I am currently a Ph.D. at South China University of Technology (SCUT), advised by Prof. Kui Jia. I received my bachelor degree from the same university (i.e. SCUT) in 2020, and expect to obtain my Ph.D. degree in 2025. I am currently an intern at Tencent AI Lab.

I mainly focus on 3D Computer Vision. My current research interests include Computer Graphics, 3D Semantic Learning and Reconstruction. Recently, I am working on Multi-View Reconstruction.

📝 Publications

ECCV 2024 (Oral)
sym

Analytic-Splatting: Anti-Aliased 3D Gaussian Splatting via Analytic Integration

Zhihao Liang, Qi Zhang, Wenbo Hu, Lei Zhu, Ying Feng, Kui Jia

Project | Code

  • In this paper, we derive an analytical solution to address the aliasing caused by discrete sampling in 3DGS. More specifically, we use a conditioned logistic function as the analytic approximation of the cumulative distribution function (CDF) in a one-dimensional Gaussian signal and calculate the Gaussian integral by subtracting the CDFs. We then introduce this approximation in the two-dimensional pixel shading, and present Analytic-Splatting, which analytically approximates the Gaussian integral within the 2D-pixel window area to better capture the intensity response of each pixel. Moreover, we use the approximated response of the pixel window integral area to participate in the transmittance calculation of volume rendering, making Analytic-Splatting sensitive to the changes in pixel footprint at different resolutions.
Arxiv 2024
sym

GS-ID: Illumination Decomposition on Gaussian Splatting via Diffusion Prior and Parametric Light Source Optimization

Kang Du, Zhihao Liang, Zeyu Wang

Project | Code

  • We present GS-ID, a novel framework for illumination decomposition on Gaussian Splatting, achieving photorealistic novel view synthesis and intuitive light editing. In this work, we first introduce intrinsic diffusion priors to estimate the attributes for physically based rendering. Then we divide the illumination into environmental and direct components for joint optimization. Last, we employ deferred rendering to reduce the computational load. Our framework uses a learnable environment map and Spherical Gaussians (SGs) to represent light sources parametrically, therefore enabling controllable and photorealistic relighting on Gaussian Splatting.
arXiv 2024
sym

Advances in 3D Generation: A Survey

Xiaoyu Li, Qi Zhang, Di Kang, Weihao Cheng, Yiming Gao, Jingbo Zhang, Zhihao Liang, Jing Liao, Yan-Pei Cao, Ying Shan

Project

  • In this survey, we aim to introduce the fundamental methodologies of 3D generation methods and establish a structured roadmap, encompassing 3D representation, generation methods, datasets, and corresponding applications. We hope this survey will help readers explore this exciting topic and foster further advancements in the field of 3D content generation.
ECCV 2024
sym

Sur2f: A Hybrid Representation for High-Quality and Efficient Surface Reconstruction from Multi-view Images

Zhangjin Huang*, Zhihao Liang*, Haojie Zhang, Yangkai Lin, Kui Jia

Project | Code

  • We propose a new hybrid representation, termed Sur2f, that can enjoy the benefits of both explicit and implicit surface representations. This is achieved by learning two parallel streams of an implicit SDF and an explicit surrogate surface mesh, both of which, by rendering, receive supervision from multi-view image observations.
CVPR 2024
sym

GS-IR: 3D Gaussian Splatting for Inverse Rendering

Zhihao Liang*, Qi Zhang*, Ying Feng, Ying Shan, Kui Jia

Project | Code

  • We present GS-IR that models a scene as a set of 3D Gaussians to achieve physically-based rendering and state-ofthe-art decomposition results for both objects and scenes.
CVPR 2023
sym

HelixSurf: A Robust and Efficient Neural Implicit Surface Learning of Indoor Scenes with Iterative Intertwined Regularization

Zhihao Liang*, Zhangjin Huang*, Changxing Ding, Kui Jia

Project | Code

  • We present a novel method of HelixSurf for reconstruction of indoor scene surface from multi-view images. HelixSurf enjoys the complementary benefits of the traditional MVS and the recent neural implicit surface learning, by regularizing the learning/optimization of one strategy iteratively using the intermediate prediction from the other.
CVPR 2022
sym

VISTA: Boosting 3D Object Detection via Dual Cross-VIew SpaTial Attention

Shengheng Deng*, Zhihao Liang*, Lin Sun, Kui Jia

Project | Code

  • We propose a novel plug-and-play fusion module Dual Cross-VIew SpaTial Attention (VISTA) to produce well-fused multi-view features to boost the performances of 3D object detector. Our proposed VISTA replaces the MLPs with convolutional operators, which is capable of better handling the local cues for attention modeling.
ICCV 2021
sym

Instance Segmentation in 3D Scenes using Semantic Superpoint Tree Networks

Zhihao Liang, Zhihao Li, Songcen Xu, Mingkui Tan, Kui Jia

Project | Code

  • We propose an end-to-end solution of Semantic Superpoint Tree Network (SSTNet) to directly propose and evaluate object instances from observed 3D scenes. By working with superpoints, our method enjoys the benefit of geometric regularity that supports consistent and sharp segmentations, especially at object boundaries.
CVPR 2020
sym

Fgn: Fully guided network for few-shot instance segmentation

Zhibo Fan, Jin-Gang Yu, Zhihao Liang, Jiarong Ou, Changxin Gao, Gui-Song Xia, Yuanqing Li

Project

  • We propose the Fully Guided Network, a novel framework for few-shot instance segmentation.

📖 Educations

  • 2021.09 - Present, Ph.D., South China University of Technology, Guangzhou.
  • 2020.09 - 2021.04, Master, South China University of Technology, Guangzhou.
  • 2016.09 - 2020.06, Undergraduate, South China University of Technology, Guangzhou.