CPU occupancy is higher when using GPU nms instead of CPU nms during training process

endernewton / tf-faster-rcnn

Tensorflow Faster RCNN for Object Detection

https://arxiv.org/pdf/1702.02138.pdf

MIT License

3.65k stars 1.57k forks source link

CPU occupancy is higher when using GPU nms instead of CPU nms during training process #357

Open czx94 opened 6 years ago

czx94 commented 6 years ago

It's much faster to train the model with gpu nms but I observe that this will occupy more CPU ressources which is not so reasonable. Anybody knows how to explain this?

By the way, anybody knows whether it is necessary to use multi-process in nms process because I notice that it's much slower when using single cpu rather than multi cpus.

engineer1109 commented 6 years ago

GPU nms is different from CPU nms. CUDA and C++