Iris segmentation using UNET

[manuel.johnson]

I need to perform efficient iris segmentation from an eye. I would like to know the expected accuracy and speed of this task on a Jetson device. Additionally, I would appreciate guidance on which model to use and whether it is available in the TAO (Train, Adapt, Optimize) toolkit.

[Morganh]

For segmentation network, you can use Mask_rcnn or Unet network.
They are available in TAO.

[manuel.johnson]

Thanks for the replay,
Can you please give me an idea about accuracy and speed in the case of UNET and MaskRCNN ?
SOM: Jetson Xavier

THANKS IN ADVANCE :)

[Morganh]

You can find the purpose-built models based on these two network in
https://docs.nvidia.com/tao/tao-toolkit/text/model_zoo/overview.html

PeopleSegNet
PeopleSemSegNet

[manuel.johnson]

Can we do any boundary level segmentation with any of the networks like MaskRCNN and UNET or TAO have any other method for that ?

[Morganh]

In TAO, there are Mask_rcnn, Unet and SegFormer.
https://docs.nvidia.com/tao/tao-toolkit/text/semantic_segmentation/index.html
https://docs.nvidia.com/tao/tao-toolkit/text/instance_segmentation/index.html

[manuel.johnson]

Thanks Morganh

[manuel.johnson]

Is there any tool for creating annotation in segmentaion from nvidia side.

[Morganh]

@manuel.johnson

creating annotation

There is not label tool from TAO.
You can refer to Sloth and Label-Studio tools.

[manuel.johnson]

Thanks