Can't export all videos in csv due to memory error

[frenchnat]

Bug description

Hi, I get a memory error in Miniforge when trying to export the analysis csv for all my videos (n=71).

Expected behaviour

I would expect to get the analysis csv files for all my videos.

Actual behaviour

I got the analysis csv files for 5 videos then the previously reported error appeared.

Your personal set up

• OS:
  Edition Windows 10 Enterprise
  Version 22H2
  Installed on ‎10/‎01/‎2024
  OS build 19045.4046
  Experience Windows Feature Experience Pack 1000.19053.1000.0

• Version(s):
  SLEAP v1.3.3

• SLEAP installation method (listed here):

  • Conda from package

Environment packages

# paste output of `pip freeze` or `conda list` here

Logs

# paste relevant logs here, if any

Screenshots

Traceback (most recent call last):
File "C:\mambaforge\envs\sleap\lib\site-packages\sleap\gui\app.py", line 499, in
lambda: self.commands.exportCSVFile(all_videos=True),
File "C:\mambaforge\envs\sleap\lib\site-packages\sleap\gui\commands.py", line 337, in exportCSVFile
self.execute(ExportAnalysisFile, all_videos=all_videos, csv=True)
File "C:\mambaforge\envs\sleap\lib\site-packages\sleap\gui\commands.py", line 242, in execute
command().execute(context=self, params=kwargs)
File "C:\mambaforge\envs\sleap\lib\site-packages\sleap\gui\commands.py", line 138, in execute
self.do_with_signal(context, params)
File "C:\mambaforge\envs\sleap\lib\site-packages\sleap\gui\commands.py", line 162, in do_with_signal
cls.do_action(context, params)
File "C:\mambaforge\envs\sleap\lib\site-packages\sleap\gui\commands.py", line 1157, in do_action
video=video,
File "C:\mambaforge\envs\sleap\lib\site-packages\sleap\io\format\csv.py", line 69, in write
csv=True,
File "C:\mambaforge\envs\sleap\lib\site-packages\sleap\info\write_tracking_h5.py", line 397, in main
) = get_occupancy_and_points_matrices(labels, all_frames, video)
File "C:\mambaforge\envs\sleap\lib\site-packages\sleap\info\write_tracking_h5.py", line 122, in get_occupancy_and_points_matrices
(frame_count, node_count, 2, track_count), np.nan, dtype=float
File "C:\mambaforge\envs\sleap\lib\site-packages\numpy\core\numeric.py", line 343, in full
a = empty(shape, dtype, order)
numpy.core._exceptions.MemoryError: Unable to allocate 13.8 GiB for an array with shape (54372, 1, 2, 17030) and data type float64

How to reproduce

1. Go to 'File'
2. Click on 'Export Analysis CSV...'
3. Scroll down to 'All Videos...'
4. See error

[roomrys]

Hi @frenchnat,

The export for pose estimation is a bit clunky in the sense that it tries to allocate space for an array of shape (frame_count, node_count, 2, track_count), populate this array, and then discard any unused portion of the array (i.e. any unused Tracks). You can imagine that since SLEAP doesn't reuse Tracks or automatically delete unused Tracks, this shape can get pretty massive. In your case, you seem to have 17030 Tracks - I doubt all of those are being used.

To get things exporting, I would try deleting all the unused Tracks (in the GUI menubar: Tracks > Delete Multiple > Unused), saving the project as, and retrying.

Let us know if that worked.

Thanks,
LIezl

[frenchnat]

Hi @roomrys,

Thanks for your reply! It didn't solve the issue because I guess I have too many frames where sleap has tracked instances where there was no animal to track (the videos are not great ahah). I was planning on cleaning the data in the csv files but I guess I need to clean it before exporting then. Would you recommend a way to do this that doesn't require me to manually check every frame of all my videos?

Thanks a lot,
Natacha

[talmo]

Hi @frenchnat,

As mentioned, the fundamental problem is that there are too many tracks predicted. You can get around this by limiting the max number of tracks being produced during inference via the GUI options or CLI.

Let us know if this is still an issue for you!

Cheers,

Talmo

[frenchnat]

Hi @talmo,

I ended up trimming my videos with Premiere and limiting the max number of tracks to 1 and it seems to be able to export the data now.

Thanks for your help!
Natacha