search

enazoe / yolo-tensorrt

TensorRT8.Support Yolov5n,s,m,l,x .darknet -> tensorrt. Yolov4 Yolov3 use raw darknet *.weights and *.cfg fils. If the wrapper is useful to you,please Star it.
MIT License
1.19k stars 315 forks source link

检测速度很慢 #10

Closed kunyao2015 closed 4 years ago

kunyao2015 commented 4 years ago

使用yolov3-tiny 进行测试，GPU：1050 ，单张测试图片检测速度 竟然是12221ms， 明显感觉不对，是什么其他地方要注意吗？

Loading pre-trained weights... Loading complete! Using previously generated plan file located at ../configs/yolov3-tiny-kFLOAT.engine Loading TRT Engine... Loading Complete! Label:2 className:car Label:7 className:truck Label:16 className:dog detect time = 12221ms id:2 prob:0.725098 rect:[220 x 89 from (466, 82)] id:7 prob:0.584577 rect:[99 x 67 from (530, 94)] id:16 prob:0.831854 rect:[257 x 299 from (124, 218)]

enazoe commented 4 years ago

@kunyao2015 这个是第一帧gpu cuda启动得问题，你试下多检测几张，第一帧后面得帧应该就正常了

kunyao2015 commented 4 years ago

谢谢！

enazoe commented 4 years ago

@kunyao2015 如果觉得有用点个star 哈😄