FPN
FPN copied to clipboard
How many iterations of training?
您好,非常感谢大神的代码!在voc2007 数据集上训练,但是mAP 只有72,请问一下你迭代了多少次呢?实验中迭代次数是7.5万次,如果换成COCO 的数据集?应该在哪里改anchor 和ratio 呢?是在配置文件中吗?
readme已经写了,用voc2007+voc2012训练 训练八万次即可
谢谢大神的回复,我对您的readme 里面有句话不是很理解 in the paper the anchor setting is Ratios: [0.5,1,2],scales :[8,]
With the setting and P2~P6, all anchor sizes are [32,64,128,512,1024],but this setting is suit for COCO dataset which has so many small targets.
But the voc dataset targets are range [128,256,512].
So, we desgin the anchor setting:Ratios: [0.5,1,2],scales :[8,16], this is very import for voc dataset.
我看了config.py 文件里面好像只有
__C.FPNRATIOS = [0.5,1,2]
__C.FPNRSCALES = 2 ** np.arange(4,6)
这里的scale 是16、32 为什么是这个呢?
我对VOC 2007的变成进行了聚类,
预测的聚类中心个数:7
[[ 75.89890956]
[ 180.74493869]
[ 310.83273847]
[ 125.66768035]
[ 386.90149049]
[ 242.98615928]
[ 34.32092669]]
0: scale:75.8989095611243
1: scale:180.74493868691016
2: scale:310.83273846745465
3: scale:125.66768034606157
4: scale:386.90149049240097
5: scale:242.98615927978665
6: scale:34.32092668661748
如果实验中采用的是scales :16、32,那么在conv_2层的anchor 就会比较大 基础的anchor 会不会都是由前面conv_2 conV_3 输出的呢?
聚类中心不大能说明啥,应该说属于一个聚类的数量才是有意义的,真正的anchor大小是scale*stride 而voc以经验起见是在128,256,512中的较多。使用聚类策略分析是一种很好的方法,你可以接着分析,找到最合适的anchor设置。