標籤:
近一兩年cnn在檢測這塊的發展突飛猛進,下面詳細review下整個cnn檢測領域模型的發展,以及在時間效能上的發展。
一、RCNN流程:
Extract region(off model) + extract features(on model) + classifyregions according feature (svm or softmax)
效能:
精度:
二、SPP-NET流程:
先做conv,再根據window提取特徵。為什麼rcnn不能也這麼做呢?原因在於spp對不同尺度進行了max pool處理能更好的滿足不同尺度window的特徵表達。
效能:
核心思想在全圖只做一次conv,這個和overfeat的思想一致
精度:
三、FAST-RCNN流程:
引入了ROI層pooling,以及multi-task同時訓練分類和檢測框。
效能:
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時間效能分析
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的基礎上,借鑒了FCN的思路,將proposal階段轉化成一個layer加進了網路一起學習。
效能:
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
著作權聲明:本文為博主原創文章,未經博主允許不得轉載。
CNN檢測摘要