michalfaber / keras_Realtime_Multi-Person_Pose_Estimation

Keras version of Realtime Multi-Person Pose Estimation project
Other
779 stars 372 forks source link

Why use float64 when predict? #111

Open DableUTeeF opened 5 years ago

DableUTeeF commented 5 years ago

In process() in demo.py, I found out that I can speedup this function from around 12s/img to 5.8s/img by add dtype='float32' in every np.zeros or np.ones. Hopefully this might help in some improvement.