Multi-Whisker Tracking

Hello, this is my first post in this community forum and I would be super thankful if you could give me some good advice.

In my setup, I record the whiskers of freely moving mice with high-speed cameras to observe and analyse the interaction with certain textures (contact times, whisker cycles, whisker kinetics, etc.). Attached you will find an exemplary image for such a marking of whiskers (in the meantime the background lighting has been optimised so that the whiskers can be seen a bit better, the electrode interface board [EIB] unfortunately sometimes blocks the view a bit).

I am now a bit unsure which settings would be the most suitable to meet these requirements and thought that I might be able to save myself a lot of optimisation work if someone here has already gained experience that they would like to share.

Specifically, I have the following questions:

  1. Would a multi-animal project (for individual whiskers) be indicated or would a single-animal project work better? I know that multi-animal projects actually need to label each individual (in this case, the whiskers), but I thought that even with fewer labels it might be able to distinguish the whiskers better across the individual frames. I wanted to label four whiskers as examples and it would be good if at least the sides (left/right) are assigned correctly.

  2. Is it super important that I catch the same whiskers for labeling in each frame or does the network prevent jumps of the markers anyway? Therefore I thought that a maDLC project might be superior to a single animal one. Should I maybe reduce myself to only one whisker per side or are more whiskers even better for differentiation?

  3. How many labels should I train per whisker? I have heard that the more the better. Are 5 labels enough?

  4. Should I use a skeleton for the whiskers and are redundant skeletons from point to point better for efficient training sets?

  5. How many frames would be recommended to reliably train the network on whiskers?

  6. In experience, are such whisker-training sets generalisable well across different mice or do I need images of each individual?

I hope I haven’t overwhelmed you with the questions and am grateful for any tips and suggestions! Have a great new year!