zhoujunr1的博客

Rich feature hierarchies for accurate object detection and semantic segmentationRCNN 物体检测模型结构整体模型包含三个自模型： * first generates category-independent region proposals, these proposals define the set...

Fast R-CNN与RCNN SPPnet对比RCNN首先finetune，使用log loss。然后，使用SVMs来训练，最后，使用bounding-box regressor。代价大慢Fast R-CNN 模型结构和训练一张图片首先经过几个卷积层和池化层产生特征向量，然后 for each object proposal a region of inter...

Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognitioncrop & warpSpatial pyramid poolingIn this paper, we introduce a spatial pyramid pooling layer to remove the f...

Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks使用卷积层计算proposals On top of these convolutional features, we construct an RPN by adding a few additional convolutional l...

caffe train 源码int train() { // 查看是否提供了solver CHECK_GT(FLAGS_solver.size(), 0) << "Need a solver definition to train."; CHECK(!FLAGS_snapshot.size() || !FLAGS_weights.size()) <...