Idmatcher.ai

[Deeksha G Krishnamoorthy]

Hi,

I've been using idtracker for a while now for my research, and it has been really helpful, thanks :)

I recently tried using idmatcher, it works perfectly well when the number of animals between the sessions are same. 

As far as I understand from the webiste (https://idtracker.ai/latest/user_guide/idmatcherai.html) , if I have a mismatch in the number of animals between the master session and the matching session, the matching should still work, but we'd have to infer the match based on the direct and indirect assignment scores.

However, this didn't really work when I tried tracking the session with lesser animals using knowledge transfer from the master session (with more animals). But I gather that we need to do the knowledge transfer for matching the sessions later?

Am I missing something?

Thanks in advance,

Deeksha

[idtrackerai]

Hi Deeksha, thanks for reaching out.

In idmatcher.ai you usually use the model from MASTER to identify animals in MATCHING and the model from MATCHING to identify animals in MASTER. When the number of animals in two sessions are different, idmatcherai relies solely on the model from the session with the largest number of animals to identify the animal images from the other session with fewer animals. The reversed matching cannot be done. Idmatcher.ai deals with it automatically so you should not care about it.

However, knowledge transfer (reusing the weights from a previous model as a starting point for another tracking session) works differently. Right now you cannot use knowledge transfer when the number of animals is different. What you can do is to track the sessions independently (without knowledge transfer) and after the tracking, match them with idmatcher.ai. In this case you might want to ensure you are using the same id_image_size and resolution_reduction in both sessions.

I've been checking this part of the code and I realised that, in the latest version 6 of the software, there's no reason to forbid knowledge transfer when the number of animals is different. I fixed it and in the next release (in a few days) it will be possible to do knowledge transfer regardless of the number of animals (in case you work with the new default contrastive algorithm).

I hope this helps,

Jordi