Object Detection Model Sizes and Real-Time Performance on Jetson

Hello,

I am currently working on a project where we are evaluating the object detection models provided by StereoLabs. I noticed that there are different model sizes available.

Could you please explain why there are different model sizes for the object detection models, such as MULTI_CLASS_BOX_FAST, MULTI_CLASS_BOX_MEDIUM, and MULTI_CLASS_BOX_ACCURATE?

Furthermore, are the smaller models specifically designed and optimized for real-time use, especially on embedded devices like the NVIDIA Jetson platform?

Best regards,
Petter Dignes

Hi @Pettedig
Bigger models are more accurate, but require higher computing time.
You must select the model that best fulfills your requirements in terms of detection frequency and detection accuracy.

Yes, this is the case