I want fast fps custom object detection. Which model should I use for transfer learning?

Autonomous Machines Jetson & Embedded Systems Jetson Nano
1

Dear Sir,

I am new to Jetson Nano, I want to train a custom object detector. I want fast FPS. Which model should I use for transfer learning on my own dataset ?

Moreover, can I train the model on Nano itself (I guess NO). Please provide me any tutorial link.

Thank you

3

Hi,

You can choose the model based on benchmark result (fps and accuracy) here:

For training, please apply the transfer learning on a desktop GPU.
Thanks.

4

Thank you @AastaLLL
I have trained SSD Mobile NET V2 using Transfer Learning.

Please provide me link from where I can convert it into tensorRT model.

Will Tensor RT affect it’s accuracy ?

Moreover for conversion, should I use Desktop or Nano itself ?

Thanks

5

Hi,

In general, TensorRT won’t affect accuracy.
For tlt model, you can check this GitHub for the conversion steps to TensorRT engine:

And this conversion need to be applied on Nano directly since the engine is not portable.
Thanks.