LiveMoments: Reselected Key Photo Restoration in Live Photos via Reference-guided Diffusion

Clara Xue, Zizheng Yan, Zhenning Shi, Yuhang Yu, Jingyu Zhuang, Qi Zhang, Jinwei Chen, Qingnan Fan.

• The first to address the problem of reselected key photo restoration in Live Photos.
• LiveMoments significantly improves perceptual quality and fidelity over existing solutions, including the recent flagships from vivo and iPhone.

BokehDiff: Neural Lens Blur with One-Step Diffusion

Chengxuan Zhu, Qingnan Fan, Qi Zhang, Jinwei Chen, Huaqi Zhang, Chao Xu, Boxin Shi.

arXiv / codes

• The first neural lens blur rendering pipeline based on pretrained diffusion priors.
• A diffusion framework with only one inference step that achieves outstanding quality compared with previous methods, especially in regions where depth prediction fails.

TSD-SR: One-Step Diffusion with Target Score Distillation for Real-World Image Super-Resolution

Linwei Dong*, Qingnan Fan*, Yihong Guo, Zhonghao Wang, Qi Zhang, Jinwei Chen, Yawei Luo, Changqing Zou.

arXiv / codes

• The first image super-resolution work that leverages the pretrained diffusion transformer (DIT) prior, specifically Stable Diffusion 3.
• TSD-SR has superior restoration results (most of the metrics perform the best) and the fastest inference speed (e.g. 40 times faster than SeeSR) compared to the past Real-ISR approaches based on pre-trained diffusion priors.

Text-Aware Real-World Image Super-Resolution via Diffusion Model with Joint Segmentation Decoders

Qiming Hu, Linlong Fan, Yiyan Luo, Yuhang Yu, Xiaojie Guo, Qingnan Fan.

arXiv / codes

• The first full-image text super-resolution work utilizing the diffusion priors.
• A novel diffusion-based SR framework, which integrates text-aware attention and joint segmentation decoders to recover not only natural details but also the structural fidelity of text regions in degraded real-world images.

AuthFace: Towards Authentic Blind Face Restoration with Face-oriented Generative Diffusion Prior

Guoqiang Liang, Qingnan Fan, Bingtao Fu, Jinwei Chen, Hong Gu, Lin Wang.

arXiv / codes

• Pioneering a new approach by enhancing the generative capabilities of pretrained T2I models for authentic face restoration, moving beyond traditional model design.
• A novel framework, namely AuthFace that achieves highly authentic face restoration results by exploring a face-oriented generative diffusion prior.

SLAM3R: Real-Time Dense Scene Reconstruction from Monocular RGB Videos

Yuzheng Liu*, Siyan Dong*, Shuzhe Wang, Yingda Yin, Yanchao Yang, Qingnan Fan, Baoquan Chen.

arXiv / video / codes

• Award: China3DV 2025, Top1 paper.
• SLAM3R is a real-time dense scene reconstruction system that regresses 3D points from video frames using feed-forward neural networks, without explicitly estimating camera parameters.

Scene-aware Activity Program Generation with Language Guidance

Zejia Su, Qingnan Fan, Xuelin Chen, Oliver van Kaick, Hui Huang, Ruizhen Hu.

project page / supp file / bibtex

• We address the problem of scene-aware activity program generation, which requires decomposing a given activity task into instructions that can be sequentially performed within a target scene to complete the activity.

C·ASE: Learning Conditional Adversarial Skill Embeddings for Physics-based Characters

Zhiyang Dou, Xuelin Chen, Qingnan Fan, Taku Komura, Wenping Wang.

arXiv / project page / video / bibtex

• We present C·ASE, an efficient and effective framework that learns conditional Adversarial Skill Embeddings for Elite physics-based characters.

VAT-Mart: Learning Visual Action Trajectory Proposals for Manipulating 3D ARTiculated Objects

Ruihai Wu*, Yan Zhao*, Kaichun Mo*, Zizheng Guo, Yian Wang, Tianhao Wu, Qingnan Fan, Xuelin Chen, Leonidas Guibas, Hao Dong.

arXiv / project page / codes / video / bibtex

• Award: WAIC 2025, Young Outstanding Paper Award
• We design an interaction-for-perception framework, VAT-MART, to learn actionable visual representations for more effective manipulation of 3D articulated objects.

ADeLA: Automatic Dense Labeling with Attention for Viewpoint Shift in Semantic Segmentation

Yanchao Yang* , Hanxiang Ren*, He Wang,
Bokui Shen, Qingnan Fan, Youyi Zheng, C. Karen Liu, Leonidas Guibas.

arXiv / codes / bibtex

• We describe a method to deal with performance drop in semantic segmentation caused by viewpoint changes within multi-camera systems, where temporally paired images are readily available, but the annotations may only be abundant for a few typical views.

CAPTRA: CAtegory-level Pose Tracking for Rigid and Articulated Objects from Point Clouds

Yijia Weng*, He Wang*, Qiang Zhou, Yuzhe Qin, Yueqi Duan, Qingnan Fan, Baoquan Chen, Hao Su, Leonidas Guibas.

arXiv / project page / codes / video / bibtex

• For the first time, we propose a unified framework that can handle 9-DoF pose tracking for novel rigid object instances as well as per-part pose tracking for 3D articulated objects.

Robust Neural Routing Through Space Partitions for Camera Relocalization in Dynamic Indoor Environments

Siyan Dong*, Qingnan Fan*, He Wang, Ji Shi, Li Yi, Thomas Funkhouser, Baoquan Chen, Leonidas Guibas.

arXiv / codes / video / bibtex

• A novel outlier-aware neural tree to tackle the camera localization challenges in dynamic indoor environments. It achieves the best performance in the RIO-10 benchmark.

Image Smoothing via Unsupervised Learning

Qingnan Fan, Jiaolong Yang, David Wipf, Baoquan Chen, Xin Tong.

arXiv / codes / supp file / bibtex

• Treat deep learning as an optimization tool to minimize the proposed image smoothing objective function in an unsupervised manner. Multiple different smoothing effects can be easily learned by adaptively changing the proposed objective function.

Revisiting Deep Intrinsic Image Decompositions

Qingnan Fan, Jiaolong Yang, Gang Hua, Baoquan Chen, David Wipf.

arXiv / codes / slides / supp file / poster / bibtex

• The first demonstration of a single basic deep architecture capable of achieving state-of-the-art results when applied to each of the major intrinsic benchmarks.

JumpCut: Non-Successive Mask Transfer and Interpolation for Video Cutout

Qingnan Fan, Fan Zhong, Dani Lischinski, Daniel Cohen-Or, Baoquan Chen.

codes / slides / video / supp file / dataset / bibtex

• An interactive real-time video segmentation algorithm. Significantly improve the video cutout accuracy and efficiency.

Build-to-Last: Strength to Weight 3D Printed Objects

Lin Lu, Andrei Sharf, Haisen Zhao, Yuan Wei, Qingnan Fan, Xuelin Chen, Yann Savoye, Changhe Tu, Daniel Cohen-Or, Baoquan Chen.

video / bibtex

• Reduce the material cost and weight of a given object while providing a durable printed model that is resistant to impact and external forces.