[caffe]深度学习之CNN检测object detection方法摘要介绍
[caffe]深度學(xué)習(xí)之CNN檢測object detection方法摘要介紹?
2015-08-17 17:44?3276人閱讀?評(píng)論(1)?收藏?舉報(bào)一兩年cnn在檢測這塊的發(fā)展突飛猛進(jìn),下面詳細(xì)review下整個(gè)cnn檢測領(lǐng)域模型的發(fā)展,以及在時(shí)間性能上的發(fā)展。
一、RCNN
流程:
Extract region(off model) + extract features(on model) + classifyregions according feature (svm or softmax)
性能:
精度:
?
二、SPP-NET
流程:
先做conv,再根據(jù)window提取特征。為什么rcnn不能也這么做呢?原因在于spp對(duì)不同尺度進(jìn)行了max pool處理能更好的滿足不同尺度window的特征表達(dá)。
性能:
核心思想在全圖只做一次conv,這個(gè)和overfeat的思想一致
?
精度:
三、FAST-RCNN
流程:
引入了ROI層pooling,以及multi-task同時(shí)訓(xùn)練分類和檢測框。
性能:
Compared to SPPnet, Fast R-CNN trains VGG163× faster, tests 10× faster, and is more accurate.
另外還額外提出了fc層SVD的思想
Vgg時(shí)間性能分析
?
精度:
The improvement of Fast R-CNN over SPPnetillustrates that even though Fast R-CNN uses single-scale training and testing,fine-tuning the conv layers provides a large improvement in mAP (from 63.1% to66.9%). Traditional R-CNN achieves a mAP of 66.0%. These results arepragmatically valuable given how much faster and easier Fast R-CNN is to trainand test, which we discuss next.
?
?
?
?
四、FASTER-RCNN
流程:
在fast-rcnn的基礎(chǔ)上,借鑒了FCN的思路,將proposal階段轉(zhuǎn)化成一個(gè)layer加進(jìn)了網(wǎng)絡(luò)一起學(xué)習(xí)。
性能:
cost-free for proposal
精度:
our detection system has a frame rate of5fps (including all steps) on a GPU, while achieving state-of-the-art objectdetection accuracy on PASCAL VOC 2007 (73.2% mAP) and 2012 (70.4% mAP) using300 proposals per image
from:?http://blog.csdn.net/sunbaigui/article/details/47728251
總結(jié)
以上是生活随笔為你收集整理的[caffe]深度学习之CNN检测object detection方法摘要介绍的全部內(nèi)容,希望文章能夠幫你解決所遇到的問題。
- 上一篇: 楼燚(yì)航的blog URL
- 下一篇: 深度学习系列:深度学习在腾讯的平台化和应