You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
1. 论文列表
2. D2D Keypoint Extraction with Describe to Detect Approach 论文理解
3. R2D2 Repeatable and Reliable Detector and Descriptor 论文理解
4. ASLFeat Learning Local Features of Accurate Shape and Localization 论文理解
5. 局部特征检索专利技术交底书
6. Attention-Based Query
2001.05027-DELG Unifying Deep Local and Global Features for Image Search 从conv4_x提取特征进入attention模块得到local feature,再经过降维从1024->40维,global feature从conv5_x经过聚合得到。端到端global和local特征提取方法
cvpr2019EGT Explore-Exploit Graph Traversal for Image Retrieval 提升搜索效率的方法
2003.13827 Co-Occurrence of Deep Convolutional Features for
Image Search 共现关系提取通道相关性tensor,bilinear pooling产生描述符
2001.07252-UR2KiD UR2KiD: Unifying Retrieval, Keypoint Detection, and Keypoint Description
without Local Correspondence Supervision 端到端,未细看
5.1907.05794 ACTNet ACTNET: end-to-end learning of feature
activations and multi-stream aggregation for
effective instance image retrieval 端到端提特征方法,可学习激活层比GeM提升,主要在描述符提取方法上改进
1803.11285 Revisiting Oxford and Paris: Large-Scale Image Retrieval Benchmarking 图像检索基准,改进数据集和度量方法
Key.Net: Keypoint Detection by Handcrafted and Learned CNN Filters 手工和CNN
HDD-Net: Hybrid Detector Descriptor with Mutual Interactive Learning 手工和CNN
D2D: Keypoint Extraction with Describe to Detect Approach.md
R2D2: Repeatable and Reliable Detector and Descriptor.md
特征提取时间:包含模型推理DELF/DELG,描述符提取(aggregation聚合)GeM、GeMmp、SPoC、REMAP、R-MAC,损失函数triple loss,ArcFace Loss
高响应高区分注意力模块具体操作(见附图2)如下:首先对输入特征图进行核大小1的ConvBnRelu操作缩减通道为512,分两个分支,其中高响应度分支对channel mean reduce后的特征和global average pooling后的特征做标准差,另外高区分度分支对特征点一定范围内的邻域特征求L2距离并加权求和,然后将两个分支进行元素乘积操作,再后核大小1的Conv操作缩减通道为1,经过softplus激活得到注意力得分图。若是训练阶段,注意力得分图还要与L2Normalize后的输入特征图做点积,得到HxWx1的特征图,reshape后进行交叉熵损失优化。