Allow Tracking-Only Pipeline (GUI & CLI)

[YiLin-Zhou]

Bug description

Hi, The tracking-only pipeline (running tracking on user-labelled or previously predicted instances without running inference first) is not working in the SLEAP v1.3.0 GUI and CLI. This issue was previously fixed for v1.2.4 here: #867.

Details

GUI: Running inference with Training/Inference Pipeline Type set to 'none' but Tracker Type set to simple or flow causes the GUI to state "An error occurred during inference. Your command line terminal may have more information about the error." The command line terminal shows the following:

Command line: Following the sleap-track documentation to run the tracking-only pipeline results in the following error:

My setup

SLEAP installation:

• Conda from package

Software versions:

• SLEAP: 1.3.0
• TensorFlow: 2.6.3
• Numpy: 1.19.5
• Python: 3.7.12

OS:

• Windows-10-10.0.19041-SP0

[roomrys]

Hi @YiLin-Zhou,

I have been looking into the problem (which seems to be a follow-up error to #1287) and am working on a solution.

Thanks for submitting the issue! Liezl

[bghogan]

Hello -

I'm running into the same problem - this makes it much more difficult to use! In fact, since I want to delete detections from some areas before tracking, it stops me being able to accomplish what I want at all. Are there any work-arounds? V 1.3.3

Ben

[bghogan]

Looking at the PR's above, it looks like the following may be a workaround:

"This mean that retracking can only be done on prediction-only projects until this PR is merged (i.e. users need to remove their user-labeled instances before retracking to avoid error)."

So, saving a copy of the dataset without user labels may allow tracking only. This isn't an great solution, but may allow me to iterate to at least find tracking parameters that in principle work well.

[maia-g]

I am also running into this same issue with trying to apply tracking after running inference and editing the predictions. I therefore have some user-labeled frames and some predicted instance frames (the majority are predicted). I have tried to run tracking-only on videos without user-labeled frames and it still does not work properly. I am using V1.3.3. Would love some input as to how to deal with this!