shicai
shicai
simply because i don't like the floor/ceil mode of conv7x7/s2 and pool3x3/s2 in caffe. ``` for input size = 225: output size of conv7x7/s2 = (225 - 7 + 2*3)...
yes. it is automatically configured by caffe: in train phase, `use_global_stats` is false, and in test phase, it is true.
@sophiazy 可以，把那一类的pos_mult设置为小于1.0的数值
按类别编号来
是的，应该是channels。 这个repo太老了，原以为没人关注，就懒得更新了。
vision_layers.hpp只是文件名没有了，其中的内容被拆分到各个单独的layer头文件中了，你自己单独建立一个新的头文件，然后在cpp和cu里include就行了
@liuzhuang13 I just tested the pretrained DenseNet-121 caffe model from Zhiqiang Shen, and the top-1/5 acc by using single center crop is only 70.8/90.3%(256xN) or 72.5/91.3%(256x256). is there something wrong?...
@liuzhuang13 thank you for your quick reply. I just manually converted the torch model into caffe format. due to some unknown reasons, the acc is about 0.2~0.3% lower than the...
@Tongcheng thanks for sharing. @liuzhuang13 I made my repo public, see: https://github.com/shicai/DenseNet-Caffe everyone can download these caffemodels freely. hope it helps.
yes, you are right. caffe use floor for conv layers, and ceil for pooling layers.