mxnet-ssd
mxnet-ssd copied to clipboard
how to reach the absolute forward time?
hi, I am trying to understand how to calculate the actual forward time, as you did...
One way is with the demo.py code, which is not ideal, since it does not offer batch sizing..
The other way is through the 'evaluate' function... But, the calculated time is for IO + Forward + NMS.. Is there something I've been missing?
p.s I tried to find where exactly the nms calculation takes place and had no success.. I'm wondering, do you have the possibility to eliminate bboxes according to a score_thresh, before entering the NMS part? In my experience with google's detection-api, this accelerates a lot the inference time
Thanks :)
You can extract the forward code from the evaluater I wrote, disable nms by setting threshold to 0, and feed in fake data.