Tidying up misidentifications

[srchild]

I intend making my installation public, but it will look silly if I leave significant misidentifications uncorrected (e.g. birds which are not known in this country)

To delete a misidentified clip/bird, is it sufficient to delete the relevant row(s) from the database. Is there anything else to do, such as a reindex after manipulating the database?

If I leave the audio files in place will they eventually get deleted automatically, or should I delete the audio file as well as the database row to avoid wasting SD card space on redundant audio files?

Thanks

[mcguirepr89]

To delete a misidentified clip/bird, is it sufficient to delete the relevant row(s) from the database. Is there anything else to do, such as a reindex after manipulating the database?

Deleting from the database will be sufficient to remove the data from the "Species Stats" page, and from the "Today's Detections," "Best Recordings," and "Recordings" pages.

If I leave the audio files in place will they eventually get deleted automatically, or should I delete the audio file as well as the database row to avoid wasting SD card space on redundant audio files?

To actually remove the audio files, I made a little bash alias that can be invoked as remove and then some really unique identifier. For instance, to remove all of the "Common Nighthawk" recordings, but keep the "Red-shouldered Hawk" recordings, I would issue "remove "Common Nighthawk"" in the terminal and that should work. If you have a species with a really unique name, you can use the unique part. For instance, removing the "Swainson's Warbler" could be accomplished with just "remove swainson" as it is unique enough to only catch the species I need to remove.

Use that tool carefully, as it does not ask before removing files.

Another method to remove the undesirable audio clips could be:

find ~/BirdSongs -type d -name '*Common Nighthawk*' -exec rm -drfv {} +

Lastly, to prevent those species from showing up in your database again in the future, you can prevent BirdNET-Pi from reporting predictions for those species by adding them to your "Excluded Species List".

To do this, go to "Tools" > "Excluded Species List"

On that page, press "Ctrl + F" to enter your browser's search mode. Type the name of the species you wish to exclude from your reporting and data collection, select from within the list of species in the box on the left and click ">>ADD>>" to have the species added to your Excluded Species List.

Please let me know if you have any other questions and I'll do my best to help out.

My very best regards,
Patrick

[srchild]

Thanks for that prompt help

[CaiusX]

How about an option to move misidentified sounds to a misidentified directory? Could be useful for future model training.

[Klogdor]

I have had a few misidentifications (it tends to identify my drill as some bird that is extinct here) .. would it not make sense to put a button on the specific recording that lets you mark something as a false positive that will then also tidy up and remove the clip + the database entry ? (or just add a false positive = 1 to the database entry?)

[srchild]

I have had just the same thought but hesitated to make yet another change request!

Yes, an option / checkbox / something that can be selected/pressed when viewing a recording, which will delete both the recording and the corresponding database entry in a single action, would be very helpful. I spent some time a few days ago checking suspect recordings, finding them in the db and deleting them, and then finding the actual recordings and deleting them.

It would be useful not only for misidentifications but also for recordings that need deleting for other reasons such as human speech being recorded.

[jds3d]

Are there instructions somewhere for connecting to the database/deleting the rows? I have an eastern screech owl record that I played from Merlin to test the original setup. I can write the sql if I have the connections details and info on what tables need to be cleared.

[slowrunner]

Superb. This works to delete mis-identifications.

There is a panel that seems for the purpose of manually identifying/correcting mis-id'd birds under species stats, that appears to offer a way to change misidentifications.

Under Species Stats:

1. select a species
2. listen to recording
3. click reviewed
4. Select "False Positive"
5. Edit/selection box with "UNKNOWN" appears, type Norther M, select Northern Mockingbird

But then what? It lacks a "Save New ID" button.

[slowrunner]

Using the Database Maintenance concept, I was able to edit the com_name and sci_name of the misidentified records, and see the effect when I checked species stats again. I can do the "tidying up misidentifications" this way also.

[CaiusX]

Hi y’all

will push this facility thru sooner than later. It’s working on joburg.birdnetpi.com under species stats page, African electricity and GitHub ignorance are retarding the workflow at the moment.

[FirstTenor]

Superb. This works to delete mis-identifications.

There is a panel that seems for the purpose of manually identifying/correcting mis-id'd birds under species stats, that appears to offer a way to change misidentifications.

Under Species Stats:

1. select a species
2. listen to recording
3. click reviewed
4. Select "False Positive"
5. Edit/selection box with "UNKNOWN" appears, type Norther M, select Northern Mockingbird

But then what? It lacks a "Save New ID" button.

I saw this too …. I have successfully used the functions to ‘delete re incorrect recording’ - but would really prefer to someone flag / report these misidentifications to improve the model. This function that @slowrunner quotes seems partially implemented… as mentioned, you can edit ‘Unknown’ to provide the real answer - but not save it.

I also wonder if that would just be ‘my local instance’ or actually reported back somewhere centrally?

Current examples (during daylight) are:

Tawny Owl = Police Car
Small Eurasian owl = Rook & Robin calling concurrently
Small toed treecreeper = someone walking dog in woods over mains hum

[danny7989]

I, too, was wondering about this review feature. I have false positives which I would like to correct/review, and this feature would be ideal if it worked. If the review influenced the AI, it would be even better. Reviewing or deleting false positives from the DB directly is not currently an option for me.

[tomwhite101]

I would be great to be able to delete a record directly from the species list.

[mcguirepr89]

I agree -- this will be possible eventually, hopefully sooner than later

[GeekNJ]

Just getting started and there was a major misidentification. We could delete the record or maybe just a field to identify as incorrect in the DB if there's ever a reason to use that data in the future for the project. Just a thought.

[hoover67]

I've deleted many misidentifications using the adminer tool, but it's a bit cumbersome. A link to delete the spectrogram, sound file and db entry for an MI would be great and I'd be happy to help implementing it (I'm a django / python guy though most of the time when it comes to web stuff).

All the best,

Uwe

[hoover67]

http://birdnetpi.local/scripts/adminer.php?sqlite=&username=&db=birds.db&select=detections&columns%5B0%5D%5Bfun%5D=&columns%5B0%5D%5Bcol%5D=&where%5B0%5D%5Bcol%5D=&where%5B0%5D%5Bop%5D=LIKE+%25%25&where%5B0%5D%5Bval%5D=sandp&where%5B01%5D%5Bcol%5D=&where%5B01%5D%5Bop%5D=%3D&where%5B01%5D%5Bval%5D=&order%5B0%5D=&limit=50&text_length=100

Here's a quick URL for your bookmarks which will take you straight to the adminer interface and offering a LIKE searchfield. Not exactly what we're asking for but it saves a few clicks.

Cheers, Uwe

[ehpersonal38]

Hi, I just implemented something that will hopefully meet your needs. Go ahead and do an update and then you should be able to delete detections from the DB by either 1) clicking the copy link in the top right of a detection on Overview or Today's Detections, then clicking X in the top right of the page that opens or 2) from the Recordings page. Let me know how that goes for you.

Ethan

[mcguirepr89]

Thanks for this, Ethan!! Super helpful

[mcguirepr89]

Not sure how to mark that as the answer?

[Phaeton]

You should be able to uncheck the prior one. BUT, the answer cannot be a reply. Ethan @ehpersonal38 would need to resubmit as an answer.

[ehpersonal38]

d2d914f offers a way to natively delete detections now in the interface! Hopefully this serves as a stepping stone to build out things further. I know there was some work done on addressing misidentifications in the verification branch for potentially retraining the model, interested to see how that progresses.

[mcguirepr89]

It (the verification branch) needs some tweaks to the workflow -- this is a great addition to it for sure!!!

[Sp4rkR4t]

This is a great addition but is there a reason the delete detection button is only in the recodings section, it would be superb if it was also available in todays detections section as that is the section I use most often to browse the days activity (it's only a single click vs 3 for that day in the recordings section)

[ehpersonal38]

I just did it that way since I think it would start looking a little crammed, and would probably look very bad on mobile. You can currently also delete detections this way though:

2LBCYzpK4a_Trim.mp4

Does that work for you?

[GeekNJ]

An X usually represents closing vs deleting. I would suggest using a trash can to represent Delete.