SFD speed ?

Nice move to include SFD detector!
How fast is your PyTorch version of SFD? I remember trying some earlier port of SFD to PyTorch and it was ~10x slower than original Caffe implementation.

The issue asks about the speed of the PyTorch SFD detector compared to the original Caffe version. Start by locating the SFD detector implementation in the codebase, likely in a module like `face_alignment/detection/sfd`. Run benchmark tests to measure inference time, comparing against any available Caffe baseline or reported figures. Check for existing performance profiling scripts or consider writing a simple timing script that uses the detector on sample images. 'Done' would be a clear report of the current speed and identification of any bottlenecks.

索引モデルが issue の本文から書いたものです。