search

kahst / BirdNET-Lite

TFLite version of BirdNET. Bird sound recognition for more than 6,000 species worldwide.
Other
147 stars 195 forks source link

Clarify relationship between BirdNET and BirdNET-Lite #14

Open haselwarter opened 2 years ago

haselwarter commented 2 years ago

Hi,

Please add some information to the readme about the relationship between the two projects. Are they completely different? Is the output format similar? Can one expect similar, or better results with the new(?) -Lite version? Is Lite the future, and everybody should switch to it?

DD4WH commented 2 years ago

BirdNET is based on tensorflow, BirdNET-Lite is based on tensorflowlite / TFLite The available classifier for BirdNET-Lite has a much wider coverage of bird species, about 65% of the global species. BirdNET-Lite also runs on less power hungry computers, have a look for example at this: https://github.com/mcguirepr89/BirdNET-Pi

lenawarbler commented 2 years ago

I think BirdNET-Lite is the way to go. Here is my experience with both:

-BirdNET-Lite processed 5 minute recordings in under 10 seconds (vs. 3 minutes with BirdNET) -As far as I can tell, you cannot disentangle the original BirdNET from eBird - unless you are a programming wiz. So if your study area does not have eBird information (like mine in northern Canada), the results are essentially useless because BirdNET cannot extract an appropriate species list. -BirdNET-Lite allows the input of a custom species list so you can get around this issue easily - as long as you have prior knowledge of the species in your region/season. -Output format: BirdNET reports 3 ranked species per detection, while BirdNET-Lite reports just one. BirdNET produces a txt file for every recording while BirdNET-Lite produces csv's. Otherwise the results are the same.