Very Slow

[dexception]

The performance is even worse than Yolo which came good 5 years back.

[YoungJoon Yoo]

@dexception This PyTorch version is not speed-optimized (s.a. nms) and is basically a parameter efficient algorithm. Speed optimized version will be available this year.

[艾因•利奇曼@利奇曼冒险团]

The model design performs lower cost of memory but higher in FLOPS.

[long_nguyen]

I recomand repo: https://github.com/biubug6/Face-Detector-1MB-with-landmark