Question about model predictions

[clee758]

I was wondering how predictions of a model are generated? I was observing some of the error data points and their corresponding frame id from the excel sheets from the Jupyter notebook Liezl made for me, but there was no yellow box in the frame of the video, so I couldn't see the extent of the error. Is there a way to see the predictions for that frame?

[talmo]

Do you mind resharing that notebook (or the relevant snippets) for context?

Is this for error metric evaluation?

If so, you'll need ground truth (user labeled) data on some frames, after which you can generate predictions on those for subsequent evaluation.

[clee758]

Yes, this is for error metric evaluation.

When I go to the corresponding frame_idx in the video, in this case 1122, there is no prediction frame in the SLEAP project, even though there are metrics generated. Why is that?

[clee758]

[clee758]

Hi, I'm just checking to see if there is any update regarding the answer?

[roomrys]

Hi @clee758,

At least for the example given, the name for the video does not line-up across your slp file and the video name(s) listed in your spreadsheet:

csv or slp	video name
csv	...0230407.mp4
slp	...0230419_1.mp4

Are you sure you are looking at the correct video in the correct slp project?

Thanks,
LIezl