Running two models in parallel on jetson nano

Autonomous Machines Jetson & Embedded Systems Jetson Nano
1

Dear all,
I want to be able to run two image models: One classification and the other regression at the same time. I looked in the tutorials but I couldn’t find an example for that. Can you please help me/point me in the right direction?
Thank you very much,
Luis

2

Hi,

What kind of framework do you use?
If you are using TensorRT, you can just launch them at the same time.

Thanks.

3

Would it be possible in PyTorch? Say I took a resnet18 model like the one shipped in the Jetson Nano sample codes -

        model = torchvision.models.resnet18(pretrained=True)
        model.fc = torch.nn.Linear(512, len(dataset.categories))

and in training I train for 2 outputs, one for regression -

        for images, category_idx, xy in iter(train_loader):
            # send data to device
            images = images.to(device)
            xy = xy.to(device)

and another for classification -

        for images, labels in iter(train_loader):
            # send data to device
            images = images.to(device)
            labels = labels.to(device)

Is this approach even possible in the PyTorch framework? I am guessing it is not, but could use some pointers.

5

Hi,

Suppose yes.
You can treat resnet18 as a feature extractor.
And add one fully-connected layer for the regression output and another one for label output.

However, this issue should be independent to the Nano device.
Please use a desktop environment for training task.

Thanks.

6

Will do, thanks!