温思嘉/Wen Sijia

Email: sijiawen@buaa.edu.cn

About Me

Sijia Wen received a Doctoral degree from the State Key Laboratory of Virtual Reality Technology and Systems, School of Computer Science and Engineering, Beihang University in 2022. He is currently an assistant professor at the School of Artificial Intelligence, at Beihang University. His research interests include computer vision, multi-source computational photography, and physical-based 3D vision.

Publications

High-Fidelity Polarimetric Implicit 3D Reconstruction with View-Dependent Physical Representation

We propose a polarimetric implicit 3D reconstruction method that integrates geometric and polarization data to generate high-quality meshes in complex scenes. A view-dependent physical representation analyzes subtle reflection properties, enhanced by a simple detection algorithm and optimized with ray tracing and polarization principles.
Yu Qiu, Sijia Wen#, Hainan Zhang, Zhiming Zheng    The 39th Annual AAAI Conference on Artificial Intelligence (AAAI), 2025.
[PDF][[Project]][[Dataset]]

Polarization State Attention Dehazing Network with a Simulated Polar-Haze Dataset

We propose a haze simulation strategy using polarization cues to generate realistic hazy data. Using the Polar-Haze dataset, our polarization state attention dehazing network (PSADNet) combines an attention-based extraction module to capture key features and a dehazing module that enhances clarity while preserving polarization accuracy.
Sijia Wen, Yinqiang Zheng, Feng Lu    IEEE Transactions on Multimedia, 2024.
[PDF][[Project]][[Dataset]]

A Cross-Consistency Strategy for Clearer Perception in Low-Light Haze

We introduce two key techniques: a cross-consistency dehazing-enhancement framework and a physically based simulation for generating a low-light hazy dataset. The framework improves image visibility by leveraging information from multiple sub-tasks, while the simulation creates ground truth datasets using a low-light hazy imaging model.
Sijia Wen, Chaoqun Zhuang, Yunfei Liu, Feng Lu    Chinese Conference on Pattern Recognition and Computer Vision (PRCV) Poster, 2024.
[PDF][[Project]][[Dataset]]

3D Photography with One-shot Portrait Relighting

In this paper, we present a physical-based framework that explicitly models 3D photography with relighting from a one-shot portrait. By extracting facial albedo and rendering realistic reflected light, our method is capable of generating novel views with different lighting conditions, which can faithfully deliver realistic rendered results.
Yunfei Liu*, Sijia Wen*, Feng Lu    IEEE International Symposium on Mixed and Augmented Reality (ISMAR) Poster, 2021.
[PDF][[Project]][[Dataset]]

Polarization Guided Specular Reflection Separation

In this paper, we derive a polarization guided model to incorporate the polarization information into a designed iteration optimization separation strategy to separate the specular reflection. The proposed polarization guided model can generate a polarization chromaticity image, which is able to reveal the intrinsic geometrical profile of the input image.
Sijia Wen, Yinqiang Zheng, Feng Lu    IEEE Transactions on Image Processing (TIP), 30: 7280-7291, 2021.
[PDF][Project][Dataset]

A Sparse Representation Based Joint Demosaicing Method for Single-Chip Polarized Color Sensor

For newly released single-chip polarized color sensor, we present a sparse representation-based optimization strategy that utilizes chromatic information and polarimetric information to jointly reconstruct the information. In addition, we build an optical data acquisition system to collect a dataset, which contains various sources of polarization.
Sijia Wen, Yinqiang Zheng, Feng Lu    IEEE Transactions on Image Processing (TIP), 30: 4171-4182, 2021.
[PDF][Project][Dataset]

Convolutional Demosaicing Network for Joint Chromatic and Polarimetric Imagery

In this letter, we collect the first chromatic-polarization dataset and propose a Chromatic-Polarization Demosaicing Network (CPDNet) to address this joint chromatic and polarimetric image demosaicing issue. The proposed CPDNet is composed of the residual block and the multi-task structure with the costumed loss function.
Sijia Wen, Yinqiang Zheng, Feng Lu, Qinping Zhao    Optics Letters, 44(22), 5646-5649, 2019.
[PDF][Project][Dataset]

Adaptive Synthesis of Indoor Scenes via Activity-Associated Object Relation Graphs

Qiang Fu, Xiaowu Chen, Xiaotian Wang, Sijia Wen, et al.
ACM Transactions on Graphics, 2017, 36(6CD):201.1-201.13.

[PDF]