XianyanLin
@luoyetx 我也在看[FaceDetect/jointCascade_py](https://github.com/FaceDetect/jointCascade_py) 的代码 运行过他的测试demo 速度有点慢 一张图花了好几秒 不过他的训练数据太少 不知道有没有关系 另外觉得你的代码比那个python代码的可读性高一些 我会跟进读懂你的代码 有什么问题也可以讨论一下 最后 还是要感谢你把代码开源出来 谢谢！
I can link caffe.lib successfully but cause an error (something like caffe parser errors) during runtime.
Thank you very much!
就觉得怎么抖得那么厉害，改完以后好很多
Data layout in caffe follows `NCWH`, but opencv is `CHW`. Before forwarding there is a [transpose](https://github.com/foreverYoungGitHub/MTCNN/blob/master/detection/MTCNN.cpp#L129), so what you get then is all transposed.
it it not a real PReLU as caffe, it is a `leakyrelu` with constant neg-slop.
> 在styling里面加上这几行就可以解决了，是一个lazysizes的问题，具体参见[aFarkas/lazysizes#508](https://github.com/aFarkas/lazysizes/issues/508) > > ```css > img[data-sizes="auto"] { > display: block; > width: 100%; > } > ``` 请问具体是哪个文件里改呢？
the same problem
https://github.com/GarrickLin/numpy-opencv-converter I have adapt this repo to opencv3