search

michalfaber / keras_Realtime_Multi-Person_Pose_Estimation

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

What is the idea behind `scale_search` ? #134

Open mazatov opened 4 years ago

mazatov commented 4 years ago

I'm wondering what is the idea behind looping over scale_search in evaluation? I looked through the paper but couldn't find a good explanation on why we use scale_search = 0.5, 1, 1.5, 2.

Moreover, if we know for sure that the image contains only one human that is the size of the image, does it make any sense to have anything but scale_search = 1 ?