search

ceccocats / tkDNN

Deep neural network library and toolkit to do high performace inference on NVIDIA jetson platforms
GNU General Public License v2.0
717 stars 209 forks source link

FPS Results using TX2 - Jetpack 4.4 #158

Closed andressasuiti closed 2 years ago

andressasuiti commented 3 years ago

Hi! I would like to know if you have any FPS results from yolov4 using Nvidia TX2, but with Jetpack 4.4. I trained my own network and the results I got were between 1 and 3 FPS slower than what you got with Jetpack 4.2

mive93 commented 3 years ago

Hi @andressasuiti Can you give more details? With which model, with which mode (FP32, FP16)? Have you enabled jetson_clock?

garyzf commented 3 years ago

I get 86-87 fps using nvidia tx2

garyzf commented 3 years ago

yolov4-tiny(320*320) , with Jetpack 4.2

garyzf commented 3 years ago

fp32,B=1

andressasuiti commented 3 years ago

Hi @mive93 I enabled jetson_clock -m 0 and tested the following resolutions: 416x416, 512x512, 608x608, and 736x736, both with FP32 and FP16. My fastest result was around 9 FPS and the slowest was 2.5 FPS. This using yolov4 and batch size 1. Is it normal?

MuhammadAsadJaved commented 3 years ago

Hi, Can anyone explain how we can know about the FPS during demo? It only show the output video but does not print FPS,

I am using Yolov3 using ./demo

Thanks

mive93 commented 3 years ago

@andressasuiti then I don't know. Here you can find further and deeper results for Tx2 with Jetpak 4.4, where the FPS are real ones (end2end, not only inference). Hope it helps.

@MuhammadAsadJaved, just let the demo finish (or stop it with CTRL+C) and you will get the numbers.

mive93 commented 2 years ago

Closing for inactivity. Feel free to reopen.