Hey @agosztolai

I refactored a bit your method, hope it's okay for you 😀.

I am testing it now with a simple Sleap project I created, but it doesn't work for now on my side: it basically never enters inside this condition, and, having manually checked, I don't see anything like this in my Sleap file. My Sleap file contains the following keys:
<KeysViewHDF5 ['frames', 'instances', 'metadata', 'points', 'pred_points', 'suggestions_json', 'tracks_json', 'videos_json']>.
The dataset under 'videos_json' does store the path to the video I used when creating my Sleap project.
Therefore video_names is empty and nothing is done inside this whole for loop neither.

Is there an issue with your code or did I do something wrong when creating my Sleap project? I am new to Sleap so maybe the mistake is on my side.
Or maybe we have different Sleap versions, which store data differently? I currently have SLEAP v1.4.1a2

Hello @maximpavliv ,

I can confirm that the code works on SLEAP v1.4.1a2, but it needs the .pkg.slp file from 'Predict - Export Labels Package... - Labelled frames' instead of the .slp project file.

By the way, I have a couple of suggestions:

1. The line 527 and 528 should be instance_dataset[i] instead of instance_dataset[i+1], the latter caused an out of range error on my end.
2. I recommend adding the line
   "Task": f"{scorer}-{sleap_file.split('labels.')[1].rsplit('.pkg.slp', 1)[0]}",
   (Assuming the sleap_file follows the format of /path/to/labels.xxx.pkg.slp)
   below config = { at line 421, without which will throw a KeyError: ‘Task' when attempting to load from the generated config.yaml.

A further issue: during the 'Analyze Videos' step in DLC, the project froze due to the generated config.yaml missing the multianimalproject: