Max Planck Institute for Informatics, Saarland Informatics Campus
Abstract:Stereo video generation has been gaining increasing attention with recent advancements in video diffusion models. However, most existing methods focus on generating 3D stereoscopic videos from monocular 2D videos. These approaches typically assume that the input monocular video is of high quality, making the task primarily about inpainting occluded regions in the warped video while preserving disoccluded areas. In this paper, we introduce a new pipeline that not only generates stereo videos but also enhances both left-view and right-view videos consistently with a single model. Our approach achieves this by fine-tuning the model on degraded data for restoration, as well as conditioning the model on warped masks for consistent stereo generation. As a result, our method can be fine-tuned on a relatively small synthetic stereo video datasets and applied to low-quality real-world videos, performing both stereo video generation and restoration. Experiments demonstrate that our method outperforms existing approaches both qualitatively and quantitatively in stereo video generation from low-resolution inputs.
Abstract:Finding correspondences between semantically similar points across images and object instances is one of the everlasting challenges in computer vision. While large pre-trained vision models have recently been demonstrated as effective priors for semantic matching, they still suffer from ambiguities for symmetric objects or repeated object parts. We propose to improve semantic correspondence estimation via 3D-aware pseudo-labeling. Specifically, we train an adapter to refine off-the-shelf features using pseudo-labels obtained via 3D-aware chaining, filtering wrong labels through relaxed cyclic consistency, and 3D spherical prototype mapping constraints. While reducing the need for dataset specific annotations compared to prior work, we set a new state-of-the-art on SPair-71k by over 4% absolute gain and by over 7% against methods with similar supervision requirements. The generality of our proposed approach simplifies extension of training to other data sources, which we demonstrate in our experiments.
Abstract:With recent advancements in neural rendering and motion capture algorithms, remarkable progress has been made in photorealistic human avatar modeling, unlocking immense potential for applications in virtual reality, augmented reality, remote communication, and industries such as gaming, film, and medicine. However, existing methods fail to provide complete, faithful, and expressive control over human avatars due to their entangled representation of facial expressions and body movements. In this work, we introduce Expressive Virtual Avatars (EVA), an actor-specific, fully controllable, and expressive human avatar framework that achieves high-fidelity, lifelike renderings in real time while enabling independent control of facial expressions, body movements, and hand gestures. Specifically, our approach designs the human avatar as a two-layer model: an expressive template geometry layer and a 3D Gaussian appearance layer. First, we present an expressive template tracking algorithm that leverages coarse-to-fine optimization to accurately recover body motions, facial expressions, and non-rigid deformation parameters from multi-view videos. Next, we propose a novel decoupled 3D Gaussian appearance model designed to effectively disentangle body and facial appearance. Unlike unified Gaussian estimation approaches, our method employs two specialized and independent modules to model the body and face separately. Experimental results demonstrate that EVA surpasses state-of-the-art methods in terms of rendering quality and expressiveness, validating its effectiveness in creating full-body avatars. This work represents a significant advancement towards fully drivable digital human models, enabling the creation of lifelike digital avatars that faithfully replicate human geometry and appearance.
Abstract:Generative diffusion models have advanced image editing with high-quality results and intuitive interfaces such as prompts and semantic drawing. However, these interfaces lack precise control, and the associated methods typically specialize on a single editing task. We introduce a versatile, generative workflow that operates in an intrinsic-image latent space, enabling semantic, local manipulation with pixel precision for a range of editing operations. Building atop the RGB-X diffusion framework, we address key challenges of identity preservation and intrinsic-channel entanglement. By incorporating exact diffusion inversion and disentangled channel manipulation, we enable precise, efficient editing with automatic resolution of global illumination effects -- all without additional data collection or model fine-tuning. We demonstrate state-of-the-art performance across a variety of tasks on complex images, including color and texture adjustments, object insertion and removal, global relighting, and their combinations.
Abstract:3D morphable models (3DMMs) are a powerful tool to represent the possible shapes and appearances of an object category. Given a single test image, 3DMMs can be used to solve various tasks, such as predicting the 3D shape, pose, semantic correspondence, and instance segmentation of an object. Unfortunately, 3DMMs are only available for very few object categories that are of particular interest, like faces or human bodies, as they require a demanding 3D data acquisition and category-specific training process. In contrast, we introduce a new method, Common3D, that learns 3DMMs of common objects in a fully self-supervised manner from a collection of object-centric videos. For this purpose, our model represents objects as a learned 3D template mesh and a deformation field that is parameterized as an image-conditioned neural network. Different from prior works, Common3D represents the object appearance with neural features instead of RGB colors, which enables the learning of more generalizable representations through an abstraction from pixel intensities. Importantly, we train the appearance features using a contrastive objective by exploiting the correspondences defined through the deformable template mesh. This leads to higher quality correspondence features compared to related works and a significantly improved model performance at estimating 3D object pose and semantic correspondence. Common3D is the first completely self-supervised method that can solve various vision tasks in a zero-shot manner.
Abstract:3D Gaussian Splatting (3DGS) is widely used for novel view synthesis due to its high rendering quality and fast inference time. However, 3DGS predominantly relies on first-order optimizers such as Adam, which leads to long training times. To address this limitation, we propose a novel second-order optimization strategy based on Levenberg-Marquardt (LM) and Conjugate Gradient (CG), which we specifically tailor towards Gaussian Splatting. Our key insight is that the Jacobian in 3DGS exhibits significant sparsity since each Gaussian affects only a limited number of pixels. We exploit this sparsity by proposing a matrix-free and GPU-parallelized LM optimization. To further improve its efficiency, we propose sampling strategies for both the camera views and loss function and, consequently, the normal equation, significantly reducing the computational complexity. In addition, we increase the convergence rate of the second-order approximation by introducing an effective heuristic to determine the learning rate that avoids the expensive computation cost of line search methods. As a result, our method achieves a $3\times$ speedup over standard LM and outperforms Adam by $~6\times$ when the Gaussian count is low while remaining competitive for moderate counts. Project Page: https://vcai.mpi-inf.mpg.de/projects/LM-IS
Abstract:This work focuses on tracking and understanding human motion using consumer wearable devices, such as VR/AR headsets, smart glasses, cellphones, and smartwatches. These devices provide diverse, multi-modal sensor inputs, including egocentric images, and 1-3 sparse IMU sensors in varied combinations. Motion descriptions can also accompany these signals. The diverse input modalities and their intermittent availability pose challenges for consistent motion capture and understanding. In this work, we present Ego4o (o for omni), a new framework for simultaneous human motion capture and understanding from multi-modal egocentric inputs. This method maintains performance with partial inputs while achieving better results when multiple modalities are combined. First, the IMU sensor inputs, the optional egocentric image, and text description of human motion are encoded into the latent space of a motion VQ-VAE. Next, the latent vectors are sent to the VQ-VAE decoder and optimized to track human motion. When motion descriptions are unavailable, the latent vectors can be input into a multi-modal LLM to generate human motion descriptions, which can further enhance motion capture accuracy. Quantitative and qualitative evaluations demonstrate the effectiveness of our method in predicting accurate human motion and high-quality motion descriptions.
Abstract:Driving a high-quality and photorealistic full-body human avatar, from only a few RGB cameras, is a challenging problem that has become increasingly relevant with emerging virtual reality technologies. To democratize such technology, a promising solution may be a generalizable method that takes sparse multi-view images of an unseen person and then generates photoreal free-view renderings of such identity. However, the current state of the art is not scalable to very large datasets and, thus, lacks in diversity and photorealism. To address this problem, we propose a novel, generalizable full-body model for rendering photoreal humans in free viewpoint, as driven by sparse multi-view video. For the first time in literature, our model can scale up training to thousands of subjects while maintaining high photorealism. At the core, we introduce a MultiHeadUNet architecture, which takes sparse multi-view images in texture space as input and predicts Gaussian primitives represented as 2D texels on top of a human body mesh. Importantly, we represent sparse-view image information, body shape, and the Gaussian parameters in 2D so that we can design a deep and scalable architecture entirely based on 2D convolutions and attention mechanisms. At test time, our method synthesizes an articulated 3D Gaussian-based avatar from as few as four input views and a tracked body template for unseen identities. Our method excels over prior works by a significant margin in terms of cross-subject generalization capability as well as photorealism.
Abstract:Egocentric motion capture with a head-mounted body-facing stereo camera is crucial for VR and AR applications but presents significant challenges such as heavy occlusions and limited annotated real-world data. Existing methods rely on synthetic pretraining and struggle to generate smooth and accurate predictions in real-world settings, particularly for lower limbs. Our work addresses these limitations by introducing a lightweight VR-based data collection setup with on-board, real-time 6D pose tracking. Using this setup, we collected the most extensive real-world dataset for ego-facing ego-mounted cameras to date in size and motion variability. Effectively integrating this multimodal input -- device pose and camera feeds -- is challenging due to the differing characteristics of each data source. To address this, we propose FRAME, a simple yet effective architecture that combines device pose and camera feeds for state-of-the-art body pose prediction through geometrically sound multimodal integration and can run at 300 FPS on modern hardware. Lastly, we showcase a novel training strategy to enhance the model's generalization capabilities. Our approach exploits the problem's geometric properties, yielding high-quality motion capture free from common artifacts in prior works. Qualitative and quantitative evaluations, along with extensive comparisons, demonstrate the effectiveness of our method. Data, code, and CAD designs will be available at https://vcai.mpi-inf.mpg.de/projects/FRAME/
Abstract:3D reconstruction of highly deformable surfaces (e.g. cloths) from monocular RGB videos is a challenging problem, and no solution provides a consistent and accurate recovery of fine-grained surface details. To account for the ill-posed nature of the setting, existing methods use deformation models with statistical, neural, or physical priors. They also predominantly rely on nonadaptive discrete surface representations (e.g. polygonal meshes), perform frame-by-frame optimisation leading to error propagation, and suffer from poor gradients of the mesh-based differentiable renderers. Consequently, fine surface details such as cloth wrinkles are often not recovered with the desired accuracy. In response to these limitations, we propose ThinShell-SfT, a new method for non-rigid 3D tracking that represents a surface as an implicit and continuous spatiotemporal neural field. We incorporate continuous thin shell physics prior based on the Kirchhoff-Love model for spatial regularisation, which starkly contrasts the discretised alternatives of earlier works. Lastly, we leverage 3D Gaussian splatting to differentiably render the surface into image space and optimise the deformations based on analysis-bysynthesis principles. Our Thin-Shell-SfT outperforms prior works qualitatively and quantitatively thanks to our continuous surface formulation in conjunction with a specially tailored simulation prior and surface-induced 3D Gaussians. See our project page at https://4dqv.mpiinf.mpg.de/ThinShellSfT.