Is there an existing issue for this?

• I have searched the existing issues

Bug description

When using a top-down pytorch model the detector is not running on the GPU when calling analyze_videos, at least for a model trained via memory replay. This is true even when providing a specific device via the device= pytorch parameter.

If I hack a line to specifically pass the device to utils.get_detector_inference_runner at line 398 of pose_estimation_pytorch/apis/analyze_videos.py then the detector does use the supplied GPU.

Operating System

Ubuntu 22.04

DeepLabCut version

3.0.0rc6

DeepLabCut mode

single animal

Device type

Nvidia RTX A5000

Steps To Reproduce

Train a top-down model (I used fine tuning with memory replay from the mouse superanimal model), then call analyze_videos.

Relevant log output

Anything else?

No response

Code of Conduct

• I agree to follow this project's Code of Conduct