GIT

目标检测——RetinaNet

论文——Focal Loss for Dense Object Detection

Posted by QITINGSHE on April 18, 2018

Classification Versus Detection

Classification:What Detection:What and Where

  • So object detection is arguably a harder problem than classification
  • Heavy computational processing: Usually a large number of image sub-windows need ti be scanned in order to localize objects
  • Challenge: In many real-world applications, running a fast object detector is as critical as running an accurate object detector.

Applications

  • MobilEye Forward Collision Warning
  • Camera detects up to faces
  • Video Analytics
  • Self-Driving Cars

Goals

  • 保持高准确率的同时,加速物体检测
  • 专注于最先进的技术
  • 专注于软件工具而不是硬件加速

#Region Proposals 通用物体检测:候选区域生成

  • 分组建议
  • 窗口评分建议
  • 度量标准和深入分析

Regionlets for Generic Object Detection

  • Regionlet representation for handing object deformations
  • Classification of region proposals based on boosted detector cascades
  • Integration with CNN features Regionlet

Regionlet