Recommended settings ?

[jpmurray]

I'm curently testing Birdnet-Go because I really like the UI, but I'm curently getting way less results than I was getting with Birdnet-Pi... Is there some recommended settings I should be setting that I might have overlooked ?

[tphakala]

What is BirdNET overlap value set to on your system? BirdNET-Go has false positive mitigation system which may reduce total number of detections, it is documented in https://github.com/tphakala/birdnet-go/blob/main/doc/wiki/guide.md#deep-detection.

In essence if you set BirdNET Overlap to 0.0 then BirdNET-Go reports every detection on first hit, you may see higher number of detections but also much higher number of false positive matches.

[tao54tao]

I just set up my own instance of BirdNet-Go using an RTSP cam as well. I am interested if you have found some settings that work well for you as I am also getting only a handful of detections.

[jpmurray]

I got the config under here. I'm getting more hit for same spicies than I had before, but it is still way less than I get with Birdnet-Pi at the moment.

(For this day, 20 detection with Birdnet-go against 109 with Birdnet-Pi.)

What kills me is that one detects a couple spicies that the other won't, and vice versa. Like BN-Go is catching a lot of crows but no mourning dove, and BN-Pi got a crow once and a lot of doves...

birdnet:
    sensitivity: 1.2
    threshold: 0.3
    overlap: 2.8
    rangefilter:
        debug: false
        model: latest
        threshold: 0.6
dynamicthreshold:
        trigger: 0.9
        min: 0.2
        validhours: 48

[tphakala]

Do not set range filter threshold to 0.6, it reduces allowed species significantly, set it to default 0.01.

[tphakala]

I added documentation about detection flow and added explanation of range filter, please see https://github.com/tphakala/birdnet-go/wiki/BirdNET%E2%80%90Go-Guide#birdnet-detection-pipeline.

@tao54tao default settings should be good for most users, avoid tweaking BirdNET setting values too much to avoid problems with detections. If you see too little detections lower threshold and overlap value, if you see too many false positives first increase overlap to 2.0 - 2.8 range and if it does not help then increase BirdNET threshold.

[jpmurray]

Awesome stuff !