与数字图像处理的区别 ·数字图像处理 ·图像/视频>图像/视频 ·图像变换、图像滤波、图像复原、图像压缩、 ·计算机视觉 ·图像/视频->模型 ·David Marr视觉计算理论的创始人 ·二维基素图->2.5维要素图->三维模型表征 Hangzhou Dianzi University抗州电子科技大学 School of Computer Science and Tecfnology计算机学院周文晖
Hangzhou Dianzi University 杭州电子科技大学 School of Computer Science and Technology 计算机学院 周文晖 与数字图像处理的区别 • 数字图像处理 • 图像/视频 ‐> 图像/视频 • 图像变换、图像滤波、图像复原、图像压缩、 … • 计算机视觉 • 图像/视频 ‐> 模型 • David Marr 视觉计算理论的创始人 • 二维基素图 ‐> 2.5维要素图 ‐> 三维模型表征
Vision is really hard ·视觉是大自然智慧的一个奇迹 ·视觉皮层面积约占整个大脑皮层面积的50% Is that a ·视力正常的人类大脑中视觉活动占整个大脑活动的 queen or a 主导地位 bishop? Hangzhou Dianzi University抗州电子科技大学 School of Computer Science and Technology计算机学院周文晖
Hangzhou Dianzi University 杭州电子科技大学 School of Computer Science and Technology 计算机学院 周文晖 Vision is really hard • 视觉是大自然智慧的一个奇迹 • 视觉皮层面积约占整个大脑皮层面积的50% • 视力正常的人类大脑中视觉活动占整个大脑活动的 主导地位 Is that a queen or a bishop?
3D from a single image and shape-from-x Action and behavior recognition Adversarial learning CVPR2021 ICCV2021 。Biometrics Computational photography 2D object recognition .Datasets and evaluation ·3 D computer vision Detection and localization in 2D and 3D .3D object recognition Efficient training and inference methods Action and behavior recognition ·Explainable Al Adversarial learning.adversarial attack and defense methods ·Faces Biometrics,face,gesture,body pose Fairness,accountability,transparency,and ethics in vision Computational photography Gestures and body pose Datasets and evaluation Image and video synthesis Efficient training and inference methods for networks Image and video retrieval Explainable Al,fairness.accountability,privacy,transparency and ethics in vision Image and video manipulation detection and integrity methods. Image and video retrieval Image and video synthesis Low-level and physics-based vision ·Image classification Machine learning architectures and formulations Low-level and physics-based vision Medical,biological,and cell microscopy Machine learning architectures and formulations 。Motion and tracking Medical,biological and cell microscopy Neural generative models 。Motion and tracking Optimization and learning methods Optimization and learning methods Recognition and classifcation ·Pose estimation Representation learning Representation learning.deep learning Scene analysis and understanding Scene analysis and understanding Scene text and document understanding Transfer,low-shot,semi-and un-supervised learning Segmentation,grouping and shape Video analysis and understanding Stereo,3D from multiview and other sensors Vision +language,vision +other modalities Transfer/Low-shot/Semi/Unsupervised Learning Vision applications and systems,vision for robotics and autonomous vehicles Video analysis and understanding ·Vision+language Vision +other modalities Hangzhou Dianzi气iversity抗州电子科技大学 School of Vision applications and systems Vision for robotics and autonomous vehicles Visual reasoning and logical representation
Hangzhou Dianzi University 杭州电子科技大学 School of Computer Science and Technology 计算机学院 周文晖 CVPR2021 ICCV2021
CVPR 2018 Submission Top 75 Keywords CVPR 2019 Submission Top 75 Keywords CVPR2018 CVPR2019 pe ECCV2020 Number of papers per area Deep Learning Applications Methodology and Theory 128 Recognition:Detection Categorization+Retrieval and 37 Matching Machine Leaming Image and Video Synthesis 46 Segmentation Grouping and Shape 430 Face+Gesture+and Body Pose Representation Leamning 45 Low-level Vision Unsupervised Learning 35 000 ach ter Science and Technolog)y计算机学院周文晦 Frequency
Hangzhou Dianzi University 杭州电子科技大学 School of Computer Science and Technology 计算机学院 周文晖 ECCV2020 CVPR2018 CVPR2019
深度学习已经成为计算机视觉中的标配？ ·首要原因是深度学习可以做到传统方法无法达到的精度： ·深度学习算法的通用性强；比如faster RCNN在人脸、行人、一般物体检测任务上 都可以取得非常好的效果； ·工程开发、优化、维护成本低。深度学习计算主要是卷积和矩阵乘，针对这种计 算优化，所有深度学习算法都可以提升性能。 Hangzhou Dianzi乙nversity杭州电子科技大学 School of Computer Science and Technology计算机学院周文晖
Hangzhou Dianzi University 杭州电子科技大学 School of Computer Science and Technology 计算机学院 周文晖 深度学习已经成为计算机视觉中的标配？ • 首要原因是深度学习可以做到传统方法无法达到的精度； • 深度学习算法的通用性强；比如faster RCNN在人脸、行人、一般物体检测任务上 都可以取得非常好的效果； • 工程开发、优化、维护成本低。深度学习计算主要是卷积和矩阵乘，针对这种计 算优化，所有深度学习算法都可以提升性能