Abstract:The field of self-supervised 3D representation learning has emerged as a promising solution to alleviate the challenge presented by the scarcity of extensive, well-annotated datasets. However, it continues to be hindered by the lack of diverse, large-scale, real-world 3D scene datasets for source data. To address this shortfall, we propose Generalizable Representation Learning (GRL), where we devise a generative Bayesian network to produce diverse synthetic scenes with real-world patterns, and conduct pre-training with a joint objective. By jointly learning a coarse-to-fine contrastive learning task and an occlusion-aware reconstruction task, the model is primed with transferable, geometry-informed representations. Post pre-training on synthetic data, the acquired knowledge of the model can be seamlessly transferred to two principal downstream tasks associated with 3D scene understanding, namely 3D object detection and 3D semantic segmentation, using real-world benchmark datasets. A thorough series of experiments robustly display our method's consistent superiority over existing state-of-the-art pre-training approaches.
Abstract:The use of synthetic data in indoor 3D object detection offers the potential of greatly reducing the manual labor involved in 3D annotations and training effective zero-shot detectors. However, the complicated domain shifts across syn-to-real indoor datasets remains underexplored. In this paper, we propose a novel Object-wise Hierarchical Domain Alignment (OHDA) framework for syn-to-real unsupervised domain adaptation in indoor 3D object detection. Our approach includes an object-aware augmentation strategy to effectively diversify the source domain data, and we introduce a two-branch adaptation framework consisting of an adversarial training branch and a pseudo labeling branch, in order to simultaneously reach holistic-level and class-level domain alignment. The pseudo labeling is further refined through two proposed schemes specifically designed for indoor UDA. Our adaptation results from synthetic dataset 3D-FRONT to real-world datasets ScanNetV2 and SUN RGB-D demonstrate remarkable mAP25 improvements of 9.7% and 9.1% over Source-Only baselines, respectively, and consistently outperform the methods adapted from 2D and 3D outdoor scenarios. The code will be publicly available upon paper acceptance.
Abstract:In this work, we tackle the task of learning generalizable 3D human Gaussians from a single image. The main challenge for this task is to recover detailed geometry and appearance, especially for the unobserved regions. To this end, we propose single-view generalizable Human Gaussian model (HGM), a diffusion-guided framework for 3D human modeling from a single image. We design a diffusion-based coarse-to-fine pipeline, where the diffusion model is adapted to refine novel-view images rendered from a coarse human Gaussian model. The refined images are then used together with the input image to learn a refined human Gaussian model. Although effective in hallucinating the unobserved views, the approach may generate unrealistic human pose and shapes due to the lack of supervision. We circumvent this problem by further encoding the geometric priors from SMPL model. Specifically, we propagate geometric features from SMPL volume to the predicted Gaussians via sparse convolution and attention mechanism. We validate our approach on publicly available datasets and demonstrate that it significantly surpasses state-of-the-art methods in terms of PSNR and SSIM. Additionally, our method exhibits strong generalization for in-the-wild images.
Abstract:Although 3D Gaussian Splatting has been widely studied because of its realistic and efficient novel-view synthesis, it is still challenging to extract a high-quality surface from the point-based representation. Previous works improve the surface by incorporating geometric priors from the off-the-shelf normal estimator. However, there are two main limitations: 1) Supervising normal rendered from 3D Gaussians updates only the rotation parameter while neglecting other geometric parameters; 2) The inconsistency of predicted normal maps across multiple views may lead to severe reconstruction artifacts. In this paper, we propose a Depth-Normal regularizer that directly couples normal with other geometric parameters, leading to full updates of the geometric parameters from normal regularization. We further propose a confidence term to mitigate inconsistencies of normal predictions across multiple views. Moreover, we also introduce a densification and splitting strategy to regularize the size and distribution of 3D Gaussians for more accurate surface modeling. Compared with Gaussian-based baselines, experiments show that our approach obtains better reconstruction quality and maintains competitive appearance quality at faster training speed and 100+ FPS rendering. Our code will be made open-source upon paper acceptance.
Abstract:Empowering 3D Gaussian Splatting with generalization ability is appealing. However, existing generalizable 3D Gaussian Splatting methods are largely confined to narrow-range interpolation between stereo images due to their heavy backbones, thus lacking the ability to accurately localize 3D Gaussian and support free-view synthesis across wide view range. In this paper, we present a novel framework FreeSplat that is capable of reconstructing geometrically consistent 3D scenes from long sequence input towards free-view synthesis.Specifically, we firstly introduce Low-cost Cross-View Aggregation achieved by constructing adaptive cost volumes among nearby views and aggregating features using a multi-scale structure. Subsequently, we present the Pixel-wise Triplet Fusion to eliminate redundancy of 3D Gaussians in overlapping view regions and to aggregate features observed across multiple views. Additionally, we propose a simple but effective free-view training strategy that ensures robust view synthesis across broader view range regardless of the number of views. Our empirical results demonstrate state-of-the-art novel view synthesis peformances in both novel view rendered color maps quality and depth maps accuracy across different numbers of input views. We also show that FreeSplat performs inference more efficiently and can effectively reduce redundant Gaussians, offering the possibility of feed-forward large scene reconstruction without depth priors.
Abstract:The recent advances in 3D Gaussian Splatting (3DGS) show promising results on the novel view synthesis (NVS) task. With its superior rendering performance and high-fidelity rendering quality, 3DGS is excelling at its previous NeRF counterparts. The most recent 3DGS method focuses either on improving the instability of rendering efficiency or reducing the model size. On the other hand, the training efficiency of 3DGS on large-scale scenes has not gained much attention. In this work, we propose DoGaussian, a method that trains 3DGS distributedly. Our method first decomposes a scene into K blocks and then introduces the Alternating Direction Method of Multipliers (ADMM) into the training procedure of 3DGS. During training, our DoGaussian maintains one global 3DGS model on the master node and K local 3DGS models on the slave nodes. The K local 3DGS models are dropped after training and we only query the global 3DGS model during inference. The training time is reduced by scene decomposition, and the training convergence and stability are guaranteed through the consensus on the shared 3D Gaussians. Our method accelerates the training of 3DGS by 6+ times when evaluated on large-scale scenes while concurrently achieving state-of-the-art rendering quality. Our project page is available at https://aibluefisher.github.io/DoGaussian.
Abstract:In this paper, we introduce X-Ray, an innovative approach to 3D generation that employs a new sequential representation, drawing inspiration from the depth-revealing capabilities of X-Ray scans to meticulously capture both the external and internal features of objects. Central to our method is the utilization of ray casting techniques originating from the camera's viewpoint, meticulously recording the geometric and textural details encountered across all intersected surfaces. This process efficiently condenses complete objects or scenes into a multi-frame format, just like videos. Such a structure ensures the 3D representation is composed solely of critical surface information. Highlighting the practicality and adaptability of our X-Ray representation, we showcase its utility in synthesizing 3D objects, employing a network architecture akin to that used in video diffusion models. The outcomes reveal our representation's superior performance in enhancing both the accuracy and efficiency of 3D synthesis, heralding new directions for ongoing research and practical implementations in the field.
Abstract:Existing multi-person human reconstruction approaches mainly focus on recovering accurate poses or avoiding penetration, but overlook the modeling of close interactions. In this work, we tackle the task of reconstructing closely interactive humans from a monocular video. The main challenge of this task comes from insufficient visual information caused by depth ambiguity and severe inter-person occlusion. In view of this, we propose to leverage knowledge from proxemic behavior and physics to compensate the lack of visual information. This is based on the observation that human interaction has specific patterns following the social proxemics. Specifically, we first design a latent representation based on Vector Quantised-Variational AutoEncoder (VQ-VAE) to model human interaction. A proxemics and physics guided diffusion model is then introduced to denoise the initial distribution. We design the diffusion model as dual branch with each branch representing one individual such that the interaction can be modeled via cross attention. With the learned priors of VQ-VAE and physical constraint as the additional information, our proposed approach is capable of estimating accurate poses that are also proxemics and physics plausible. Experimental results on Hi4D, 3DPW, and CHI3D demonstrate that our method outperforms existing approaches. The code is available at \url{https://github.com/boycehbz/HumanInteraction}.
Abstract:3D point cloud completion is designed to recover complete shapes from partially observed point clouds. Conventional completion methods typically depend on extensive point cloud data for training %, with their effectiveness often constrained to object categories similar to those seen during training. In contrast, we propose a zero-shot framework aimed at completing partially observed point clouds across any unseen categories. Leveraging point rendering via Gaussian Splatting, we develop techniques of Point Cloud Colorization and Zero-shot Fractal Completion that utilize 2D priors from pre-trained diffusion models to infer missing regions. Experimental results on both synthetic and real-world scanned point clouds demonstrate that our approach outperforms existing methods in completing a variety of objects without any requirement for specific training data.
Abstract:In this paper, we investigate Open-Vocabulary 3D Instance Segmentation (OV-3DIS) with free-form language instructions. Earlier works that rely on only annotated base categories for training suffer from limited generalization to unseen novel categories. Recent works mitigate poor generalizability to novel categories by generating class-agnostic masks or projecting generalized masks from 2D to 3D, but disregard semantic or geometry information, leading to sub-optimal performance. Instead, generating generalizable but semantic-related masks directly from 3D point clouds would result in superior outcomes. In this paper, we introduce Segment any 3D Object with LanguagE (SOLE), which is a semantic and geometric-aware visual-language learning framework with strong generalizability by generating semantic-related masks directly from 3D point clouds. Specifically, we propose a multimodal fusion network to incorporate multimodal semantics in both backbone and decoder. In addition, to align the 3D segmentation model with various language instructions and enhance the mask quality, we introduce three types of multimodal associations as supervision. Our SOLE outperforms previous methods by a large margin on ScanNetv2, ScanNet200, and Replica benchmarks, and the results are even close to the fully-supervised counterpart despite the absence of class annotations in the training. Furthermore, extensive qualitative results demonstrate the versatility of our SOLE to language instructions.