Abstract:Accurately estimating scene lighting is critical for applications such as mixed reality. Existing works estimate illumination by generating illumination maps or regressing illumination parameters. However, the method of generating illumination maps has poor generalization performance and parametric models such as Spherical Harmonic (SH) and Spherical Gaussian (SG) fall short in capturing high-frequency or low-frequency components. This paper presents MixLight, a joint model that utilizes the complementary characteristics of SH and SG to achieve a more complete illumination representation, which uses SH and SG to capture low-frequency ambient and high-frequency light sources respectively. In addition, a special spherical light source sparsemax (SLSparsemax) module that refers to the position and brightness relationship between spherical light sources is designed to improve their sparsity, which is significant but omitted by prior works. Extensive experiments demonstrate that MixLight surpasses state-of-the-art (SOTA) methods on multiple metrics. In addition, experiments on Web Dataset also show that MixLight as a parametric method has better generalization performance than non-parametric methods.
Abstract:Depth completion, aiming to predict dense depth maps from sparse depth measurements, plays a crucial role in many computer vision related applications. Deep learning approaches have demonstrated overwhelming success in this task. However, high-precision depth completion without relying on the ground-truth data, which are usually costly, still remains challenging. The reason lies on the ignorance of 3D structural information in most previous unsupervised solutions, causing inaccurate spatial propagation and mixed-depth problems. To alleviate the above challenges, this paper explores the utilization of 3D perceptual features and multi-view geometry consistency to devise a high-precision self-supervised depth completion method. Firstly, a 3D perceptual spatial propagation algorithm is constructed with a point cloud representation and an attention weighting mechanism to capture more reasonable and favorable neighboring features during the iterative depth propagation process. Secondly, the multi-view geometric constraints between adjacent views are explicitly incorporated to guide the optimization of the whole depth completion model in a self-supervised manner. Extensive experiments on benchmark datasets of NYU-Depthv2 and VOID demonstrate that the proposed model achieves the state-of-the-art depth completion performance compared with other unsupervised methods, and competitive performance compared with previous supervised methods.
Abstract:Reconstructing 3D objects from extremely sparse views is a long-standing and challenging problem. While recent techniques employ image diffusion models for generating plausible images at novel viewpoints or for distilling pre-trained diffusion priors into 3D representations using score distillation sampling (SDS), these methods often struggle to simultaneously achieve high-quality, consistent, and detailed results for both novel-view synthesis (NVS) and geometry. In this work, we present Sparse3D, a novel 3D reconstruction method tailored for sparse view inputs. Our approach distills robust priors from a multiview-consistent diffusion model to refine a neural radiance field. Specifically, we employ a controller that harnesses epipolar features from input views, guiding a pre-trained diffusion model, such as Stable Diffusion, to produce novel-view images that maintain 3D consistency with the input. By tapping into 2D priors from powerful image diffusion models, our integrated model consistently delivers high-quality results, even when faced with open-world objects. To address the blurriness introduced by conventional SDS, we introduce the category-score distillation sampling (C-SDS) to enhance detail. We conduct experiments on CO3DV2 which is a multi-view dataset of real-world objects. Both quantitative and qualitative evaluations demonstrate that our approach outperforms previous state-of-the-art works on the metrics regarding NVS and geometry reconstruction.
Abstract:The recent neural surface reconstruction by volume rendering approaches have made much progress by achieving impressive surface reconstruction quality, but are still limited to dense and highly accurate posed views. To overcome such drawbacks, this paper pays special attention on the consistent surface reconstruction from sparse views with noisy camera poses. Unlike previous approaches, the key difference of this paper is to exploit the multi-view constraints directly from the explicit geometry of the neural surface, which can be used as effective regularization to jointly learn the neural surface and refine the camera poses. To build effective multi-view constraints, we introduce a fast differentiable on-surface intersection to generate on-surface points, and propose view-consistent losses based on such differentiable points to regularize the neural surface learning. Based on this point, we propose a jointly learning strategy for neural surface and camera poses, named SC-NeuS, to perform geometry-consistent surface reconstruction in an end-to-end manner. With extensive evaluation on public datasets, our SC-NeuS can achieve consistently better surface reconstruction results with fine-grained details than previous state-of-the-art neural surface reconstruction approaches, especially from sparse and noisy camera views.
Abstract:High-fidelity 3D scene reconstruction from monocular videos continues to be challenging, especially for complete and fine-grained geometry reconstruction. The previous 3D reconstruction approaches with neural implicit representations have shown a promising ability for complete scene reconstruction, while their results are often over-smooth and lack enough geometric details. This paper introduces a novel neural implicit scene representation with volume rendering for high-fidelity online 3D scene reconstruction from monocular videos. For fine-grained reconstruction, our key insight is to incorporate geometric priors into both the neural implicit scene representation and neural volume rendering, thus leading to an effective geometry learning mechanism based on volume rendering optimization. Benefiting from this, we present MonoNeuralFusion to perform the online neural 3D reconstruction from monocular videos, by which the 3D scene geometry is efficiently generated and optimized during the on-the-fly 3D monocular scanning. The extensive comparisons with state-of-the-art approaches show that our MonoNeuralFusion consistently generates much better complete and fine-grained reconstruction results, both quantitatively and qualitatively.
Abstract:Previous online 3D dense reconstruction methods often cost massive memory storage while achieving unsatisfactory surface quality mainly due to the usage of stagnant underlying geometry representation, such as TSDF (truncated signed distance functions) or surfels, without any knowledge of the scene priors. In this paper, we present DI-Fusion (Deep Implicit Fusion), based on a novel 3D representation, called Probabilistic Local Implicit Voxels (PLIVoxs), for online 3D reconstruction using a commodity RGB-D camera. Our PLIVox encodes scene priors considering both the local geometry and uncertainty parameterized by a deep neural network. With such deep priors, we demonstrate by extensive experiments that we are able to perform online implicit 3D reconstruction achieving state-of-the-art mapping quality and camera trajectory estimation accuracy, while taking much less storage compared with previous online 3D reconstruction approaches.
Abstract:In this paper, we present a novel tightly-coupled probabilistic monocular visual-odometric Simultaneous Localization and Mapping algorithm using wheels and a MEMS gyroscope, which can provide accurate, robust and long-term localization for the ground robot moving on a plane. Firstly, we present an odometer preintegration theory that integrates the wheel encoder measurements and gyroscope measurements to a local frame. The preintegration theory properly addresses the manifold structure of the rotation group SO(3) and carefully deals with uncertainty propagation and bias correction. Then the novel odometer error term is formulated using the odometer preintegration model and it is tightly integrated into the visual optimization framework. Furthermore, we introduce a complete tracking framework to provide different strategies for motion tracking when (1) both measurements are available, (2) visual measurements are not available, and (3) wheel encoder experiences slippage, which leads the system to be accurate and robust. Finally, the proposed algorithm is evaluated by performing extensive experiments, the experimental results demonstrate the superiority of the proposed system.
Abstract:This paper presents a novel tightly-coupled monocular visual-inertial Simultaneous Localization and Mapping algorithm, which provides accurate and robust localization within the globally consistent map in real time on a standard CPU. This is achieved by firstly performing the visual-inertial extended kalman filter(EKF) to provide motion estimate at a high rate. However the filter becomes inconsistent due to the well known linearization issues. So we perform a keyframe-based visual-inertial bundle adjustment to improve the consistency and accuracy of the system. In addition, a loop closure detection and correction module is also added to eliminate the accumulated drift when revisiting an area. Finally, the optimized motion estimates and map are fed back to the EKF-based visual-inertial odometry module, thus the inconsistency and estimation error of the EKF estimator are reduced. In this way, the system can continuously provide reliable motion estimates for the long-term operation. The performance of the algorithm is validated on public datasets and real-world experiments, which proves the superiority of the proposed algorithm.