search

jkjung-avt / tensorrt_demos

TensorRT MODNet, YOLOv4, YOLOv3, SSD, MTCNN, and GoogLeNet
https://jkjung-avt.github.io/
MIT License
1.74k stars 545 forks source link

How come we have low FPS even after optimizing the models #524

Closed abhinav0927 closed 2 years ago

abhinav0927 commented 2 years ago

Hi

I was going to follow your project and then read the FPS you have mentioned after optimizing the models. How come we are facing low FPS even after optimizing the models. ?

jkjung-avt commented 2 years ago

What platform are you testing the code with? What FPS number are you expecting? And what's the result?

abhinav0927 commented 2 years ago

Hi, I am planning to test it on Jetson Nano. I havent tested it but was going through your documentation and realized that you mentioned FPS to be around 10 which I feel is a bit low. I am expecting the FPS to be around 25-30. Do you think is that possible?

I am already getting 30-32FPS with SSD Mobilenet models but I feel SSD detection accuracy is a bit low and it cannot perform detection on smaller objects, thats why I want to switch to other models like Yolo

jkjung-avt commented 2 years ago

Please check out the following posts about my opinions on FPS. If FPS is really important to your application, you might consider adopting DeepStream instead.