Technical University of Munich
Abstract:We introduce MultiDiff, a novel approach for consistent novel view synthesis of scenes from a single RGB image. The task of synthesizing novel views from a single reference image is highly ill-posed by nature, as there exist multiple, plausible explanations for unobserved areas. To address this issue, we incorporate strong priors in form of monocular depth predictors and video-diffusion models. Monocular depth enables us to condition our model on warped reference images for the target views, increasing geometric stability. The video-diffusion prior provides a strong proxy for 3D scenes, allowing the model to learn continuous and pixel-accurate correspondences across generated images. In contrast to approaches relying on autoregressive image generation that are prone to drifts and error accumulation, MultiDiff jointly synthesizes a sequence of frames yielding high-quality and multi-view consistent results -- even for long-term scene generation with large camera movements, while reducing inference time by an order of magnitude. For additional consistency and image quality improvements, we introduce a novel, structured noise distribution. Our experimental results demonstrate that MultiDiff outperforms state-of-the-art methods on the challenging, real-world datasets RealEstate10K and ScanNet. Finally, our model naturally supports multi-view consistent editing without the need for further tuning.
Abstract:Learning 3D head priors from large 2D image collections is an important step towards high-quality 3D-aware human modeling. A core requirement is an efficient architecture that scales well to large-scale datasets and large image resolutions. Unfortunately, existing 3D GANs struggle to scale to generate samples at high resolutions due to their relatively slow train and render speeds, and typically have to rely on 2D superresolution networks at the expense of global 3D consistency. To address these challenges, we propose Generative Gaussian Heads (GGHead), which adopts the recent 3D Gaussian Splatting representation within a 3D GAN framework. To generate a 3D representation, we employ a powerful 2D CNN generator to predict Gaussian attributes in the UV space of a template head mesh. This way, GGHead exploits the regularity of the template's UV layout, substantially facilitating the challenging task of predicting an unstructured set of 3D Gaussians. We further improve the geometric fidelity of the generated 3D representations with a novel total variation loss on rendered UV coordinates. Intuitively, this regularization encourages that neighboring rendered pixels should stem from neighboring Gaussians in the template's UV space. Taken together, our pipeline can efficiently generate 3D heads trained only from single-view 2D image observations. Our proposed framework matches the quality of existing 3D head GANs on FFHQ while being both substantially faster and fully 3D consistent. As a result, we demonstrate real-time generation and rendering of high-quality 3D-consistent heads at $1024^2$ resolution for the first time.
Abstract:The creation of high-fidelity, digital versions of human heads is an important stepping stone in the process of further integrating virtual components into our everyday lives. Constructing such avatars is a challenging research problem, due to a high demand for photo-realism and real-time rendering performance. In this work, we propose Neural Parametric Gaussian Avatars (NPGA), a data-driven approach to create high-fidelity, controllable avatars from multi-view video recordings. We build our method around 3D Gaussian Splatting for its highly efficient rendering and to inherit the topological flexibility of point clouds. In contrast to previous work, we condition our avatars' dynamics on the rich expression space of neural parametric head models (NPHM), instead of mesh-based 3DMMs. To this end, we distill the backward deformation field of our underlying NPHM into forward deformations which are compatible with rasterization-based rendering. All remaining fine-scale, expression-dependent details are learned from the multi-view videos. To increase the representational capacity of our avatars, we augment the canonical Gaussian point cloud using per-primitive latent features which govern its dynamic behavior. To regularize this increased dynamic expressivity, we propose Laplacian terms on the latent features and predicted dynamics. We evaluate our method on the public NeRSemble dataset, demonstrating that NPGA significantly outperforms the previous state-of-the-art avatars on the self-reenactment task by 2.6 PSNR. Furthermore, we demonstrate accurate animation capabilities from real-world monocular videos.
Abstract:As large language models (LLMs) evolve, their integration with 3D spatial data (3D-LLMs) has seen rapid progress, offering unprecedented capabilities for understanding and interacting with physical spaces. This survey provides a comprehensive overview of the methodologies enabling LLMs to process, understand, and generate 3D data. Highlighting the unique advantages of LLMs, such as in-context learning, step-by-step reasoning, open-vocabulary capabilities, and extensive world knowledge, we underscore their potential to significantly advance spatial comprehension and interaction within embodied Artificial Intelligence (AI) systems. Our investigation spans various 3D data representations, from point clouds to Neural Radiance Fields (NeRFs). It examines their integration with LLMs for tasks such as 3D scene understanding, captioning, question-answering, and dialogue, as well as LLM-based agents for spatial reasoning, planning, and navigation. The paper also includes a brief review of other methods that integrate 3D and language. The meta-analysis presented in this paper reveals significant progress yet underscores the necessity for novel approaches to harness the full potential of 3D-LLMs. Hence, with this paper, we aim to chart a course for future research that explores and expands the capabilities of 3D-LLMs in understanding and interacting with the complex 3D world. To support this survey, we have established a project page where papers related to our topic are organized and listed: https://github.com/ActiveVisionLab/Awesome-LLM-3D.
Abstract:We present Mesh2NeRF, an approach to derive ground-truth radiance fields from textured meshes for 3D generation tasks. Many 3D generative approaches represent 3D scenes as radiance fields for training. Their ground-truth radiance fields are usually fitted from multi-view renderings from a large-scale synthetic 3D dataset, which often results in artifacts due to occlusions or under-fitting issues. In Mesh2NeRF, we propose an analytic solution to directly obtain ground-truth radiance fields from 3D meshes, characterizing the density field with an occupancy function featuring a defined surface thickness, and determining view-dependent color through a reflection function considering both the mesh and environment lighting. Mesh2NeRF extracts accurate radiance fields which provides direct supervision for training generative NeRFs and single scene representation. We validate the effectiveness of Mesh2NeRF across various tasks, achieving a noteworthy 3.12dB improvement in PSNR for view synthesis in single scene representation on the ABO dataset, a 0.69 PSNR enhancement in the single-view conditional generation of ShapeNet Cars, and notably improved mesh extraction from NeRF in the unconditional generation of Objaverse Mugs.
Abstract:Recently, significant progress has been achieved in sensing real large-scale outdoor 3D environments, particularly by using modern acquisition equipment such as LiDAR sensors. Unfortunately, they are fundamentally limited in their ability to produce dense, complete 3D scenes. To address this issue, recent learning-based methods integrate neural implicit representations and optimizable feature grids to approximate surfaces of 3D scenes. However, naively fitting samples along raw LiDAR rays leads to noisy 3D mapping results due to the nature of sparse, conflicting LiDAR measurements. Instead, in this work we depart from fitting LiDAR data exactly, instead letting the network optimize a non-metric monotonic implicit field defined in 3D space. To fit our field, we design a learning system integrating a monotonicity loss that enables optimizing neural monotonic fields and leverages recent progress in large-scale 3D mapping. Our algorithm achieves high-quality dense 3D mapping performance as captured by multiple quantitative and perceptual measures and visual results obtained for Mai City, Newer College, and KITTI benchmarks. The code of our approach will be made publicly available.
Abstract:We introduce LightIt, a method for explicit illumination control for image generation. Recent generative methods lack lighting control, which is crucial to numerous artistic aspects of image generation such as setting the overall mood or cinematic appearance. To overcome these limitations, we propose to condition the generation on shading and normal maps. We model the lighting with single bounce shading, which includes cast shadows. We first train a shading estimation module to generate a dataset of real-world images and shading pairs. Then, we train a control network using the estimated shading and normals as input. Our method demonstrates high-quality image generation and lighting control in numerous scenes. Additionally, we use our generated dataset to train an identity-preserving relighting model, conditioned on an image and a target shading. Our method is the first that enables the generation of images with controllable, consistent lighting and performs on par with specialized relighting state-of-the-art methods.
Abstract:Recently, progress in acquisition equipment such as LiDAR sensors has enabled sensing increasingly spacious outdoor 3D environments. Making sense of such 3D acquisitions requires fine-grained scene understanding, such as constructing instance-based 3D scene segmentations. Commonly, a neural network is trained for this task; however, this requires access to a large, densely annotated dataset, which is widely known to be challenging to obtain. To address this issue, in this work we propose to predict instance segmentations for 3D scenes in an unsupervised way, without relying on ground-truth annotations. To this end, we construct a learning framework consisting of two components: (1) a pseudo-annotation scheme for generating initial unsupervised pseudo-labels; and (2) a self-training algorithm for instance segmentation to fit robust, accurate instances from initial noisy proposals. To enable generating 3D instance mask proposals, we construct a weighted proxy-graph by connecting 3D points with edges integrating multi-modal image- and point-based self-supervised features, and perform graph-cuts to isolate individual pseudo-instances. We then build on a state-of-the-art point-based architecture and train a 3D instance segmentation model, resulting in significant refinement of initial proposals. To scale to arbitrary complexity 3D scenes, we design our algorithm to operate on local 3D point chunks and construct a merging step to generate scene-level instance segmentations. Experiments on the challenging SemanticKITTI benchmark demonstrate the potential of our approach, where it attains 13.3% higher Average Precision and 9.1% higher F1 score compared to the best-performing baseline. The code will be made publicly available at https://github.com/artonson/autoinst.
Abstract:3D asset generation is getting massive amounts of attention, inspired by the recent success of text-guided 2D content creation. Existing text-to-3D methods use pretrained text-to-image diffusion models in an optimization problem or fine-tune them on synthetic data, which often results in non-photorealistic 3D objects without backgrounds. In this paper, we present a method that leverages pretrained text-to-image models as a prior, and learn to generate multi-view images in a single denoising process from real-world data. Concretely, we propose to integrate 3D volume-rendering and cross-frame-attention layers into each block of the existing U-Net network of the text-to-image model. Moreover, we design an autoregressive generation that renders more 3D-consistent images at any viewpoint. We train our model on real-world datasets of objects and showcase its capabilities to generate instances with a variety of high-quality shapes and textures in authentic surroundings. Compared to the existing methods, the results generated by our method are consistent, and have favorable visual quality (-30% FID, -37% KID).
Abstract:We introduce Motion2VecSets, a 4D diffusion model for dynamic surface reconstruction from point cloud sequences. While existing state-of-the-art methods have demonstrated success in reconstructing non-rigid objects using neural field representations, conventional feed-forward networks encounter challenges with ambiguous observations from noisy, partial, or sparse point clouds. To address these challenges, we introduce a diffusion model that explicitly learns the shape and motion distribution of non-rigid objects through an iterative denoising process of compressed latent representations. The diffusion-based prior enables more plausible and probabilistic reconstructions when handling ambiguous inputs. We parameterize 4D dynamics with latent vector sets instead of using a global latent. This novel 4D representation allows us to learn local surface shape and deformation patterns, leading to more accurate non-linear motion capture and significantly improving generalizability to unseen motions and identities. For more temporal-coherent object tracking, we synchronously denoise deformation latent sets and exchange information across multiple frames. To avoid the computational overhead, we design an interleaved space and time attention block to alternately aggregate deformation latents along spatial and temporal domains. Extensive comparisons against the state-of-the-art methods demonstrate the superiority of our Motion2VecSets in 4D reconstruction from various imperfect observations, notably achieving a 19% improvement in Intersection over Union (IoU) compared to CaDex for reconstructing unseen individuals from sparse point clouds on the DeformingThings4D-Animals dataset. More detailed information can be found at https://vveicao.github.io/projects/Motion2VecSets/.