That would be greatly appreciated!

From my basic understanding, it's a mix of num_frames x num_components, 1 x num_components, 1 x num_frames, and potentially img_height x img_width x num_components output arrays that are usually used (at least for the output data used in my lab), where num_components could be in the range of 100s.

Correct me if I'm wrong, I looked quickly through ndx-pose and the PoseEstimationSeries only saves either num_frames x 3 or num_frames x 2, which might not be a good fit for facemap outputs.

@tuanpham96 what does "component" mean in this context? I was expecting num_frames x num_nodes

from my understanding, it's the components (default = 500) when using the SVD option on the face region of the video. E.g. component 1 could represent some pixel covariation that's related to whisking motion, component 2 could represent some other pixel covariation that's related to nose motion, ... but nose and whiskers are not marked like in pose estimation. I think there's some option with pose estimation and tracking as well but we didn't use that (yet) in the lab.

One of the outputs array would be a time series of each of these components. For example, in cell 18 of their tutorial notebook.

Another would be an image mask that represents each component, see cell 17 of their tutorial notebook.

@tuanpham96 I see. I was more thinking about capturing the pose estimation, e.g. this image from the readme:

I see that this software also does SVD of the face, so we should think about how best to support both.

@Atika-Syeda or @carsen-stringer, would one of you mind sharing small example output file(s)? Ideally <10MB, as that will help us integrate this into our testing suite. I see a reference to test_cam1.avi and test_cam2.avi here but I can't find those files.