kahst
Using segments to remove false positives
Hello,
Thanks for all the great work you've put in to BirdNET. It is an incredibly useful tool.
I would like to be able to use the segments function to remove false positives from the results. Currently, the validation process makes you choose a balance between recall and accuracy. It would be nice to be able to manually check the detections I've decided to keep and use results that are 100% accurate.
To do this, it would be nice to have two features:
- A setting that allows you to pull all segments above a set confidence score
- Include the 'begin time', aka the offset from the start of the audio file, in the segment file names.
We're working on a review feature for the GUI, in the meantime, we recommend using Raven to do selection review based on selection table outputs.
Adding the begin time to each segment should be easily possible @Josef-Haupt no?
Perhaps my #1 stocking stuffer wish would be a UI that allowed one to group detections in to one viewable event based on a user specified setting of inter-detection "gap". E.g. if detections are withing X seconds of one another, treat it as one vocal event. My #1 would be to allow user-defined fields to be attached to the detections and then exported seperately along with the name of the original audio file that the detection came from. Sorry for asking :) I really appreciate all that you folks are doing already.
The current repo includes a review tab, that allows users to seperate positive and negative segments.
