regerssion based object detection.pptxVIP

Regression-based object detection directfly 2016-8-2 Deep Neural Networks for Object Detection A DNN-based regression which outputs a binary mask of the object bounding box Deep Neural Networks for Object Detection A multi-scale box inference followed by a refinement step to produce precise detections Deep Neural Networks for Object Detection Loss function Deep Neural Networks for Object Detection Multiple Masks for Robust Localization One network to predict the object box mask and four additional networks to predict four halves of the box: bottom, top, left and right halves. Object Localization from DNN Output Sliding windows Score with masks + DNN classification Non-maximum suppression Multi-scale Refinement of DNN Localizer Deep Neural Networks for Object Detection Deep Neural Networks for Object Detection 用DNN（CNN）找mask，滑窗生成的bb与mask打分，好的bb就是结果。 为了增强效果，做了refine和多mask Scalable Object Detection using Deep Neural Networks we use a Deep Neural Network (DNN), which outputs a fixed number of bounding boxes. it outputs a score for each box expressing the network confidence of this box containing an object. Scalable Object Detection using Deep Neural Networks Loss function Scalable Object Detection using Deep Neural Networks 训练时加入了一个匹配的问题，把bb和ground truth进行匹配再训练。 网络训练每个类别的bb和confidence OverFeat: Integrated Recognition, Localization and Detection using Convolutional Networks An integrated framework for using Convolutional Networks for classification, localization and detection. OverFeat: Integrated Recognition, Localization and Detection using Convolutional Networks resolution augmentation OverFeat: Integrated Recognition, Localization and Detection using Convolutional Networks Sliding Window Efficiency OverFeat: Integrated Recognition, Localization and Detection using Convolutional Networks OverFeat: Integrated Recognition, Localization and Detection using Convolutional Networks Resolution augmentation的方法感觉挺有意思的，可惜对结果帮助不是很大…… 检测还是老套路，全连接到4个bb的坐标进行回归 You Only Look Once