Device Argument Not Working as Expected

[timlac]

Device Argument Not Working as Expected (cuda:0)

Issue 1: --device cuda in the CLI does not actually enable GPU in image mode

In process_image(), the device passed from the CLI is ignored because the function resets it internally:

def process_image(..., device='cpu'):
    # Initialize models with specified device
    device = 'cpu'   # ← this line overrides the CLI argument

As a result:
openface detect image.jpg --device cuda
still runs on CPU.

Issue 2: cuda:0 is rejected in LandmarkDetector

The constructor expects device="cuda" and a separate device_ids=[0], which is not how PyTorch normally handles device strings. Passing a standard string like cuda:0 results in:

Error: Invalid device type 'cuda:0'. Use 'cpu' or 'cuda'.

Just passing --device cuda results in:

Error: When using 'cuda', provide at least one valid device ID.
Aborted! 

Expected Behavior

--device cuda:0 should be accepted and run on GPU 0.

process_image() should not override the device passed from CLI.

Suggested Fixes

Remove or modify the hardcoded device='cpu' line in process_image().

Change device checks to device.startswith("cuda") to support cuda:0, cuda:1, etc.