Tracking multiple mice using deeplabcut

Hi all,

I read that in the paper that the network was trained to keep track of body parts on one mouse but it was possible to label videos with multiple mice during testing.

I am a bit curious about this. If during training, there is only one set of labels (e.g. left ear, right ear, nose, tail), how does the labels get duplicated when testing on a video with more than one mouse?

Many thanks!

1 Like

Hi @edgarbc in the paper we show you can threshold the scoremaps to locate multiple instances of a part, i.e. “nose”, “tail”, etc.

In short, we will soon be releasing an update that allows this to be more straight-forward for the end user. In the meantime, you would need to edit the code for analyze videos to grab > 1 maximum point off the scoremap.

Here are a few issues opened around this topic too: https://github.com/AlexEMG/DeepLabCut/issues/50
https://github.com/AlexEMG/DeepLabCut/issues/228

1 Like

Great. Thanks! I will have a look.

Hi,
Is this feature included in the 2.06 release?

nope :slight_smile: but we will announce it when it;s out :wink:

Any update on a soft timeline for this? We are about to enter a new phase of experiments in our lab and are excited about the potential of DeepLab Cut, but typically need to track multiple animals.

We are most interested in identifying instances of mating, so perhaps we could train to ID just one BIG blob instead of two smaller posed blobs haha

Hi! Can you update what line do we need to change in the current version of predict.py?

By the way, is there any deeplabcut-development branch for this?

Here is how it works: https://github.com/AlexEMG/DeepLabCut/pull/321

Hi @MWMathis,

I tried to recognize several identical objects in one image at the same time. Unfortunately, I only recognized several very close positions at one point of an object. The detection then jumped over to the other object, but it was never visible at the same time.

I followed the announcement that a multi object detection should appear in the next days. Will this new release detect several similar objects at the same time? Thanks in advance!

Yes, that’s what the next update deliver!

Hi @MWMathis,

do you have a date for release?

Thank you!

Finally, here is the beta version:

pip install deeplabcut==2.2b5

1 Like