Example

The following example shows how to use the pre-trained neural network on an unlabeled sample video. To follow along, ensure that you have read the Installation section and set up the required environment.

First, download the sample video 2022-08-19T14-07-11--Basler acA720-520uc-x264.mp4 (68 MiB) from .

Note: The video is about 44 seconds long and shows a mouse running on a wheel. The playback speed of the video has been slowed down by a factor of ~30 for illustrative purposes.

Next, open an Anaconda terminal and activate the KineWheelSystem Anaconda environment:

conda activate kwa

In the Anaconda terminal, navigate to the Inference folder in the repository root folder.

cd C:\repository\root\KineWheelSystem\Inference

Note: Adapt the command lines below to reflect the proper locations of the downloaded sample video and the DLC config file on your system.

Run inference.py to predict paw locations in the sample video using the pre-trained neural network.

python inference.py predict -v "c:\path\to\sample\2022-08-19t14-07-11--basler aca720-520uc-x264.mp4" -c "c:\repository\root\dlc-skeleton\config.yaml"

Finally, create the labeled video based on the predictions made above.

python inference.py label -v "c:\path\to\sample\2022-08-19t14-07-11--basler aca720-520uc-x264.mp4" -c "c:\repository\root\dlc-skeleton\config.yaml"

The labeled video (as well as the raw predictions in CSV and HDF5 format) can be found in the folder of the sample video.

For reference purposes, the labeled video and raw predictions are also provided on Zenodo (see link on top of page).