MLIFeat: Multi-Level Information Fusion based Deep Local Features | |
Yuyang Zhang1; Jinge Wang3; Shibiao Xu1,2; Xiao Liu3; Xiaopeng Zhang1,2 | |
2020-11 | |
会议日期 | 2020.11.30-2020.12.04 |
会议地点 | Virtual Kyoto |
页码 | 403-419 |
国家 | Japan |
英文摘要 | Accurate image keypoints detection and description are of central importance in a wide range of applications. Although there are various studies proposed to address these challenging tasks, they are far from optimal. In this paper, we devise a model named MLIFeat with two novel light-weight modules for multi-level information fusion based deep local features learning, to cope with both the image keypoints detection and description. On the one hand, the image keypoints are robustly detected by our Feature Shuffle Module (FSM), which can efficiently utilize the multi-level convolutional feature maps with marginal computing cost. On the other hand, the corresponding feature descriptors are generated by our well-designed Feature Blend Module (FBM), which can collect and extract the most useful information from the multi-level convolutional feature vectors. To study in-depth about our MLIFeat and other state-of-the-art methods, we have conducted thorough experiments, including image matching on HPatches and FM-Bench, and visual localization on Aachen-Day-Night, which verifies the robustness and effectiveness of our proposed model. |
会议录 | Computer Vision – ACCV 2020 |
语种 | 英语 |
内容类型 | 会议论文 |
源URL | [http://ir.ia.ac.cn/handle/173211/44771] |
专题 | 模式识别国家重点实验室_三维可视计算 |
通讯作者 | Shibiao Xu |
作者单位 | 1.National Laboratory of Pattern Recognition, Institute of Automation, Chinese Academy of Sciences 2.School of Artificial Intelligence, University of Chinese Academy of Sciences 3.Megvii Technology |
推荐引用方式 GB/T 7714 | Yuyang Zhang,Jinge Wang,Shibiao Xu,et al. MLIFeat: Multi-Level Information Fusion based Deep Local Features[C]. 见:. Virtual Kyoto. 2020.11.30-2020.12.04. |
个性服务 |
查看访问统计 |
相关权益政策 |
暂无数据 |
收藏/分享 |
除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。
修改评论