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
1.18k stars 316 forks source link

yolov4 inference speed is slow #164

Closed Tristesse-stk closed 2 years ago

Tristesse-stk commented 2 years ago

Hi, thank you for sharing a great open source project about TensorRT. There is a question after I run the sample_detector with yolov4 successfully. The inference time I tested is longer than the time you mentioned in your benchmark. My environment: Jetson XAVIER NX TensorRT7.1.3 cuda10.2 cudnn8.0.0 opencv4.1.1

My result table is below.

精度 | 图像尺寸 | 速度 -- | -- | -- FP32 | 416*416 | 180ms FP16 | 416*416 | 90ms

It is seen that it is slower than yours. Image

If you can help me suggest some possible solutions, I will be very grateful and give you a star.

enazoe commented 2 years ago

@Tristesse-stk modify the power mode to 2 core 15W, and turbo the cpu

Tristesse-stk commented 2 years ago

Great！Your solution is helpful and the inference speed is as fast as yours.I just gave the star to your project.

denred0 commented 2 years ago

Hi! Can you say how to "modify the power mode to 2 core 15W, and turbo the cpu". ? RTX 2080 Ti TensorRT 8.0.6 CUDA 11.3 cuDNN 8.2.1

Thank you!

Tristesse-stk commented 2 years ago

In the upper right corner of the interface. MODE_set

Nuzhny007 commented 2 years ago

@denred0 but it works only for Nvidia Jetson

denred0 commented 2 years ago

@Nuzhny007 Yes, I understood