3DV 2025 Papers

This paper presents a novel framework for compactly representing a 3D indoor scene using a set of polycuboids through a deep learning-based fitting method. Indoor scenes mainly consist of man-made objects, such as furniture, which often exhibit rectilinear geometry. This property allows indoor scenes to be represented using combinations of polycuboids, providing a compact representation that benefits downstream applications like furniture rearrangement. Our framework takes a noisy point cloud as input and first detects six types of cuboid faces using a transformer network. Then, a graph neural network is used to validate the spatial relationships of the detected faces to form potential polycuboids. Finally, each polycuboid instance is reconstructed by forming a set of boxes based on the aggregated face labels. To train our networks, we introduce a synthetic dataset encompassing a diverse range of cuboid and polycuboid shapes that reflect the characteristics of indoor scenes. Our framework generalizes well to real-world indoor scene datasets, including Replica, ScanNet, and scenes captured with an iPhone. The versatility of our method is demonstrated through practical applications, such as virtual room tours and scene editing.

We present a method for reconstructing a clear Neural Radiance Field (NeRF) even with fast camera motions. To address blur artifacts, we leverage both (blurry) RGB images and event camera data captured in a binocular configuration. Importantly, when reconstructing our clear NeRF, we consider the camera modeling imperfections that arise from the simple pinhole camera model as learned embeddings for each camera measurement, and further learn a mapper that connects event camera measurements with RGB data. As no previous dataset exists for our binocular setting, we introduce an event camera dataset with captures from a 3D-printed stereo configuration between RGB and event cameras. Empirically, we evaluate on our introduced dataset and EVIMOv2 and show that our method leads to improved reconstructions. We are committed to making our code and dataset public.

We introduce SAMPro3D for zero-shot instance segmentation of 3D scenes. Given the 3D point cloud and multiple posed RGB-D frames of 3D scenes, our approach segments 3D instances by applying the pretrained Segment Anything Model (SAM) to 2D frames. Our key idea involves locating SAM prompts in 3D to align their projected pixel prompts across frames, ensuring the view consistency of SAM-predicted masks. Moreover, we suggest selecting prompts from the initial set guided by the information of SAM-predicted masks across all views, which enhances the overall performance. We further propose to consolidate different prompts if they are segmenting different surface parts of the same 3D instance, bringing a more comprehensive segmentation. Notably, our method does **not** require any additional training. Extensive experiments on diverse benchmarks show that our method achieves comparable or better performance compared to previous zero-shot or fully supervised approaches, and in many cases surpasses human annotations. Furthermore, since our fine-grained predictions often lack annotations in available datasets, we present ScanNet200-Fine50 test data which provides more fine-grained annotations on 50 scenes from ScanNet200 dataset.

This paper introduces a robust translation synchronization approach which takes relative directions between pairs of images as inputs and outputs absolute image translations. Our approach is based on a generalized eigenvalue problem, where the formulation contains edge weights in relative directions and vertex weights in absolute image translations. We present a rigorous stability analysis to determine how to set these weights optimally. Specifically, optimal vertex weights are always $1$, while optimal edge weights depend on magnitudes of relative transformations and variances of relative directions. These results lead to an iterative translation synchronization formulation, which progressively removes outliers in the inputs by adaptively adjusting the edge weights. We present exact and robust recovery conditions for our approach under a standard noise model. Experimental results justify our theoretical results and show that our approach outperforms state-of-the-art baseline approaches on both synthetic and real datasets.

The latest regularized Neural Radiance Field (NeRF) approaches produce poor geometry and view extrapolation for large scale sparse view scenes, such as ETH3D. Density-based approaches tend to be under-constrained, while surface-based approaches tend to miss details. In this paper, we take a density-based approach, sampling patches instead of individual rays to better incorporate monocular depth and normal estimates and patch-based photometric consistency constraints between training views and sampled virtual views. Loosely constraining densities based on estimated depth aligned to sparse points further improves geometric accuracy. While maintaining similar view synthesis quality, our approach significantly improves geometric accuracy on the ETH3D benchmark, e.g. increasing the F1@2cm score by 4x-8x compared to other regularized density-based approaches, with much lower training and inference time than other approaches.

Point clouds are a set of data points in space to represent the 3D geometry of objects. A fundamental step in the processing is to identify a subset of points to represent the shape. While traditional sampling methods often ignore to incorporate geometrical information, recent developments in learning-based sampling models have achieved significant levels of performance. With the integration of geometrical priors, the ability to learn and preserve the underlying structure can be enhanced when sampling. To shed light into the shape, a qualitative skeleton serves as an effective descriptor to guide sampling for both local and global geometries. In this paper, we introduce MorphoSkel3D as a new technique based on morphology to facilitate an efficient skeletonization of shapes. With its low computational cost, MorphoSkel3D is a unique, rule-based algorithm to benchmark its quality and performance on two large datasets, ModelNet and ShapeNet, under different sampling ratios. The results show that training with MorphoSkel3D leads to an informed and more accurate sampling in the practical application of object classification and point cloud retrieval.

We present a spatio-temporal perspective on category-agnostic 3D lifting of 2D keypoints over a temporal sequence. Our approach differs from existing state-of-the-art methods that are either: (i) object-agnostic, but can only operate on individual frames, or (ii) can model space-time dependencies, but are only designed to work with a single object category. Our approach is grounded in two core principles. First, general information about similar objects can be leveraged to achieve better performance when there is little object-specific training data. Second, a temporally-proximate context window is advantageous for achieving consistency throughout a sequence. These two principles allow us to outperform current state-of-the-art methods on per-frame and per-sequence metrics for a variety of animal categories. Lastly, we release a new synthetic dataset containing 3D skeletons and motion sequences for a variety of animal categories.

This paper introduces FlowMap, an end-to-end differentiable method that solves for precise camera poses, camera intrinsics, and per-frame dense depth of a video sequence. Our method performs per-video gradient-descent minimization of a simple least-squares objective that compares the optical flow induced by depth, intrinsics, and poses against correspondences obtained via off-the-shelf optical flow and point tracking. Alongside the use of point tracks to encourage long-term geometric consistency, we introduce a differentiable re-parameterization of depth, intrinsics, and pose that is amenable to first-order optimization. We empirically show that camera parameters and dense depth recovered by our method enable photo-realistic novel view synthesis on 360° trajectories using Gaussian Splatting. Our method not only far outperforms prior gradient-descent based bundle adjustment methods, but surprisingly performs on par with COLMAP, the state-of-the-art SfM method, on the downstream task of 360° novel view synthesis - even though our method is purely gradient-descent based, fully differentiable, and presents a complete departure from conventional SfM. Our result opens the door to the self-supervised training of neural networks that perform camera parameter estimation, 3D reconstruction, and novel view synthesis.

3D Gaussian Splatting (3D-GS) has recently attracted great attention with real-time and photo-realistic renderings. This technique typically takes perspective images as input and optimizes a set of 3D elliptical Gaussians by splatting them onto the image planes, resulting in 2D Gaussians. However, applying 3D-GS to panoramic inputs presents challenges in effectively modeling the projection onto the spherical surface of ${360^\circ}$ images using 2D Gaussians. In practical applications, input panoramas are often sparse, leading to unreliable initialization of 3D Gaussians and subsequent degradation of 3D-GS quality. In addition, due to the under-constrained geometry of texture-less planes (e.g., walls and floors), 3D-GS struggles to model these flat regions with elliptical Gaussians, resulting in significant floaters in novel views. To address these issues, we propose 360-GS, a novel layout-guided $360^{\circ}$ Gaussian splatting for a limited set of panoramic inputs. Instead of splatting 3D Gaussians directly onto the spherical surface, 360-GS projects them onto the tangent plane of the unit sphere and then maps them to the spherical projections. This adaptation enables the representation of the projection using Gaussians. We guide the optimization of 3D Gaussians by exploiting layout priors within panoramas, which are simple to obtain and contain strong structural information about the indoor scene. Our experimental results demonstrate that 360-GS allows panoramic rendering and outperforms state-of-the-art methods with fewer artifacts in novel view synthesis, thus providing immersive roaming in indoor scenarios.

Traditional 3D garment creation is labor-intensive, involving sketching, modeling, UV mapping, and texturing, which are time-consuming and costly. Recent advances in diffusion-based generative models have enabled new possibilities for 3D garment generation from text prompts, images, and videos. However, existing methods either suffer from inconsistencies among multi-view images or require additional processes to separate cloth from the underlying human model. In this paper, we propose GarmentDreamer, a novel method that leverages 3D Gaussian Splatting (GS) as guidance to generate wearable, simulation-ready 3D garment meshes from text prompts. In contrast to using multi-view images directly predicted by generative models as guidance, our 3DGS guidance ensures consistent optimization in both garment deformation and texture synthesis. Our method introduces a novel garment augmentation module, guided by normal and RGBA information, and employs implicit Neural Texture Fields (NeTF) combined with Variational Score Distillation (VSD) to generate diverse geometric and texture details. We validate the effectiveness of our approach through comprehensive qualitative and quantitative experiments, showcasing the superior performance of GarmentDreamer over state-of-the-art alternatives.

We present Rig3DGS, a novel technique for creating reanimatable 3D portraits from short monocular smartphone videos. Rig3DGS learns to reconstruct a set of controllable 3D gaussians from a monocular video of a dynamic subject captured with varying head poses and facial expressions in an in-the-wild scene. In contrast to synchronized multi-view studio captures, this in-the-wild, single camera setup brings fresh challenges to learning high quality 3D gaussians. We address these challenges by learning to deform 3D gaussians from a fixed canonical space to the deformed space that is consistent with the target facial expression and head-pose. Our key contribution is a carefully designed deformation model that is guided by a 3D face morphable model. This deformation not only enables control over facial expression and head-poses but also allows our method to generates high quality photorealistic renders of the whole scene. Once trained, Rig3DGS is able to generate photorealistic renders of a subject and their scene for novel facial expression, head-poses, and viewing directions. Through extensive experiments we demonstrate that Rig3DGS significantly outperforms prior art while being orders of magnitude faster.

Cross-modal place recognition methods are flexible GPS-alternatives under varying environment conditions and sensor setups. However, this task is non-trivial since extracting consistent and robust global descriptors from different modalities is challenging. To tackle this issue, we propose \textit{Voxel-Cross-Pixel (VXP)}, a novel camera-to-LiDAR place recognition framework that enforces local similarities in a self-supervised manner and effectively brings global context from images and LiDAR scans into a shared feature space. Specifically, VXP is trained in three stages: first, we deploy a visual transformer to compactly represent input images. Secondly, we establish local correspondences between image-based and point cloud-based feature spaces using our novel geometric alignment module. We then aggregate local similarities into an expressive shared latent space. Extensive experiments on the three benchmarks (Oxford RobotCar, ViViD++ and KITTI) demonstrate that our method surpasses the state-of-the-art cross-modal retrieval by a large margin. Our evaluations show that the proposed method is accurate, efficient and light-weight. Our project page is available at: \href{https://yunjinli.github.io/projects-vxp/}{https://yunjinli.github.io/projects-vxp/}.

3D Gaussian Splatting (3DGS) has attracted great attention in novel view synthesis because of its superior rendering efficiency and high fidelity. However, the trained Gaussians suffer from severe zooming degradation due to non-adjustable representation derived from single-scale training. Though some methods attempt to tackle this problem via post-processing techniques such as selective rendering or filtering techniques towards primitives, the scale-specific information is not involved in Gaussians. In this paper, we propose a unified optimization method to make Gaussians adaptive for arbitrary scales by self-adjusting the primitive properties (e.g., color, shape and size) and distribution (e.g., position). Inspired by the mipmap technique, we design pseudo ground-truth for the target scale and propose a scale-consistency guidance loss to inject scale information into 3D Gaussians. Our method is a plug-in module, applicable for any 3DGS models to solve the zoom-in and zoom-out aliasing. Extensive experiments demonstrate the effectiveness of our method. Notably, our method outperforms 3DGS in PSNR by an average of 9.25 dB for zoom-in and 10.40 dB for zoom-out on the NeRF Synthetic dataset.

3D Gaussian Splatting (3DGS) has recently enabled real-time rendering of unbounded 3D scenes for novel view synthesis. However, this technique requires dense training views to accurately reconstruct 3D geometry. A limited number of input views will significantly degrade reconstruction quality, resulting in artifacts such as "floaters'' and "background collapse'' at unseen viewpoints. In this work, we introduce SparseGS, an efficient training pipeline designed to address the limitations of 3DGS in scenarios with sparse training views. SparseGS incorporates depth priors, novel depth rendering techniques, and a pruning heuristic to mitigate floater artifacts, alongside an Unseen Viewpoint Regularization module to alleviate background collapses. Our extensive evaluations on the Mip-NeRF360, LLFF, and DTU datasets demonstrate that SparseGS achieves high-quality reconstruction in both unbounded and forward-facing scenarios, with as few as 12 and 3 input images, respectively, while maintaining fast training and real-time rendering capabilities.

To evaluate end-to-end autonomous driving systems, a simulation environment based on Novel View Synthesis (NVS) techniques is essential, which synthesizes photo-realistic images and point clouds from previously recorded sequences under new vehicle poses, particularly in cross-lane scenarios. Therefore, the development of a multi-lane dataset and benchmark is necessary. While recent synthetic scene-based NVS datasets have been prepared for cross-lane benchmarking, they still lack the realism of captured images and point clouds. To further assess the performance of existing methods based on NeRF and 3DGS, we present the first multi-lane dataset registering parallel scans specifically for novel driving view synthesis dataset derived from real-world scans, comprising 25 groups of associated sequences, including 16,000 front-view images, 64,000 surround-view images, and 16,000 LiDAR frames. All frames are labeled to differentiate moving objects from static elements. Using this dataset, we evaluate the performance of existing approaches in various testing scenarios at different lanes and distances. Additionally, our method provides the solution for solving and assessing the quality of multi-sensor poses for multi-modal data alignment for curating such a dataset in real-world. We plan to continually add new sequences to test the generalization of existing methods across different scenarios. The dataset is released publicly at the project page: https://nizqleo.github.io/paralane-dataset/.

Test-Time Training (TTT) proposes to adapt a pre-trained network to changing data distributions on-the-fly. In this work, we propose the first TTT method for 3D semantic segmentation, TTT-KD, which models Knowledge Distillation (KD) from foundation models (e.g. DINOv2) as a self-supervised objective for adaptation to distribution shifts at test-time. Given access to paired image-pointcloud (2D-3D) data, we first optimize a 3D segmentation backbone for the main task of semantic segmentation using the pointclouds and the task of 2D → 3D KD by using an off-the-shelf 2D pre-trained foundation model. At test-time, our TTT-KD updates the 3D segmentation backbone for each test sample, by using the self-supervised task of knowledge distillation, before performing the final prediction. Extensive evaluations on multiple indoor and outdoor 3D segmentation benchmarks show the utility of TTT-KD, as it improves performance for both in-distribution (ID) and out-of-distribution (OOD) test datasets. We achieve a gain of up to 13 % mIoU (7 % on average) when the train and test distributions are similar and up to 45 % (20 % on average) when adapting to OOD test samples

Implicit neural representations (INRs) have become an important method for encoding various data types, such as 3D objects or scenes, images, and videos. They have proven to be particularly effective at representing 3D content, e.g., 3D scene reconstruction from 2D images, novel 3D content creation, as well as the representation, interpolation, and completion of 3D shapes. With the widespread generation of 3D data in an INR format, there is a need to support effective organization and retrieval of INRs saved in a data store. A key aspect of retrieval and clustering of INRs in a data store is the formulation of similarity between INRs that would, for example, enable retrieval of similar INRs using a query INR. In this work, we propose INRet, a method for determining similarity between INRs that represent shapes, thus enabling accurate retrieval of similar shape INRs from an INR data store. INRet flexibly supports different INR architectures such as INRs with octree grids, triplanes, and hash grids, as well as different implicit functions including signed/unsigned distance function and occupancy field. We demonstrate that our method is more general and accurate than the existing INR retrieval method, which only supports simple MLP INRs and requires the same architecture between the query and stored INRs. Furthermore, compared to converting INRs to other representations (e.g., point clouds or multi-view images) for 3D shape retrieval, INRet achieves higher accuracy while avoiding the conversion overhead.

We present a method to generate 3D objects in styles. Our method takes a text prompt and a style reference image as input and reconstructs a neural radiance field to synthesize a 3D model with the content aligning with the text prompt and the style following the reference image. To simultaneously generate the 3D object and perform style transfer in one go, we propose a stylized score distillation loss to guide a text-to-3D optimization process to output visually plausible geometry and appearance. Our stylized score distillation is based on a combination of an original pretrained text-to-image model and its modified sibling with the key and value features of self-attention layers manipulated to inject styles from the reference image. Comparisons with state-of-the-art methods demonstrated the strong visual performance of our method, further supported by the quantitative results from our user study.

Building a foundation model for 3D vision is a complex challenge that remains unsolved. Towards that goal, it is important to understand the 3D reasoning capabilities of current models as well as identify the gaps between these models and humans. Therefore, we construct a new 3D visual understanding benchmark named UniQA-3D. UniQA-3D covers fundamental 3D vision tasks in the Visual Question Answering (VQA) format. We evaluate state-of-the-art Vision-Language Models (VLMs), specialized models, and human subjects on it. Our results show that VLMs generally perform poorly, while the specialized models are accurate but not robust, failing under geometric perturbations. In contrast, human vision continues to be the most reliable 3D visual system. We further demonstrate that neural networks align more closely with human 3D vision mechanisms compared to classical computer vision methods, and Transformer-based networks such as ViT align more closely with human 3D vision mechanisms than CNNs. We hope our study will benefit the future development of foundation models for 3D vision. Code is available at https://github.com/princeton-vl/UniQA-3D.