Using deeplabcut with prelabeled data

I have a video of 100,000 frames of a spider building its web. I’m trying to track the spider’s legs as it builds.
I have labeled 10,000 frames of the video already, for use with other tracking methods.
I’d like to use the same 10,000 frames with the same labels in deeplabcut.
How can I create a labeled training dataset without using the label_frames() function?
I’ve tried putting all my label data into the “CollectedData_.csv" file but that seems insufficient to load the labels. The structure of the "CollectedData_.h5” file is not intuitive but I imagine I’ll need to create a similar *.h5 file with my label data.

With which method did you label your data and in which format are your labels now?

the .h5 is what is used, the .csv is just the human readable format. If you put your data into tthe .csv format, you can use the fucntion: deeplabcut.convertcsv2h5 to make the .h5

Exactly what I needed. Thanks!

We are having the same issue but the convert code is giving the error “Attention: Folder does not appear to have labeled data!”
We used the extract frames code to extract all of the frames in the video then ran the label frames code and did not label any frames to produce a blank file. From here we copied our prelabeled data into the CollectedData_ .cvs file and changed the path of the config file to that folder.

Update: we got it to work by changing the config.yaml file for the path of the video. Now the issue is that in the label_frams command when the video is loaded in, the coordinates for the Y-axis are flipped where 0 starts at the top.