Difference to BirdNET Lite

[gitalm]

Dear Stefan,

thanks for the wonderful new software. Makes it easy with analyzing in locale and segements. But the results differ quite from the old BirdNET-Lite. I used a fork that can manage to analze a folder https://github.com/UCSD-E4E/BirdNET-Lite and the results have been quite better, ring ouzel was right and also pygmy owl. The parameters have been just the same. python3 analyze.py --i /home/jalmer/Musik/TestHofbauernalm/ --lat 47.856 --lon 12.344 --week 16 --overlap 1.5 --min_conf 0.7 --sensitivity 1.25 --locale de --o /home/jalmer/Musik/TestHofbauernalm/

[DD4WH]

Yes: the problem is, that the models are not comparable, because they do not cover the same species ensemble. Pygmy owl for example is not covered at all by the new 2.1 model. It would be highly desirable to have access to a new model which covers ALL (Central) European species.

[gitalm]

Thanks for the reply. Because the file handling with BirdNET-Analyzer is much nicer. It would be nice, if it contains the time stamp in the output. Thanks for your help.

[kahst]

I am working on the 3k species model which will finally include pygmy owls :) The new model should provide better results than the Lite model for most species because of a different model architecture, improved training process and vastly more data to train on (esp. soundscape data). However, you might have to figure out the best parameter settings again in order to achieve the best possible results. Not sure why it didn't pick up the Ring Ouzel, I would suspect the location filter is at fault here because it is a bit more restrictive than in the Lite version. But it could also be that this is one of the edge cases in which the new version does not perform as well as the old one.

What kind of time stamp would you like to see in the output?

[JacobGlennAyers]

Stefan kindly gave me an overview of the difference between the models in https://github.com/kahst/BirdNET-Analyzer/issues/20

[gitalm]

Dear Stefan, thanks for the support. Than i wait until the complete version, but your work is incredible. Especially for owls .🦉 it is very comfortable to sleep and let some device do the recording ;-)

As time stamp the creation date of the file + the time from the recording would be nice. So you can directly draw diagrams out of the csv-file. The other project is a solar powered Recorder with instant detection and upload via LTE

[kahst]

Ok, I understand, but shouldn't the filename of the recording already have a time stamp? We're just appending .BirdNET.selection.table.txt to the filename.

[Josef-Haupt]

No activity -> closing this issue