- 博客(7)
- 收藏
- 关注
原创 OHEM论文解读
参考:http://blog.csdn.net/yaoqi_isee/article/details/63796423Training Region-based Object Detectors with Online Hard Example Mining论文解读:在FRCN中,定义background roi为和任意gt box的IOU在[bg_lo,0.5)的region propo...
2018-02-10 17:26:50 1587
原创 mAP,ROC
AP:precision:预测的正类中实际的正类/预测正类recall:预测正类中实际的正类/实际正类对于每一个检测的结果都会输出一个score值,我们都将它作为一个阈值,大于这个score值为预测正样本,小于这个score值为预测负样本,AP@IOU=0.5时,实际当框框与真值框大于0.5时为正样本,当框框与真值框小于0.5时为实际负样本,就得到一个recall和precisio
2018-02-05 10:16:29 746
原创 mxnet使用记录
1、在服务器上跑时,在要运行的py文件上,在import os和import mxnet前加上os.environ['MXNET_GPU_MEM_POOL_RESERVE'] = '95'能使模型所占内存更稳定,加上os.environ['MXNET_ENABLE_GPU_P2P'] = '0'2、训练的时候报错:terminate called after throwing an ins
2018-02-03 20:47:40 450
原创 Face Detection using Deep Learning: An Improved Faster RCNN Approach论文解读
Flowchart of the training procedureFirst of all, we train the CNN model of Faster RCNN using the WIDER FACE dataset [30]. Wefurther use the same dataset to test the pre-trained model so as t
2018-02-03 17:08:15 353
原创 mxnet中test时输出speed:samples/sec
mxnet中test时输出speed:samples/sec代表在所有gpu上每秒能训练的image数目
2018-02-02 20:59:30 1390
原创 Face R-CNN论文解读
网络结构:这篇文章在基于faster rcnn做了一些改进:First, based on a newly developed loss function called center loss [33], we design a new multi-taskloss function in the Fast R-CNN model to supervise the learning
2018-02-01 18:55:55 972
空空如也
空空如也
TA创建的收藏夹 TA关注的收藏夹
TA关注的人