中国斜学我术大草 实验结果: 为了更好的观察focal loss在reweighting example的效果,作者随机选取了l07个负样本框和10^5个 正样本框,然后通过网络之后分别计算这些正负样本的1oss,最后,分别对于正样本和负样本,把所有 框的1oss进行归一化(softmax),画出累计loss随样本数目的增长曲线。 y=0 y=0 0.8 y=0.5 0.8 y=0.5 Y= Y= 0.6 y-2 sso]pazilewjou 0.6 y-2 0.4 0.4 annejnwno 0.2 anne]nwno 0.2 0 0 0 2 .4 .6 .8 0 2 .4 .6 .8 fraction of foreground examples fraction of background examples Figure 4.Cumulative distribution functions of the normalized loss for positive and negative samples for different values of for a converged model.The effect of changing y on the distribution of the loss for positive examples is minor.For negatives,however,increasing y heavily concentrates the loss on hard examples,focusing nearly all attention away from easy negatives.实验结果: 为了更好的观察focal loss在reweighting example的效果,作者随机选取了10^7个负样本框和10^5个 正样本框,然后通过网络之后分别计算这些正负样本的loss,最后,分别对于正样本和负样本,把所有 框的loss进行归一化(softmax),画出累计loss随样本数目的增长曲线