python script for batdetect2 via command line interface

[d-singer]

Hi,

with the help of ChatGPT I was able to develop a small python script for using batdetect2 via the command line interface, adding some functionality that was helpful for our large-scale monitoring:

https://github.com/d-singer/batdetect2_CLI

Maybe you like to test it and integrate it as an option in the batdetect2 repository in future.

Best,
David

[mbsantiago]

Hi @d-singer!

Huge thanks for this, the script is working really well. I did have to install tqdm separately, so might be good to add it to the instructions.

The batch processing and recovering from partial progress are great ideas, and I'll definitely incorporate them. Just to clarify: we do have a CLI for BD2 already (see batdetect2/cli.py), which covers some of this functionality, but it doesn't yet support this "checkpointing".

In the upcoming update, we're also expanding the CLI to include commands for training and data management, so it would be great to integrate your approach. Are there any options or patterns from the BirdNET CLI that you've found especially useful and think would be worth implementing here?

Cheers,
Santiago

[d-singer]

Hi @mbsantiago

thank you for your response - I did not notice the existing CLI, otherwise I would have used it as a baseline...

What I like about the BirdNET CLI is mainly the option to skip existing results as an large scale analysis on a work station may stop due to different reasons. Saves time and energy to not re-run it completely after a stop.

What they have not implemented within BirdNET CLI is to run the analysis in batches per subfolder (=study site in our audio archive). I think this is very helpful when analysing large data as you can be sure which site has been completely analysed - at least I like to organise the audio data in one folder per site.

BirdNET also does not include a row into the results when no calls are detected. However, by including such "true negatives" (depending on the capabilities of the classifier...) in the results, you finally get a complete list of audio files that have been sucessfully analysed - what is particularly helpful when analysing temporal patterns where "true zeros" are also highly informative. Would be nice to have the option to include also files with "true negatives" in the results.

A list of files that could not be successfully analysed due to corrupt audio files etc. would also be great to report temporal coverage/data gaps.

Seeing the progress per batch and sites is also a helpful feature for me.

Feel free to integrate (parts of) my script :)

Best,
David

Edit: An CLI implementation of the option to get scores for all species per call would also be a nice feature, as my colleague @batthiloreng already communicated: #53