基于器官造型的植物精细重建
项目来源
项目主持人
项目受资助机构
立项年度
立项时间
项目编号
项目级别
研究期限
受资助金额
学科
学科代码
基金类别
关键词
参与者
参与机构
项目受资助省
项目结题报告(全文)
1.Efficient Joint Gradient Based Attack against SOR Defense for 3D Point Cloud Classification
- 关键词：
- Deep learning;Statistics;Decision boundary;Defense strategy;Hyper-parameter;Linear approximations;Non-differentiable optimization;Objective functions;Outlier removals;State of the art
- Ma, Chengcheng;Meng, Weiliang;Wu, Baoyuan;Xu, Shibiao;Zhang, Xiaopeng
- 《28th ACM International Conference on Multimedia, MM 2020》
- 2020年
- October 12, 2020 - October 16, 2020
- Virtual, Online, United states
- 会议
Deep learning based classifiers on 3D point cloud data have been shown vulnerable to adversarial examples, while a defense strategy named Statistical Outlier Removal (SOR) is widely adopted to defend adversarial examples successfully, by discarding outlier points in the point cloud. In this paper, we propose a novel white-box attack method, Joint Gradient Based Attack (JGBA), aiming to break the SOR defense. Specifically, we generate adversarial examples by optimizing an objective function containing both the original point cloud and its SOR-processed version, for the purpose of pushing both of them towards the decision boundary of classifier at the same time. Since the SOR defense introduces a non-differentiable optimization problem, we overcome the problem by introducing a linear approximation of the SOR defense and successfully compute the joint gradient. Moreover, we impose constraints on perturbation norm for each component point in the point cloud instead of for the entire object, to further enhance the attack ability against the SOR defense. Our JGBA method can be directly extended to the semi white-box setting, where the values of hyper-parameters in the SOR defense are unknown to the attacker. Extensive experiments validate that our JGBA method achieves the highest performance to break both the SOR defense and the DUP-Net defense (a recently proposed defense which takes SOR as its core procedure), compared with state-of-the-art attacks on four victim classifiers, namely PointNet, PointNet++(SSG), PointNet++(MSG), and DGCNN. © 2020 ACM.
...2.Detailed 3D Face Reconstruction from Single Images Via Self-supervised Attribute Learning
- 关键词：
- 3D modeling;Three dimensional computer graphics;3D face reconstruction;3D Morphable model;High-fidelity;Human face model;Recovery scheme;RGB images;Single images
- Yang, Mingxin;Guo, Jianwei;Ye, Juntao;Zhang, Xiaopeng
- 《SIGGRAPH Asia 2020 Posters - International Conference on Computer Graphics and Interactive Techniques, SA 2020》
- 2020年
- December 4, 2020 - December 13, 2020
- Virtual, Online, Korea, Republic of
- 会议
We present a novel approach to reconstruct high-fidelity geometric human face model from a single RGB image. The main idea is to add details into a coarse 3D Morphable Model (3DMM) based model in a self-supervised way. Our observation is that most of the facial details like wrinkles are driven by expression and intrinsic facial characteristics which here we refer to as the facial attribute. To this end, we propose an expression related details recovery scheme and a facial attribute representation. © 2020 Owner/Author.
...