Qingnan Fan (樊庆楠)

I am a Lead Researcher in the Imaging Algorithm Center of VIVO. Our group is the core algorithm team responsible for advancing the photographic quality in the flagship smartphones with the cutting-edge technologies (3D, AIGC, etc).

I was a Senior Researcher in the Visual Computing Center of Tencent AI Lab between 2021 to 2023.

I was a Postdoctoral Researcher in Stanford University supervised by Prof. Leonidas Guibas between 2019 to 2021.

I obtained my PhD degree in the Computer Science and Technology School of Shandong University at 2019. I was supervised by Prof. Baoquan Chen.

If you are interested in the internship in our group for either publishing academic papers or landing the latest technologies, feel free to drop me an email.

Email  /  CV  /  Biography  /  Google Scholar  /  LinkedIn  /  Twitter  /  Github

VIVO X200: BlueImage Photography Algorithm for Large-model Image Quality Enhancement

I am the host of BlueTalk, which invites the expert speakers in the fields of computer vision, graphics and computational photography. The mission of BlueTalk is to promote the communication between academic and industrial communities, and explore the possibilities of landing the most innovative research ideas into any AI-powered industrial applications. If you are insterested in joining this family to give a talk, feel free to let me know.

2024-12-09: Du Chen, PolyU, Real-world image super-resolution: solutions and challenges

2024-10-30: Mengwei Ren, Adobe, Relightful harmonization: lighting-aware portrait background replacement

2024-08-26: Jianqi Ma, PolyU, AI vision research based on text

2023-09-08: Xintao Wang, Tencent, Visual generation and editing via diffusion models

2023-08-15: Huan Ling, Nvidia, Generative models and vision perception using diffusion models

2023-07-25: Jianmin Bao, MSRA, 2D & 3D visual synthesis and manipulation via diffusion models

My research focus lies in computer graphics, 3D vision, image processing, and human-computer interaction. My recent effort has been spent on pushing the limit of 3D vision and reinforcement learning technologies to implement an intelligent embodied agent in both forms of physical robots and digital humans. For more recent research since 2024, please refer to my google scholar page.

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.

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

We propose a framework for the challenging 3D-aware Object goal navigation task based on two straightforward sub-policies. The two sub-polices, namely corner-guided exploration policy and category-aware identification policy, simultaneously perform by utilizing online fused 3D points as observation.

We propose a novel learning framework, DualAfford, to learn collaborative affordance for dual-gripper manipulation tasks. The core design of the approach is to reduce the quadratic problem for two grippers into two disentangled yet interconnected subtasks for efficient learning.

In this work, we explicitly model the camera and scene uncertainty components to solve the problem of active camera localization by reinforcement learning. Our algorithm improves over the state-of-the-art Markov Localization based approaches by a large margin on the fine-scale camera pose accuracy.

In this paper, we propose a novel framework, named AdaAfford, that learns to perform very few test-time interactions for quickly adapting the affordance priors to more accurate instance-specific posteriors.

We propose a novel multi-robot active mapping algorithm by reducing the problem to bipartite graph matching, solved by the proposed multiplex graph neural network (mGNN) via reinforcement learning.

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.

We design an interaction-for-perception framework, VAT-MART, to learn actionable visual representations for more effective manipulation of 3D articulated objects.

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.

We propose a novel contrastive learning objective, TupleInfoNCE. It contrasts tuples based not only on positive and negative correspondences, but also by composing new negative tuples using modalities describing different scenes.

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.

A data-driven framework to convert a digital illustration into three corresponding components: manga line drawing, regular screentone, and irregular screen texture. These components can be directly composed into manga images and can be further retouched for more plentiful manga creations.

A journal extension of our ECCV 2018 paper. We further propose a cheap parameter-tuning version of the decouple learning framework that enables real-time alternation between different image operators.

A dynamic graph learning algorithm for autonomous part assembly. It learns to reason an assembly-oriented dynamically-evolved relation graph, which indicates the assembly process which is guided by the major parts (chair leg&seat).

we propose a new plugin module, “Adaptive Filterbank Pyramid”, which can be inserted into a backbone network to support multiple operators and continuous parameter tuning.

A deep-learning based optical flow approach designed to handle heavy rain.

A novel semi-supervised framework for X-ray classification which is based on a graph-based optimisation model. A new multi-class classification functional that strengthens the synergy between the limited number of labels and the huge amount of unlabelled data.

A simple and tractable user interactive tool for single image reflection removal, which is facilitated with a spatially-aware prior term solved by an efficient half-quadratic splitting optimization approach.

A novel end-to-end gated context aggregation network GCANet that outperforms all the existing appraoches by a large margin on both image dehazing and deraining tasks.

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.

The first decouple learning framework that is capable of successfully incorporating many different parameterized image operators into a single network without requirement of retraining or fintuning any other networks.

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.

An advanced deep architecture for low-level vision tasks; A novel reflection image synthesis approach which enables outstanding generalization ability to real images with trained newtork.

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

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

Our algorithm works by learning a deep neural network to optimize the target with joint constraints enhanced among multiple input images during the training phase, but is able to eliminate reflections only from a single input for evaluation.

We formulate the part assembly task as a concrete reinforcement learning problem and propose a pipeline for robots to learn to assemble a diverse set of chairs.

We propose an IF-Defense framework to directly optimize the coordinates of input points with geometry-aware and distribution-aware constraints, to address the 3D adversarial attacks.

We propose contrasting “pairs of point-pixel pairs”, where positives include pairs of RGB-D points in correspondence, and negatives include pairs where one of the two modalities has been disturbed and/or the two RGB-D points are not in correspondence.

APR. 2022: Active 3D scene understanding and its applications
“三维视觉与智能图形”前沿论坛, 图图名师讲堂, China

OCT. 2021: Visual Localization
Embodied AI Workshop, Valse, China

JAN. 2019: Deep Learning in Computational Photography
USC ICT/UW Reality Lab/Berkeley/Stanford/Google/MSR, US

DEC. 2018: Deep Learning for Single Image Artifact Removal
ACCV Tutorial 2018, Australia

DEC. 2018: Image Smoothing via Unsupervised Learning
SIGGRAPH Asia 2018, Japan; GAMES Webinar, China

AUG. 2018: Discovering Unsupervised Learning in Image Processing
CIA, Cambridge University, UK

JUN. 2018: Revisiting Deep Intrinsic Image Decomposition
CVPR 2018, USA

NOV. 2015: Interactive Real-time Video Segmentation
SIGGRAPH Asia 2015, Japan

2022: Tencent Outstanding Contributor

2020: CCF Doctorial Dissertation Award Nominee (CCF 优博提名)

2019: Outstanding Academic Achievement Award of Shandong University

2018: Academic Star Nominee of Shandong University (10/20000)

2018: National Scholarship

2016: Outstanding Academic Achievement Award of Shandong University

2015: Presidential Scholarship of Shandong University (35/20000) (Highest honor for students in SDU, only 35 elected among around 20000 candidates)

2015: National Scholarship

2015: Pacemaker to Outstanding Graduate Student of Shandong University