Combine custom classifiers.

[WillFMitch]

Hi there, apologies if this has been answered elsewhere.

I have been training new classifiers for several bird species which aren't currently included in the standard birdnet model. I have been using the GUI.

I would like to carry out a batch analysis and search for all focal species within a large dataset of field recordings. Is there a way to combine several custom classifiers into a single tflite file or perhaps into a species list, to allow me to search for each species in one go, instead of searching with each custom classifier individually?

I know I can append a custom classifier to the standard birdNET model but I just want the classifiers I've made and not the standard model.

I've little experience with python but willing to have a crack if someone could provide some pointers

Thanks!

[WillFMitch]

Hi again, @kahst any thoughts on the above problem? thanks again

[Mattk70]

https://birdnet-team.github.io/BirdNET-Analyzer/best-practices/training.html

[WillFMitch]

Thanks for your reply,

I have read these instructions but I'm still not sure if this will be sufficient to solve my problem.

Many of the species I am working with co-occur and have very similar calls. In such cases, I need to include example calls of species 1 as negative class training data for species 2 and vice versa, to achieve good classifier performance. This means I need to provide 'background' class data that is specific to each separate class in the training data.

With the instructions provided I cannot see that this is possible. Can you offer any further insight?

Thanks again for your time

[tphakala]

You should not supply negative classes as background class, you do it by adding a class with - prefix, like below

Folder structure should be like

background <- here you put all common negative samples like wind, traffic noise, AC hum, etc. environmental noises
Pica pica_Eurasian magpie <- this is positive class for magpie
-Pica pica_Eurasian magpie <- here you put all samples which are NOT magpie samples, these could include other species and it will not count as global negative sample, only negative sample for magpie

[WillFMitch]

Thank you so much! I will give this a go now.
I assume this is only command line and won't work in the GUI?
Really appreciate your help

[tphakala]

This is not CLI specific, it works just as well with GUI version.

[MariusBSc]

Question can one run detection on other species as well ? As we have a unit placed at a wildlife reserve? Kind Regards Marius BuysFounder AM Risk And Training On 14 Aug 2025, at 02:30, WillFMitch ***@***.***> wrote: @max-mauermann @kahst Any suggestions for the above? Even if just to confirm it's not possible using the GUI I've used birdnet R to search through my custom classifiers in a loop but this is very inefficient Thanks in advance —Reply to this email directly, view it on GitHub, or unsubscribe.You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>