YOLOv8 object recognition model not suitable for mobile?

Unity Engine
,
1

Hello, I’m using the Unity Yolov8 example on Hugging Face Found Here to create a real-time mobile object recognition application, this was a great starting point.

I connected a video of my own to the project to test it out, and it seems that Sentis recognition is really resource intensive. I added an FPS checker just to see how hard the lag was hitting (it looked like a slide show on my phone), and it goes down form ~121FPS to 27FPS in the editor, and down from 60 FPS (capped) to anywhere from 1-3FPS.

Am I using the project wrong? The only changes I made were substitute the video that was there for my own, add the yolov8n asset to the Resources folder, then link it via Resources.Load. Is there a better way to get object recognition running on mobile?

Let me know if there’s anything I can provide to make this easier, thank you so much for all your help!

2 Likes
2

This is most likely related to the NMS.
We’ll have a performance improvement for that in an upcoming release.
This task is known internally as Task 211.

1 Like
3

Hello ,
Can you please share the video of your performance? We are also doing the similiar task but with the camera video.Not sure if it is also possible. All examples are based on static images and video.
Regards,

4

Hello, I’m using camera video too but the framerate is approximately 0.5FPS, so I’m waiting for the performance improvement that ilutaurasvysniauskas_unity mentioned

1 Like