-
Hi, I started using sleap to perform some tracking analysis and I found the software great, in particular the GUI is really great. Great job. I have just a small problem, I can't get the prediction data in the hdf5 file. In particular, I'm using the single_animal model to track a single animal in the video. I perform the training and the inference, selecting even the flow tracker in the inference window of the GUI. But after inference I get just the instance prediction with no track prediction, and when I save the HDF5 analysis file the 'track' dataset is empty. I can get the track by using the command 'set instance track' of Tracks widget in the menu bar, but I should do this for each frame of the movie. I already solved the problem by using the command sleap-track directly from the terminal, but is there a way to do so with a single command from the GUI? Moreover, the problem does not appear if instead of the single animal model I use the top-down model. Am I doing something wrong? Help is much appreciated. Giorgio |
Beta Was this translation helpful? Give feedback.
Replies: 1 comment
-
Hi @GioLetti, For single animal projects, we suggest not using However, the logic for checking whether the user has a single animal project is not as direct as looping through all the user's labels to check that there are not more than 1 annotation per frame. Instead, we (in this case, incorrectly) assume that the In the GUI, can you Track > Delete Multiple Tracks > Unused. Save As. Retry the export. Sorry for the troubles! Please let us know if that helped. Thanks, |
Beta Was this translation helpful? Give feedback.
Hi @GioLetti,
For single animal projects, we suggest not using
Track
s at all. The main purpose ofTrack
s is to sort which pose estimation/annotation belongs to which animal throughout an entire video. However, if there is just a single pose estimation/annotation in eachLabeledFrame
, then we already know that the pose estimation belongs to the one and only animal being annotated (i.e. single animal project).However, the logic for checking whether the user has a single animal project is not as direct as looping through all the user's labels to check that there are not more than 1 annotation per frame. Instead, we (in this case, incorrectly) assume that the
Labels
/slp project will have noT…