obtain improved generalization. However, using these datasets to evaluate dense geometry tasks, such as depth rendering, can be problematic as the meshes of the dataset are often incomplete and may produce wrong ground truth to evaluate the details. In this paper, we propose SCRREAM, a dataset annotation framework that allows annotation of fully dense meshes of objects in the scene and registers camera poses on the real image sequence, which can produce accurate ground truth for both sparse 3D as well as dense 3D tasks. We show the details of the dataset annotation pipeline and showcase four possible variants of datasets that can be obtained from our framework with example scenes, such as indoor reconstruction and SLAM, scene editing & object removal, human reconstruction and 6d pose estimation. Recent pipelines for indoor reconstruction and SLAM serve as new benchmarks. In contrast to previous indoor dataset, our design allows to evaluate dense geometry tasks on eleven sample scenes against accurately rendered ground truth depth maps.

Digital cameras transform sensor RAW readings into RGB images by means of their Image Signal Processor (ISP). Computational photography tasks such as image denoising and colour constancy are commonly performed in the RAW domain, in part due to the inherent hardware design, but also due to the appealing simplicity of noise statistics that result from the direct sensor readings. Despite this, the availability of RAW images is limited in comparison with the abundance and diversity of available RGB data. Recent approaches have attempted to bridge this gap by estimating the RGB to RAW mapping: handcrafted model-based methods that are interpretable and controllable usually require manual parameter fine-tuning, while end-to-end learnable neural networks require large amounts of training data, at times with complex training procedures, and generally lack interpretability and parametric control. Towards addressing these existing limitations, we present a novel hybrid model-based and data-driven ISP that builds on canonical ISP operations and is both learnable and interpretable. Our proposed invertible model, capable of bidirectional mapping between RAW and RGB domains, employs end-to-end learning of rich parameter representations, i.e. dictionaries, that are free from direct parametric supervision and additionally enable simple and plausible data augmentation. We evidence the value of our data generation process by extensive experiments under both RAW image reconstruction and RAW image denoising tasks, obtaining state-of-the-art performance in both. Additionally, we show that our ISP can learn meaningful mappings from few data samples, and that denoising models trained with our dictionary-based data augmentation are competitive despite having only few or zero ground-truth labels.

This paper is concerned with contrastive learning (CL) for low-level image restoration and enhancement tasks. We propose a new label-efficient learning paradigm based on residuals, residual contrastive learning (RCL), and derive an unsupervised visual representation learning framework, suitable for low-level vision tasks with noisy inputs. While supervised image reconstruction aims to minimize residual terms directly, RCL alternatively builds a connection between residuals and CL by defining a novel instance discrimination pretext task, using residuals as the discriminative feature. Our formulation mitigates the severe task misalignment between instance discrimination pretext tasks and downstream image reconstruction tasks, present in existing CL frameworks. Experimentally, we find that RCL can learn robust and transferable representations that improve the performance of various downstream tasks, such as denoising and super resolution, in comparison with recent self-supervised methods designed specifically for noisy inputs. Additionally, our unsupervised pre-training can significantly reduce annotation costs whilst maintaining performance competitive with fully-supervised image reconstruction.

High dynamic range (HDR) imaging is of fundamental importance in modern digital photography pipelines and used to produce a high-quality photograph with well exposed regions despite varying illumination across the image. This is typically achieved by merging multiple low dynamic range (LDR) images taken at different exposures. However, over-exposed regions and misalignment errors due to poorly compensated motion result in artefacts such as ghosting. In this paper, we present a new HDR imaging technique that specifically models alignment and exposure uncertainties to produce high quality HDR results. We introduce a strategy that learns to jointly align and assess the alignment and exposure reliability using an HDR-aware, uncertainty-driven attention map that robustly merges the frames into a single high quality HDR image. Further, we introduce a progressive, multi-stage image fusion approach that can flexibly merge any number of LDR images in a permutation-invariant manner. Experimental results show our method can produce better quality HDR images with up to 1.1dB PSNR improvement to the state-of-the-art, and subjective improvements in terms of better detail, colours, and fewer artefacts.

The main challenge in Super Resolution (SR) is to discover the mapping between the low-and high-resolution manifolds of image patches, a complex ill-posed problem which has recently been addressed through piecewise linear regression with promising results. In this paper we present a novel regression-based SR algorithm that benefits from an extended knowledge of the structure of both manifolds. We propose a transform that collapses the 16 variations induced from the dihedral group of transforms (i.e. rotations, vertical and horizontal reflections) and antipodality (i.e. diametrically opposed points in the unitary sphere) into a single primitive. The key idea of our transform is to study the different dihedral elements as a group of symmetries within the high-dimensional manifold. We obtain the respective set of mirror-symmetry axes by means of a frequency analysis of the dihedral elements, and we use them to collapse the redundant variability through a modified symmetry distance. The experimental validation of our algorithm shows the effectiveness of our approach, which obtains competitive quality with a dictionary of as little as 32 atoms (reducing other methods' dictionaries by at least a factor of 32) and further pushing the state-of-the-art with a 1024 atoms dictionary.

Recent research in piecewise linear regression for Super-Resolution has shown the positive impact of training regressors with densely populated clusters whose datapoints are tight in the Euclidean space. In this paper we further research how to improve the locality condition during the training of regressors and how to better select them during testing time. We study the characteristics of the metrics best suited for the piecewise regression algorithms, in which comparisons are usually made between normalized vectors that lie on the unitary hypersphere. Even though Euclidean distance has been widely used for this purpose, it is suboptimal since it does not handle antipodal points (i.e. diametrically opposite points) properly, as vectors with same module and angle but opposite directions are, for linear regression purposes, identical. Therefore, we propose the usage of antipodally invariant metrics and introduce the Half Hypersphere Confinement (HHC), a fast alternative to Multidimensional Scaling (MDS) that allows to map antipodally invariant distances in the Euclidean space with very little approximation error By doing so, we enable the usage of fast search structures based on Euclidean distances without undermining their speed gains with complex distance transformations. The performance of our method, which we named HHC Regression (HHCR), applied to SuperResolution (SR) improves both in quality (PSNR) and it is faster than any other state-of-the-art method. Additionally, under an application-agnostic interpretation of our regression framework, we also test our algorithm for denoising and depth upscaling with promising results.

Dictionary-based super-resolution (SR) algorithms usually select dictionary atoms based on the distance or similarity metrics. Although the optimal selection of the nearest neighbors is of central importance for such methods, the impact of using proper metrics for SR has been overlooked in literature, mainly due to the vast usage of Euclidean distance. In this paper, we present a very fast regression-based algorithm, which builds on the densely populated anchored neighborhoods and sublinear search structures. We perform a study of the nature of the features commonly used for SR, observing that those features usually lie in the unitary hypersphere, where every point has a diametrically opposite one, i.e., its antipode, with same module and angle, but the opposite direction. Even though, we validate the benefits of using antipodally invariant metrics, most of the binary splits use Euclidean distance, which does not handle antipodes optimally. In order to benefit from both the worlds, we propose a simple yet effective antipodally invariant transform that can be easily included in the Euclidean distance calculation. We modify the original spherical hashing algorithm with this metric in our antipodally invariant spherical hashing scheme, obtaining the same performance as a pure antipodally invariant metric. We round up our contributions with a novel feature transform that obtains a better coarse approximation of the input image thanks to iterative backprojection. The performance of our method, which we named antipodally invariant SR, improves quality (Peak Signal to Noise Ratio) and it is faster than any other state-of-the-art method.

This paper presents a fast, high-performance method for super resolution with external learning. The first contribution leading to the excellent performance is a bimodal tree for clustering, which successfully exploits the antipodal invariance of the coarse-to-high-res mapping of natural image patches and provides scalability to finer partitions of the underlying coarse patch space. During training an ensemble of such bimodal trees is computed, providing different linearizations of the mapping. The second and main contribution is a fast inference algorithm, which selects the most suitable mapping function within the tree ensemble for each patch by adopting a Local Naive Bayes formulation. The experimental validation shows promising scalability properties that reflect the suitability of the proposed model, which may also be generalized to other tasks. The resulting method is beyond one order of magnitude faster and performs objectively and subjectively better than the current state of the art.

We present a noise-aware single-image super-resolution (SI-SR) algorithm, which automatically cancels additive noise while adding detail learned from lower-resolution scales. In contrast with most SI-SR techniques, we do not assume the input image to be a clean source of examples. Instead, we adapt the recent and efficient in-place cross-scale self-similarity prior for both learning fine detail examples and reducing image noise. Our experiments show a promising performance, despite the relatively simple algorithm. Both objective evaluations and subjective validations show clear quality improvements when upscaling noisy images.

The performance of dictionary-based super-resolution (SR) strongly depends on the\ncontents of the training dataset. Nevertheless, many dictionary-based SR methods randomly select patches from of a larger set of training images to build their dictionaries\n[\n8\n,\n14\n,\n19\n,\n20\n], thus relying on patches being diverse enough. This paper describes\na dictionary building method for SR based on adaptively selecting an optimal subset of\npatches out of the training images. Each training image is divided into sub-image entities,\nnamed regions, of such a size that texture consistency is preserved and high-frequency\n(HF) energy is present. For each input patch to super-resolve, the best-fitting region is\nfound through a Bayesian selection. In order to handle the high number of regions in\nthe training dataset, a local Naive Bayes Nearest Neighbor (NBNN) approach is used.\nTrained with this adapted subset of patches, sparse coding SR is applied to recover the\nhigh-resolution image. Experimental results demonstrate that using our adaptive algo-\nrithm produces an improvement in SR performance with respect to non-adaptive training.

This paper presents a new method for estimating a super-resolved version of an observed image by exploiting cross-scale self-similarity. We extend prior work on single-image super-resolution by introducing an adaptive selection of the best fitting upscaling and analysis filters for example learning. This selection is based on local error measurements obtained by using each filter with every image patch, and contrasts with the common approach of a constant metric in both dictionary-based and internal learning super-resolution. The proposed method is suitable for interactive applications, offering low computational load and a parallelizable design that allows straight-forward GPU implementations. Experimental results also show how our method generalizes better to different datasets than dictionary-based super-resolution and comparably to internal learning with adaptive post-processing.