Is there an existing issue for this?

• I have searched the existing issues

Operating System

Win10

DeepLabCut version

3.0.0rc8

What engine are you using?

pytorch

DeepLabCut mode

multi animal

Device type

GeForce 2060 RTX

Bug description 🐛

Hi all,

I followed you fantastic colab for the BUCTD and CTD tracking. I am trying to detect and track 5 animals, with 5 keypoints. However, I am dealing with some problems:

• Some animals are not detected for a few frames and, in these moments, there are several false positives (i.e. DLC tracks a fixed region of the image, for example).
• When this problem happens, after a while, a given animal becomes visible again but the DLC is not able to detect it because it is fixed in that specific region of the image.

Therefore, I would like to ask some points to increase the performance of my project:

1. Should I increase the number of labeled frames (I have 80 frames labelled), or the number of bodyparts?
2. When only some parts of the animal are available, should I label these parts or only when the animal has its full body available?
3. How to increase the CTD performance when the animal is occluded for a while? I explored those CTD parameters but it seems that there is no difference. Could you explain how I change the CTD parameters to increase the performance, in this case?
4. Is it possible to export the bounding box coordinates, after analyzing the video?

Thank you aigain,

Felipe.

Steps To Reproduce

No response

Relevant log output

Anything else?

No response

Code of Conduct

• I agree to follow this project's Code of Conduct