Slow inference runtime? about hand-cnn HOT 4 CLOSED

Hello. The inference time is around 2-3 frames/ second on Titan X.

from hand-cnn.

hi
How can I make it faster about inference time?

from hand-cnn.

The current architecture uses ResNet101 backbone, so it's a big network and the inference time is high. If you want low inference times, you can trade-off the performance using lighter backbones such as ResNet-50. In that case, you should train a new architecture with a new backbone.

from hand-cnn.

thanks for your reply
then how can I make mask-rcnn model off about masks part?
I mean I just want to use this model for only getting rectangle hand bbox
I really appreciate for your help

from hand-cnn.