search

ShiqiYu / libfacedetection

An open source library for face detection in images. The face detection speed can reach 1000FPS.
Other
12.27k stars 3.05k forks source link

FLOPs #306

Closed nttstar closed 2 years ago

nttstar commented 3 years ago

Hi, how about the FLOPs of your new v3 CNN model(under 640x480 input)?

fengyuentau commented 3 years ago

For CONV layers, we calculate FLOPs as the following:

FLOPs_conv = (2 * C_in * K * K - bias) * (C_out // groups) * H_out * W_out

So the FLOPs of CONV layers in our 85K-param model is about 0.767 GFLOPs under a 640x480 input.

More details will be released on https://github.com/ShiqiYu/libfacedetection.train once we clear our code.

nttstar commented 3 years ago

@fengyuentau Thanks! So for pytorch style computation, it should be around 0.38G FLOPs.