Hujun Bao’s research while affiliated with Zhejiang University and other places

This paper proposes a concise, elegant, and robust pipeline to estimate smooth camera trajectories and obtain dense point clouds for casual videos in the wild. Traditional frameworks, such as ParticleSfM~\cite{zhao2022particlesfm}, address this problem by sequentially computing the optical flow between adjacent frames to obtain point trajectories. They then remove dynamic trajectories through motion segmentation and perform global bundle adjustment. However, the process of estimating optical flow between two adjacent frames and chaining the matches can introduce cumulative errors. Additionally, motion segmentation combined with single-view depth estimation often faces challenges related to scale ambiguity. To tackle these challenges, we propose a dynamic-aware tracking any point (DATAP) method that leverages consistent video depth and point tracking. Specifically, our DATAP addresses these issues by estimating dense point tracking across the video sequence and predicting the visibility and dynamics of each point. By incorporating the consistent video depth prior, the performance of motion segmentation is enhanced. With the integration of DATAP, it becomes possible to estimate and optimize all camera poses simultaneously by performing global bundle adjustments for point tracking classified as static and visible, rather than relying on incremental camera registration. Extensive experiments on dynamic sequences, e.g., Sintel and TUM RGBD dynamic sequences, and on the wild video, e.g., DAVIS, demonstrate that the proposed method achieves state-of-the-art performance in terms of camera pose estimation even in complex dynamic challenge scenes.

High-quality real-time rendering using user-affordable capture rigs is an essential property of human performance capture systems for real-world applications. However, state-of-the-art performance capture methods may not yield satisfactory rendering results under a very sparse (e.g., four) capture setting. Specifically, neural radiance field (NeRF)-based methods and 3D Gaussian Splatting (3DGS)-based methods tend to produce local geometry errors for unseen performers, while occupancy field (PIFu)-based methods often produce unrealistic rendering results. In this paper, we propose a novel generalizable neural approach to reconstruct and render the performers from very sparse RGBD streams in high quality. The core of our method is a novel point-based generalizable human (PGH) representation conditioned on the pixel-aligned RGBD features. The PGH representation learns a surface implicit function for the regression of surface points and a Gaussian implicit function for parameterizing the radiance fields of the regressed surface points with 2D Gaussian surfels, and uses surfel splatting for fast rendering. We learn this hybrid human representation via two novel networks. First, we propose a novel point-regressing network (PRNet) with a depth-guided point cloud initialization (DPI) method to regress an accurate surface point cloud based on the denoised depth information. Second, we propose a novel neural blending-based surfel splatting network (SPNet) to render high-quality geometries and appearances in novel views based on the regressed surface points and high-resolution RGBD features of adjacent views. Our method produces free-view human performance videos of 1K resolution at 12 fps on average. Experiments on two benchmarks show that our method outperforms state-of-the-art human performance capture methods.

This paper aims to address the challenge of reconstructing long volumetric videos from multi-view RGB videos. Recent dynamic view synthesis methods leverage powerful 4D representations, like feature grids or point cloud sequences, to achieve high-quality rendering results. However, they are typically limited to short (1~2s) video clips and often suffer from large memory footprints when dealing with longer videos. To solve this issue, we propose a novel 4D representation, named Temporal Gaussian Hierarchy, to compactly model long volumetric videos. Our key observation is that there are generally various degrees of temporal redundancy in dynamic scenes, which consist of areas changing at different speeds. Motivated by this, our approach builds a multi-level hierarchy of 4D Gaussian primitives, where each level separately describes scene regions with different degrees of content change, and adaptively shares Gaussian primitives to represent unchanged scene content over different temporal segments, thus effectively reducing the number of Gaussian primitives. In addition, the tree-like structure of the Gaussian hierarchy allows us to efficiently represent the scene at a particular moment with a subset of Gaussian primitives, leading to nearly constant GPU memory usage during the training or rendering regardless of the video length. Moreover, we design a Compact Appearance Model that mixes diffuse and view-dependent Gaussians to further minimize the model size while maintaining the rendering quality. We also develop a rasterization pipeline of Gaussian primitives based on the hardware-accelerated technique to improve rendering speed. Extensive experimental results demonstrate the superiority of our method over alternative methods in terms of training cost, rendering speed, and storage usage. To our knowledge, this work is the first approach capable of efficiently handling hours of volumetric video data while maintaining state-of-the-art rendering quality.

Unbiased Monte Carlo path tracing that is extensively used in realistic rendering produces undesirable noise, especially with low samples per pixel (spp). Recently, several methods have coped with this problem by importing unbiased noisy images and auxiliary features to neural networks to either predict a fixed-sized kernel for convolution or directly predict the denoised result. Since it is impossible to produce arbitrarily high spp images as the training dataset, the network-based denoising fails to produce high-quality images under high spp. More specifically, network-based denoising is inconsistent and does not converge to the ground truth as the sampling rate increases. On the other hand, the post-correction estimators yield a blending coefficient for a pair of biased and unbiased images influenced by image errors or variances to ensure the consistency of the denoised image. As the sampling rate increases, the blending coefficient of the unbiased image converges to 1, that is, using the unbiased image as the denoised results. However, these estimators usually produce artifacts due to the difficulty of accurately predicting image errors or variances with low spp. To address the above problems, we take advantage of both kernel-predicting methods and post-correction denoisers. A novel kernel-based denoiser is proposed based on distribution-free kernel regression consistency theory, which does not explicitly combine the biased and unbiased results but constrain the kernel bandwidth to produce consistent results under high spp. Meanwhile, our kernel regression method explores bandwidth optimization in the robust auxiliary feature space instead of the noisy image space. This leads to consistent high-quality denoising at both low and high spp. Experiment results demonstrate that our method outperforms existing denoisers in accuracy and consistency.

In this paper, we propose a novel multi-view stereo (MVS) framework that gets rid of the depth range prior. Unlike recent prior-free MVS methods that work in a pair-wise manner, our method simultaneously considers all the source images. Specifically, we introduce a Multi-view Disparity Attention (MDA) module to aggregate long-range context information within and across multi-view images. Considering the asymmetry of the epipolar disparity flow, the key to our method lies in accurately modeling multi-view geometric constraints. We integrate pose embedding to encapsulate information such as multi-view camera poses, providing implicit geometric constraints for multi-view disparity feature fusion dominated by attention. Additionally, we construct corresponding hidden states for each source image due to significant differences in the observation quality of the same pixel in the reference frame across multiple source frames. We explicitly estimate the quality of the current pixel corresponding to sampled points on the epipolar line of the source image and dynamically update hidden states through the uncertainty estimation module. Extensive results on the DTU dataset and Tanks&Temple benchmark demonstrate the effectiveness of our method. The code is available at our project page: https://zju3dv.github.io/GD-PoseMVS/.