[BUG][CLI] Fix error in cli script

[surya3214]

Description

Training with coco_dir option fails due to a small bug from L#52

NameError: name 'device_supports_cuda' is not defined

Type of change

• Bug fix (non-breaking change which fixes an issue)

How has this change been tested, please provide a testcase or example of how you tested the change?

"python3 cli/main.py --coco_dir " won't fail anymore due to "NameError" from "device_supports_cuda"

Any specific deployment considerations

None

Docs

None

[CLAassistant]

All committers have signed the CLA.

[surya3214]

@yeldarby @probicheaux I'd appreciate if you could please check this

[nok]

Hello @surya3214 , alternatively you can use the variable DEVICE from the config.py#L11:

def train_from_coco_dir(coco_dir: str):
    from rfdetr.config import DEVICE

    rf_detr = RFDETRBase()
    rf_detr.train(
        dataset_dir=coco_dir,
        epochs=1,
        device=DEVICE,
    )

[surya3214]

from rfdetr.config import DEVICE

Thanks @nok I've updated it. I see device_supports_cuda() is used in train_from_rf_project() as well. Should that be updated too?

[Abdul-Mukit]

Why should it not be defined by the user?
e.g. In most CLI commands for DETRs we can usually set it by --device option.

parser.add_argument('--device', default='cuda',
                        help='device to use for training / testing')

[surya3214]

Why should it not be defined by the user? e.g. In most CLI commands for DETRs we can usually set it by --device option.

parser.add_argument('--device', default='cuda',
                        help='device to use for training / testing')

This PR only keeps the status quo wrt device selection and avoids build error.

While that suggestion is really useful, I shall consider it in a different PR. WDYT?