I tested the algorithm with a V100 gpu and get about 60 ms for one prediction. This is with batch size 1. Are there any other ways to improve speed? Maybe changing iou threshold or nms threshold? I also tried using a pytorch implementation and it was about 5 ms.
I tested the algorithm with a V100 gpu and get about 60 ms for one prediction. This is with batch size 1. Are there any other ways to improve speed? Maybe changing iou threshold or nms threshold? I also tried using a pytorch implementation and it was about 5 ms.