Decrease inference time

[SaharGezer]

Can it run with higher FPS, What can I change in order to improve the inference time?

Thank you

[bladeszasza]

Hello @SaharGezer
at a fast solution I would advised to use the smaller models google/owlv2-base-patch16-ensemble and the facebook/sam2.1-hiera-tiny

you are also able to specify the sampling fps and whatever it should use GPU

| --fps | (Optional) Frame sampling rate (frames per second) for video inputs.
| --device | (Optional) Compute device ("cuda" or "cpu").

but as an honest answer I have lot of stuff run in sequence yet, so paralelization is promising processing boost. I have in plans also to extend our SAM family with (TAM)[https://github.com/yformer/EfficientTAM].

thanks for raising it!

[SaharGezer]

I used your project and it works great, I will try to change the owl model and update you about the results.

Edit:
It still takes 1.5 sec for one image @bladeszasza

[bladeszasza]

This is the next on my plate to optimize the code and function calling in order to reduce the inference time let you know once the PR is merged