Kolkir/mlcpp

Speed of Mask R-CNN During Inference

canerozer opened this issue · 1 comments

Hello, thank you for creating this project.

I see that demo.cpp has been designed for single-image inference, however can it be modified for inference of multi-images, let's say that images in a particular folder? I'd like to see if there is an improvement or any changes in FPS rate to compare with Detectron and matterport's Mask R-CNN.

Thank you.

Sorry but I don't have plans to implement such functionality, it should be easy to add a loop for this. But there is an another problem - this implementation support batch sizes of size one only.