About Me
I am a Ph.D. candidate in Control Science and Engineering at the Department of Automation, University of Science and Technology of China (USTC), advised by Prof. Tianzhu Zhang. I received my bachelor's degree from the Department of Electronic Engineering and Information Science at USTC in 2022 as a member of the Talent Program in Artificial Intelligence. I expect to graduate in 2027.
My research focuses on robust and trustworthy visual understanding across natural images, 3D point clouds, medical images, and vision-language models. I develop methods for fine-grained perception and multimodal reasoning under distribution shifts, with particular interests in domain adaptation and generalization, confidence calibration, semi-supervised learning, and efficient and reliable multimodal inference.
I am currently a research intern at Tencent Youtu Lab, working on fine-grained visual understanding and image-to-code capabilities for multimodal large language models.
Research Experience
Fine-Grained Visual Understanding Across Multiple Modalities (Natural Images / 3D Point Clouds / Medical Images)
Natural Images
- Towards Robust Pseudo-Label Learning in Semantic Segmentation: An Encoding PerspectiveNeurIPS 2025 · CCF A · Top-tier machine learning conference · First author · Domain adaptation
- Towards Unsupervised Domain Bridging via Image Degradation in Semantic SegmentationNeurIPS 2025 · CCF A · Top-tier machine learning conference · First author · Domain adaptation · Semi-supervised learning
- Balanced Learning for Domain Adaptive Semantic SegmentationICML 2025 · CCF A · Top-tier machine learning conference · First author · Domain adaptation
- DA-Cal: Towards Cross-Domain Calibration in Semantic SegmentationTIP 2026 · SCI Q1 · CCF A · Top-tier computer vision journal · First author · Cross-domain calibration
- Exploring Weather-Aware Aggregation and Adaptation for Semantic Segmentation Under Adverse ConditionsICCV 2025 · CCF A · Top-tier computer vision conference · Domain adaptation
- Towards Unbiased Learning in Semi-Supervised Semantic SegmentationICLR 2025 · CCF A · Top-tier machine learning conference · Semi-supervised learning
- Beyond Confidence: Exploiting Homogeneous Patterns for Semi-Supervised Semantic SegmentationICML 2025 · CCF A · Top-tier machine learning conference · Semi-supervised learning
- Two Losses, One Goal: Balancing Conflict Gradients for Semi-Supervised Semantic SegmentationICCV 2025 · CCF A · Highlight · Top-tier computer vision conference · Semi-supervised learning
- From Softmax to Dirichlet: Evidential Learning for Semi-Supervised Semantic SegmentationCVPR 2026 · CCF A · Top-tier computer vision conference · Semi-supervised learning
- Dual-Agent Optimization Framework for Cross-Domain Few-Shot SegmentationCVPR 2025 · CCF A · Top-tier computer vision conference · Few-shot learning
3D Point Clouds
- Adaptive Augmentation-Aware Latent Learning for Robust LiDAR Semantic SegmentationICLR 2026 · CCF A · Top-tier machine learning conference · First author · Domain generalization
- BeyondMix: Leveraging Structural Priors and Long-Range Dependencies for Domain-Invariant LiDAR SegmentationNeurIPS 2025 · CCF A · Top-tier machine learning conference · Domain adaptation
- DA2-LiDAR: A Generic Density-Adaptive Framework for Unsupervised Domain Adaptation in LiDAR SegmentationTIP · SCI Q1 · CCF A · Top-tier computer vision journal · Domain adaptation
- Localization and Expansion: A Decoupled Framework for Point Cloud Few-Shot Semantic SegmentationECCV 2024 · Top-tier computer vision conference · Few-shot learning
- Generalized Few-Shot Point Cloud Segmentation via LLM-Assisted Hyper-Relation MatchingICCV 2025 · CCF A · Top-tier computer vision conference · Few-shot learning
Medical and Biological Images
- MAUNet: Modality-Aware Anti-Ambiguity U-Net for Multi-Modality Cell SegmentationNeurIPS Competition Track 2022 · CCF A Workshop · First author · Domain generalization · Semi-supervised learning
- Enhancing Cell Detection in Histopathology Images: A ViT-Based U-Net ApproachMICCAI Workshop 2023 · Top-tier medical imaging conference workshop · Cell detection
- SAM-Glomeruli: Enhanced Segment Anything Model for Precise Glomeruli SegmentationMICCAI Workshop 2024 · Top-tier medical imaging conference workshop · Glomeruli segmentation
- Alleviate and Mining: Rethinking Unsupervised Domain Adaptation for Mitochondria Segmentation from a Pseudo-Label PerspectiveAAAI 2025 · CCF A · Top-tier artificial intelligence conference · Domain adaptation
Multimodal Understanding, Alignment, and Efficient Inference for Vision-Language Models
- Cycle-Consistent Test-Time Adaptation for Reasoning SegmentationEMNLP 2026 · Under review · Top-tier natural language processing conference · First author · Reasoning segmentation · Multimodal adaptation
- Principled Semantic Representation Optimization for Text-Assisted Image ClusteringNeurIPS 2026 · CCF A · Under review · Top-tier machine learning conference · Co-first author · Multimodal alignment · Image clustering
- Spectral Heat Flow for Conservative Token Condensation in Vision-Language ModelsICML 2026 · CCF A · Top-tier machine learning conference · Visual token condensation · Efficient inference
- Beyond Logits: Coherent Hallucination Mitigation via Attention Contrastive DecodingICML 2026 · CCF A · Top-tier machine learning conference · Hallucination mitigation · Contrastive decoding
- Beyond Blind Noising: Disentangled Visual Rectification for Hallucination Mitigation in MLLMsICML 2026 · CCF A · Top-tier machine learning conference · Hallucination mitigation · Visual rectification
Selected Publications
- Wangkai Li, Rui Sun, Zhaoyang Li, Yujia Chen, Tianzhu Zhang. “DA-Cal: Towards Cross-Domain Calibration in Semantic Segmentation.” IEEE Transactions on Image Processing (TIP), 2026.
- Wangkai Li, Zhaoyang Li, Yuwen Pan, Rui Sun, Yujia Chen, Tianzhu Zhang. “Adaptive Augmentation-Aware Latent Learning for Robust LiDAR Semantic Segmentation.” International Conference on Learning Representations (ICLR), 2026.
- Wangkai Li, Rui Sun, Zhaoyang Li, Tianzhu Zhang. “Towards Robust Pseudo-Label Learning in Semantic Segmentation: An Encoding Perspective.” Conference on Neural Information Processing Systems (NeurIPS), 2025.
- Wangkai Li, Rui Sun, Huayu Mai, Tianzhu Zhang. “Towards Unsupervised Domain Bridging via Image Degradation in Semantic Segmentation.” Conference on Neural Information Processing Systems (NeurIPS), 2025.
- Wangkai Li, Rui Sun, Bohao Liao, Zhaoyang Li, Tianzhu Zhang. “Balanced Learning for Domain Adaptive Semantic Segmentation.” International Conference on Machine Learning (ICML), 2025.
Academic Competitions
- 1st Place, WSI-Level Glomeruli Detection in Kidney Pathology Image Segmentation Challenge (MICCAI 2024)
- 2nd Place, NTIRE Efficient Super-Resolution Challenge (CVPR 2024)
- Second Prize (3rd Place; CNY 200,000), Video Interpolation in Fast-Motion Scenes, 2nd Guangdong-Hong Kong-Macao Greater Bay Area Algorithm Case Competition
- 1st Place, Cell Detection from Cell-Tissue Interaction Challenge (MICCAI 2023)
- Spotlight Tracker, Best Robustness, Visual Object Tracking and Segmentation Challenge (ICCV 2023)
- Meritorious Winner, Rank: 4/102, Weakly Supervised Cell Segmentation in Multimodal High-Resolution Microscopy Images Challenge (NeurIPS 2022)
- 1st Place, Visual Object Tracking Short-Term Bounding-Box Challenge (ECCV 2022)
Awards
- Jianghuai NIO Scholarship, USTC 2025
- First-Class Graduate Scholarship, USTC (awarded five times) 2022–2026
- First-Class Scholarship, Talent Program in Artificial Intelligence, USTC 2021
- First Prize, 12th Chinese Mathematics Competition (Non-Mathematics Major) Oct. 2020
- Outstanding Undergraduate Scholarship, Silver Award, USTC (awarded three times) 2019–2021
Professional Service
Reviewer for TPAMI, TMM, CVPR (2024–2026), AAAI (2025–2026), ICLR (2025–2026), ICCV 2025, ICML 2026, ECCV 2026, and NeurIPS 2026.
关于我
我是中国科学技术大学自动化系控制科学与工程专业博士研究生,导师为张天柱教授,预计于 2027 年毕业。我于 2022 年本科毕业于中国科学技术大学电子工程与信息科学系,并入选人工智能英才班。
我的研究聚焦于自然图像、三维点云、医学影像及视觉语言模型中的鲁棒可信视觉理解,致力于提升模型在分布偏移和复杂场景下的细粒度感知与多模态推理能力。具体研究方向包括域适应与泛化、置信度校准、半监督学习,以及高效可靠的多模态推理。
目前,我在腾讯优图实验室担任研究实习生,开展多模态大模型细粒度视觉理解与 image-to-code 方向的研究。
研究经历
面向多种模态数据(自然图像 / 三维点云 / 医学影像)的细粒度视觉理解
自然图像
- Towards Robust Pseudo-Label Learning in Semantic Segmentation: An Encoding PerspectiveNeurIPS 2025 · CCF A · 机器学习顶会 · 独立一作 · 跨域适应
- Towards Unsupervised Domain Bridging via Image Degradation in Semantic SegmentationNeurIPS 2025 · CCF A · 机器学习顶会 · 独立一作 · 跨域适应 · 半监督学习
- Balanced Learning for Domain Adaptive Semantic SegmentationICML 2025 · CCF A · 机器学习顶会 · 独立一作 · 跨域适应
- DA-Cal: Towards Cross-Domain Calibration in Semantic SegmentationTIP 2026 · SCI 一区 · CCF A · 计算机视觉顶刊 · 独立一作 · 跨域校准
- Exploring Weather-Aware Aggregation and Adaptation for Semantic Segmentation Under Adverse ConditionsICCV 2025 · CCF A · 计算机视觉顶会 · 跨域适应
- Towards Unbiased Learning in Semi-Supervised Semantic SegmentationICLR 2025 · CCF A · 机器学习顶会 · 半监督学习
- Beyond Confidence: Exploiting Homogeneous Patterns for Semi-Supervised Semantic SegmentationICML 2025 · CCF A · 机器学习顶会 · 半监督学习
- Two Losses, One Goal: Balancing Conflict Gradients for Semi-Supervised Semantic SegmentationICCV 2025 · CCF A · Highlight · 计算机视觉顶会 · 半监督学习
- From Softmax to Dirichlet: Evidential Learning for Semi-Supervised Semantic SegmentationCVPR 2026 · CCF A · 计算机视觉顶会 · 半监督学习
- Dual-Agent Optimization Framework for Cross-Domain Few-Shot SegmentationCVPR 2025 · CCF A · 计算机视觉顶会 · 小样本学习
三维点云
- Adaptive Augmentation-Aware Latent Learning for Robust LiDAR Semantic SegmentationICLR 2026 · CCF A · 机器学习顶会 · 独立一作 · 跨域泛化
- BeyondMix: Leveraging Structural Priors and Long-Range Dependencies for Domain-Invariant LiDAR SegmentationNeurIPS 2025 · CCF A · 机器学习顶会 · 跨域适应
- DA2-LiDAR: A Generic Density-Adaptive Framework for Unsupervised Domain Adaptation in LiDAR SegmentationTIP · SCI 一区 · CCF A · 计算机视觉顶刊 · 跨域适应
- Localization and Expansion: A Decoupled Framework for Point Cloud Few-Shot Semantic SegmentationECCV 2024 · 计算机视觉顶会 · 小样本学习
- Generalized Few-Shot Point Cloud Segmentation via LLM-Assisted Hyper-Relation MatchingICCV 2025 · CCF A · 计算机视觉顶会 · 小样本学习
医学与生物影像
- MAUNet: Modality-Aware Anti-Ambiguity U-Net for Multi-Modality Cell SegmentationNeurIPS Competition Track 2022 · CCF A Workshop · 独立一作 · 跨域泛化 · 半监督学习
- Enhancing Cell Detection in Histopathology Images: A ViT-Based U-Net ApproachMICCAI Workshop 2023 · 医学图像顶会 Workshop · 细胞检测
- SAM-Glomeruli: Enhanced Segment Anything Model for Precise Glomeruli SegmentationMICCAI Workshop 2024 · 医学图像顶会 Workshop · 肾小球分割
- Alleviate and Mining: Rethinking Unsupervised Domain Adaptation for Mitochondria Segmentation from a Pseudo-Label PerspectiveAAAI 2025 · CCF A · 人工智能顶会 · 跨域适应
面向视觉语言大模型的多模态理解与对齐、推理加速及幻觉缓解
- Cycle-Consistent Test-Time Adaptation for Reasoning SegmentationEMNLP 2026 · 在投 · 自然语言处理顶会 · 独立一作 · 推理分割 · 多模态适应
- Principled Semantic Representation Optimization for Text-Assisted Image ClusteringNeurIPS 2026 · CCF A · 在投 · 机器学习顶会 · 共同一作 · 多模态对齐 · 图像聚类
- Spectral Heat Flow for Conservative Token Condensation in Vision-Language ModelsICML 2026 · CCF A · 机器学习顶会 · 视觉 token 压缩 · 推理加速
- Beyond Logits: Coherent Hallucination Mitigation via Attention Contrastive DecodingICML 2026 · CCF A · 机器学习顶会 · 幻觉缓解 · 对比解码
- Beyond Blind Noising: Disentangled Visual Rectification for Hallucination Mitigation in MLLMsICML 2026 · CCF A · 机器学习顶会 · 幻觉缓解 · 视觉校正
代表性论文
- 李汪楷,孙锐,李钊阳,陈钰佳,张天柱。“DA-Cal: Towards Cross-Domain Calibration in Semantic Segmentation.” IEEE Transactions on Image Processing (TIP),2026。
- 李汪楷,李钊阳,潘裕文,孙锐,陈钰佳,张天柱。“Adaptive Augmentation-Aware Latent Learning for Robust LiDAR Semantic Segmentation.” International Conference on Learning Representations (ICLR),2026。
- 李汪楷,孙锐,李钊阳,张天柱。“Towards Robust Pseudo-Label Learning in Semantic Segmentation: An Encoding Perspective.” Conference on Neural Information Processing Systems (NeurIPS),2025。
- 李汪楷,孙锐,麦华煜,张天柱。“Towards Unsupervised Domain Bridging via Image Degradation in Semantic Segmentation.” Conference on Neural Information Processing Systems (NeurIPS),2025。
- 李汪楷,孙锐,廖柏皓,李钊阳,张天柱。“Balanced Learning for Domain Adaptive Semantic Segmentation.” International Conference on Machine Learning (ICML),2025。
竞赛经历
- 第一名,肾脏病理图像 WSI 级肾小球检测挑战赛(MICCAI 2024)
- 第二名,NTIRE 高效超分辨率挑战赛(CVPR 2024)
- 二等奖(第三名,奖金 20 万元),快速运动场景下视频插帧挑战赛(第二届粤港澳大湾区算法算例大赛)
- 第一名,细胞-组织交互下的细胞检测挑战赛(MICCAI 2023)
- Spotlight Tracker,最佳鲁棒性,视觉目标跟踪与分割挑战赛(ICCV 2023)
- 优胜奖(排名 4/102),多模态高分辨率显微图像弱监督细胞分割挑战赛(NeurIPS 2022)
- 第一名,视觉目标跟踪短时边框挑战赛(ECCV 2022)
所获奖项
- 中国科学技术大学江淮蔚来奖学金 2025
- 中国科学技术大学研究生一等奖学金(五次) 2022–2026
- 中国科学技术大学人工智能英才班一等奖学金 2021
- 第十二届全国大学生数学竞赛一等奖(非数学类) 2020 年 10 月
- 中国科学技术大学本科生优秀学生奖学金(银奖,三次) 2019–2021
学术服务
担任 TPAMI、TMM、CVPR(2024–2026)、AAAI(2025–2026)、ICLR(2025–2026)、ICCV 2025、ICML 2026、ECCV 2026 和 NeurIPS 2026 审稿人。