search

YonghaoHe / LFD-A-Light-and-Fast-Detector

LFD is a big update upon LFFD. Generally, LFD is a multi-class object detector characterized by lightweight, low inference latency and superior precision. It is for real-world appilcations.
418 stars 82 forks source link

why trt mode consumed more GPU and MEM space？ #45

Open chainfitness opened 2 years ago

chainfitness commented 2 years ago

thx this awesome work! I just want to deploy this awesome work in practice, but i find that trt mode consume more GPU and MEM space.The GPU refered above is Tesla T4, i'd appreciate if the prompt repley.

YonghaoHe commented 2 years ago

@chainfitness We do not encounter your case before. Would you provide more information?

chainfitness commented 2 years ago

I'm obilged for quikly reply.I want to deploy face detection from this work to nvidia tx platform. But the official installing TensorRT version from TX is 8.0.1.6,which is too new to run predict_tensorrt.py. So,i move code to Tesla T4 that installed TensorRT 7.2.2.3, it works! But i find that trt mode consumed more GPU and MEM space, GPU and MEM consumption running predict.py was 977M , 2123M respectively,however, running predict_tensorrt.py was 1486M and 3714M. The pretrained weight was downloaded from image predict_tensorrt.py is only modified the path of pretrained weight.