D2Det: Towards High Quality Object Detection and Instance Segmentation

本文轉載自查看原文 2020-12-09 15:10 513 D2Det/ Dense local regression/ Faster-RCNN/ Two-stage/ Discriminative ROI pooling/ Machine Learning/Computer Vision/ ROI Pooling

D2Det是一種two-stage算法,類似於Faster-RCNN,在Faster-RCNN的基礎上進行了一些改進,總體框架如下圖(a)所示:

和Faster-RCNN相比,改進的地方在於:

1. Dense local regression

如上圖(b)所示,Faster-RCNN是對RPN提出的ROI進行卷積操作,對提出的box進行NMS操作,得到最后的結果,而D2Det是對ROI內所有的點提出的box進行平均運算,得到最后的box,而且並不是對所有的點進行平均,而是預測ROI內所有的點的前景/背景分類,只有前景的點才參與平均運算,背景點不參與平均運算;

2. Discriminative ROI pooling

如上圖(c)所示,類似於Deformable ROI Pooling,改進之處在於,先進行一種輕量級的偏移量預測,預測偏移量之后,對於4個采樣點采用加權pooling,即哈達瑪積(Hadamard).

免責聲明！

本站轉載的文章為個人學習借鑒使用，本站對版權不負任何法律責任。如果侵犯了您的隱私權益，請聯系本站郵箱yoyou2525@163.com刪除。

猜您在找 論文閱讀《Towards End-to-End Lane Detection: an Instance Segmentation Approach》目標檢測 | 經典算法 Cascade R-CNN: Delving into High Quality Object Detection Towards Universal Object Detection by Domain Attention Learning Rich Features from RGB-D Images for Object Detection and Segmentation論文筆記 [ICCV 2019] Weakly Supervised Object Detection With Segmentation Collaboration semantic segmentation 和instance segmentation 中文版 Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks 目標檢測（四）Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks 【CV論文閱讀】Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks論文理解