北京联合大学 · 人工智能 → 北京交通大学(保研)Beijing Union University · AI → Beijing Jiaotong University

任义Yi Ren Yi Ren任义

我做底层视觉与图像复原,关注算法在真实、复杂的退化条件下依然稳定、色彩可信。目前研究方向是图像去雾与屏幕内容图像超分辨率(SCISR)。

I work on low-level vision and image restoration, aiming for methods that stay stable and color-faithful under real, complex degradations. My current directions are image dehazing and screen content image super-resolution (SCISR).

看研究See research GitHub

01简介About

我毕业于沈阳市第二十中学(辽宁省重点高中、省示范高中),2023 年进入北京联合大学人工智能专业,大一加入景竑元副教授团队,开始图像复原研究,熟悉 CNN、Transformer、Mamba 等架构,参与过算法设计、实验迭代与论文写作。在图像去雾上,我关心去雾后的颜色保真与残余雾;接下来会把同样的思路带到屏幕内容图像超分辨率。

I graduated from Shenyang No. 20 High School (a provincial key and model school) and entered Beijing Union University in 2023 to study AI. I joined Hongyuan Jing's group in my first year and have worked on image restoration since, across CNN, Transformer and Mamba architectures, from algorithm design to experiments and writing. In dehazing I care about color fidelity and residual haze, and I am carrying the same thinking into screen content super-resolution.

  • Python
  • C / C++
  • PyTorch
  • Git
  • CNN
  • Transformer
  • Mamba
4.18GPA / 5.0
1/36专业排名Major rank
9国家级及以上奖项National-level awards
16省部级奖项Provincial awards
北京联合大学校徽 / Beijing Union University emblem

北京联合大学Beijing Union University

机器人学院(人工智能学院)· 人工智能专业 · 本科 2023.09 至今College of Robotics · B.Eng. in Artificial Intelligence · 2023.09 – present

  • 北京市属高校Beijing municipal university
  • 多个北京市重点实验室Several Beijing key labs

导师Advisor 景竑元 副教授Assoc. Prof. Hongyuan Jing

北京联合大学机器人学院副教授,英国莱斯特大学工学博士,硕士生导师;ICGIP、ICCCS 等国际会议技术委员会成员、分论坛主席。研究方向为目标检测、图像复原(去雾、去噪、超分辨率)与多模态融合;近五年在 IEEE TIM、IoT 等 SCI 期刊发表论文 10 余篇,承担欧盟 FP7、国家“十四五”重点研发子课题、北京市自然科学基金等项目 8 项。

Associate professor at the College of Robotics, Beijing Union University; Ph.D. from the University of Leicester; master's supervisor; technical committee member and session chair for ICGIP and ICCCS. Works on object detection, image restoration (dehazing, denoising, super-resolution) and multimodal fusion; 10+ SCI papers in five years (IEEE TIM, IoT, etc.) and 8 funded projects including EU FP7 and a national 14th Five-Year key R&D subtopic.

北京交通大学校徽 / Beijing Jiaotong University emblem

北京交通大学Beijing Jiaotong University

国家卓越工程师学院 · 计算机技术专业硕士(专硕)· 推免录取National Elite Engineers College · professional M.Eng. in Computer Technology · recommended admission

  • 211
  • 双一流Double First-Class
  • 教育部直属Ministry of Education

导师Advisor 白慧慧 教授Prof. Huihui Bai

北京交通大学信息科学研究所教授、博士生导师,国家级高层次人才,教育部特聘长江学者。研究方向为计算机视觉、视频分析与视频编码;主持国家自然科学基金重点项目、国家重点研发计划重点专项等,发表论文 100 余篇(TPAMI、TIP、TCSVT、CVPR 等),获北京市科技进步二等奖、中国电子学会青年科学家奖。官方主页 ↗

Professor and doctoral supervisor at the Institute of Information Science, Beijing Jiaotong University; national high-level talent and Changjiang Scholar (MOE). Works on computer vision, video analysis and video coding; led an NSFC key project and a national key R&D program, 100+ papers (TPAMI, TIP, TCSVT, CVPR, etc.), Beijing Science and Technology Progress Award (2nd prize) and CIE Young Scientist Award. Official page ↗

02研究方向Research

拖动中间的把手,看两个方向各自想解决什么。

Drag the handle to see what each direction is about.

Clear scene Hazy scene 有雾 I(x)Hazy I(x) 清晰 J(x)Clear J(x)

图像去雾Image Dehazing

I(x) = J(x)·t(x) + A·(1 − t(x))

雾让场景光被衰减,并混入大气光 A。我关注去雾后的颜色保真与残余雾:HSV-DehazeNet 从色相一致性和雾浓度监督入手,TAE-Net 与 PEFA-Net 则从注意力的聚焦与并行效率入手。

Haze attenuates scene light and mixes in atmospheric light A. I care about color fidelity and residual haze: HSV-DehazeNet works on hue consistency and haze-density supervision, while TAE-Net and PEFA-Net tackle attention focus and parallel efficiency.

左:有雾图像;右:对应的无雾真值图像(配对数据,非模型输出)。 Left: hazy image; right: the paired haze-free ground truth (not model output).
模糊屏幕图像 LRBlurry screen image LR 高分辨率 HRHigh-res HR

屏幕内容图像超分辨率 SCISRScreen Content Image Super-Resolution

LR → HR · text · UI · charts

屏幕内容(文字、图标、界面、图表)有锐利的边缘、细线和大面积平坦区域,与自然图像的统计特性很不一样,简单插值后笔画极易发糊。这是我正在转向的新方向。

Screen content (text, icons, interfaces, charts) has sharp edges, thin strokes and large flat areas, unlike natural images; plain interpolation blurs strokes easily. This is the new direction I am moving into.

示意图由页面实时绘制,非模型输出。 Illustration drawn live in the page, not model output.

03论文Publications

SCI 期刊论文,按我的作者位次排序。

SCI journal papers, ordered by my authorship position.

SCI · JCR Q3第一作者First author2026

HSV-DehazeNet: Hue Consistency Calibration and Haze Density Supervision for Image Dehazing

Yi Ren, Hongyuan Jing, Songhao Wu, Wenlu Yang, Mengfei Han, Jinjin Hu, Kehong Li*, Mengmeng Zhang (* corresponding author)

Pattern Analysis and Applications, 29:113

针对去雾后的颜色失真与残余雾问题,从 HSV 空间出发:发现去雾前后色相(H)通道变化很小,据此设计 HCCM 模块约束色相一致性;同时把饱和度差异作为雾浓度线索,作为训练时的辅助监督(HDS),使网络直接学习雾浓度的变化。

Targets color shift and residual haze. In HSV space the hue channel barely changes before and after dehazing, so HCCM enforces hue consistency; the saturation discrepancy serves as a haze-density cue for auxiliary supervision (HDS) during training.

SCI · JCR Q2学生第二作者Second student author2026

FSDG-Net: A Frequency-Spatial Parallel Network with Density Guidance for Image Dehazing

Wenlu Yang, Hongyuan Jing*, Yi Ren, Jinjin Hu, Mengfei Han, Mengmeng Zhang (* corresponding author)

Signal Processing: Image Communication, vol. 148, 117667

频域与空域并行处理:FSPP 模块在保持空域结构稳定的同时,分别处理频域的高、低频成分来恢复细节;雾浓度估计模块(HDEB)学习雾的空间分布并自适应调整去雾权重;多阶段特征融合(MF)促进互补信息的流动。

Frequency and spatial branches in parallel: the FSPP block keeps spatial structure stable while processing high and low frequencies separately to restore detail; a Haze Density Estimation Block (HDEB) learns the spatial haze distribution and adapts the dehazing weights; a multi-stage fusion block (MF) promotes complementary information flow.

SCI · JCR Q2合作作者Co-author

EGCDNet: Edge-Guided Cross-Domain Image Restoration Network

Mengfei Han, Hongyuan Jing*, Mengmeng Zhang, Jinjin Hu, Yi Ren, Wenlu Yang (* corresponding author)

The Visual Computer (Springer)

边缘引导的跨域图像复原网络。

An edge-guided cross-domain image restoration network.

会议论文(EI)Conference papers (EI)

04动态News

05竞赛与荣誉Competitions & Honors

CVPR NTIRE 2025Day & Night Raindrop Removal · 冠军Winner
中国机器人及人工智能大赛 2026CRAIC 2026人工智能创新赛 · 全国总决赛一等奖AI Innovation Track · National Final First Prize

另有软件著作权 4 项,实审发明专利 1 项。

Also 4 software copyrights and 1 patent under substantive examination.

06照片Moments

竞赛答辩Competition defense
日常Off duty