Is it possible to run your model on a Smartphone ?

[Avant-Gardiste]

Hi, I was wondering if it might be possible to run your model even on the limited computing power of a smartphone. Before diving too deep into it, I wanted to get the opinion of more experienced people in the field.

If you think it is possible, is a rough estimation for fps possible ?

If you think it is not possible could you briefly elaborate on the reasons please ?

Any insights are appreciated. Thank you very much!

[thohemp]

Yolov5 is too big to run it on a smartphone in reasonable time. I would suggest to train it on Yolov5-tiny.

[Avant-Gardiste]

Thank you, I'll try that ! Otherwise, I think your repo is incomplete, I'm not enable to start training with your model because a missing file is needed ! @thohemp could you answer please ? I already opened an issue about this !

[thohemp]

I added the related files. Please try again.

[Avant-Gardiste]

Thank you so much !

[Avant-Gardiste]

Hi Mr. @thohemp, I want to let you know that I successfully deployed OBB model on an Android device ! Now, I need to optimize my model in order to increase the speed and latency, and from our last discussion, you suggested that I should use YoloV5-tiny. But, just out of curious, I did not find the Tiny model in your models repo, and same for weights, I just want to know if you had done this project before the Tiny version was released, or, you just forgot to push them into the repo ?
Thank you in advance and sorry for borthering you again !

[thohemp]

Great stuff!
In my case, the performance of the smaller models weren't satisfying for me, so I stuck with the m model. But depending on the specific task, setting and requirements one of the smaller models (5n6, 5s6) may fit your needs.

[thohemp]

The models are defined here: https://github.com/thohemp/cube_detector/tree/master/yolov5/models
Changing the config for training should do the magic. E.g: --cfg yolov5n.yaml