search

mlfpm / deepof

DeepLabCut based data analysis package including pose estimation and representation learning mediated behavior recognition
MIT License
39 stars 6 forks source link

[JOSS] Interface with DLC #17

Closed cellistigs closed 1 year ago

cellistigs commented 1 year ago

I have some questions about how deepof interfaces with the outputs of DLC that I do not believe are addressed in the documentation or software paper:

  1. What versions of DLC is DeepOF compatible with/tested with? If the answer is all, it would be great to see that stated explicitly.
  2. What is the recommendation for usage of DLC-native post processing routines (e.g. simple smoothing, multi-animal based techniques) vs. deepof? Should a user apply both, or default to the choice of one?
  3. What portions of the code specifically will not function if a user does not have the pre-specified set of labels expected for deepof?

Thank you.

cellistigs commented 1 year ago

https://github.com/openjournals/joss-reviews/issues/5394

lucasmiranda42 commented 1 year ago

Dear @cellistigs,

thank you for your questions! Replying in order:

1) We tested DeepOF with single and multi-animal projects on DeepLabCut 2.X (up until the most recent version when writing this comment, 2.3.4). I added a statement in the README and the documentation landing page to clarify this :)

2) All experiments presented relegate preprocessing to DeepOF, ignoring DLC post-processing routines. I don't see a problem with applying both, however, but it has not been tested. Shall we clarify this in the documentation?

3) It depends. In the supervised pipeline, if key labels for certain behaviors are missing, those annotations will be skipped. In the unsupervised pipeline, graph representations (described in the corresponding tutorial) will not be available, and the user will need to rely on matrix representations alone.

Hope this helps! I'm closing the issue to mark it as taken care of for now, but of course feel free to comment / reopen if needed. Best! Lucas

cellistigs commented 1 year ago

Hi Lucas- thanks for your reply, that's great. I would suggest clarifying point 2 in the documentation. Likewise, I think that point 3 would be good to have stated somewhere (I was not able to figure this out from looking at the documentation or software paper).