ZhiqiJiang
Results
6
comments of
ZhiqiJiang
Thank you @qiaoguan.
@ love1314pds No I didn't. Maybe I modified something correspondingly.
@love1314pds 我已经不止一次遇到跨框架引起精度损失这个问题了,就算是利用作者提供的pretrained model.我一直还没有耐下性子去定位分析乃至解决这个问题.
@chuanqi305 why don't convolution depth wise use engine: CUDNN instead of engine: Caffe? In the layers of conv1/dw and conv2/dw and so on in MobileNet-SSD/train.prototxt, setting engine: CAFFE that means...
Because P-net only have one pooling layer, namely shifting one step in output feature map is equal to shifting two step in input layer.