Replies: 0 comments 14 replies
-
Frank, you're awesome! Thank you, first of all, for dedicating so much of yourself to conservation efforts and for educating others in the subject -- we need more folks like you! 1. Installation/network Two possible solutions to this which I will explore at some point:
Thank you for the feedback on the AnyDesk bit-version issue -- I will change my documentation to recommending the NoMachine setup you've had success with and will keep an eye out for other solutions.
2. Settings for more flexibility
3. More features
Thank you so much for your feedback and providing some insight from the biologists' perspective ~~ very cool. My best, |
Beta Was this translation helpful? Give feedback.
-
Patrick, thank you so much for your detailed answer and the suggestions. I have to find some time to try some of your suggestions and I will report my success here. Just a quick answer on your question regarding this:
RPi code is here: and in action it pretty much looks like this: https://birdnet.cornell.edu/live/ I have seen it running live at a conference where Stefan Kahl presented BirdNET with a Raspberry Pi runnning that code, absolutely amazing! Have a nice weekend! Frank DD4WH |
Beta Was this translation helpful? Give feedback.
-
@DD4WH I have worked a bit with your suggestions and have the rpitesting branch working with a Caddyfile that allows the Raspberry Pi to connect to my cell phone's hotspot, then I can reach the BirdNET-system at http://raspberrrypi.local -- I was hopeful that the setup would work for you out of the box, but the installation that you used sets the first Caddyfile block to only listen to http://localhost and then the rest of the http services are broadcast via mDNS. While the mDNS features aren't available using this hotspot connection, the main Extractions page is available with this method. I will be pulling the rpitesting changes into the main BirdNET-system-for-raspi4 branch in the next few days after more testing. Lots of great success today, but I seem to always overlook something, so I'm waiting a bit longer before pushing these changes. Once I have pushed the changes, I hope you'll consider reinstalling to add this new feature. |
Beta Was this translation helpful? Give feedback.
-
sounds good! As soon as you have a version ready on github and want me to test, just drop me a note and I will follow the installation procedure on another SD card and test the new functionalities for you! |
Beta Was this translation helpful? Give feedback.
-
Patrick, thank you so much ! I will be testing this out as soon as I find some more time this weekend and report back to you here! Thanks again for all your hard work! Frank DD4WH |
Beta Was this translation helpful? Give feedback.
-
Hi Patrick, today I installed the new installer, here come my results and comments: 1.) installed the recommended RPi OS with balenaEtcher to a 16GB microSD card So, in summary, it works!!!
A little bit complicated, but the basic functionality is there! Thanks, Patrick for making this available! My impression is that the major obstacle is the Android smartphone not connecting to mDNS services . . but I do not know anything about these networking questions, so this is -at most- sophisticated guessing . . . Keep up the good work! Best regards, Frank |
Beta Was this translation helpful? Give feedback.
-
@mcguirepr89
Thanks for being so active with this BirdNET implementation!
My profession is (among other things) teaching ornithological methods for nature conservationists, so this list of ideas/suggestions is surely biased and emphasizes the use of your system, not necessarily the technological perspective :-).
I collected some ideas for improving performance, flexibility and promoting professional use of your BirdNET-system and grouped them into three categories:
1) Installation/network
• The use of the installation script is super easy and makes it possible for an unexperienced RPi-owner to install this system
• In my home network, after installation, the system does not work without a LAN-connection to the RPi. The Pi is connected to my home WLAN, but when I pull the LAN-cable, it is no longer possible to connect to the RPi from another computer in the network [however, I can connect to it via a remote desktop application like NoMachine]
• It would be good to fully integrate the possibility of remote desktop access into the installation process: I initially tried Anydesk, but that does not work, because Anydesk does not (yet) offer an installer for a 64-bit RPi-system (and using the 32-bit-RPi-installer installs something, but it is not possible to start Anydesk --> many sources on the web report this without any solution to it). I helped myself with the alternative NoMachine, which does have an installer (use the one for ARMv8 architecture) and allows remote access even from my smartphone and with functional audio streaming!
2) Settings for more flexibility
• BirdNET-system uses the eBird distribution data to filter the species list for the internal BirdNET classification process. There should be an option to switch this ON/OFF during runtime or at least in the config.py-file. I plan to use the system with mobile phone connectivity making it impractical to regularly download such an amount of data. Additionally, for conservation work, it is important to have classifications for rarer species even if they have not been included in the eBird data. For example, I am monitoring pygmy owl (which is quite frequent here), but there are no records on eBird for this region, so it is never recognized by the system as it is at the moment. I helped myself by setting the eBird threshold to 0.0, but then ALL species in the world find entry into my classification results. And still, the eBird data is being downloaded!
• A must for me is the possibility to use custom species lists, so that you can use the/a global BirdNET model, but filter this for my region. By the way, I have a list for European species (about 700 species) derived from the species list of the BirdNET-Lite classifier species list with 5000+ species. If anybody is interested, I can post it here or elsewhere
• There should be the possibility to make settings in real time on the web interface: Settings should include: eBird threshold, spectrogram overlap, spectrogram minimum length, minimum confidence, prediction sensitivity
3) More features
• The points mentioned above have higher priorities from my perspective, but here come some ideas, which could make the system even more attractive, but will probably be a little more complex to implement:
• A simple bargraph with the number of extracted recordings per species (as a first overview graph) would be very nice: this could help immensely in separating wrong classifications from securely identified species. This could also be implemented very fast on the birdstats.local page as a simple figure after the species
• Stefan Kahl has already programmed a beautiful streaming interface which shows a running realtime spectrogram of the incoming audio, simultaneously broadcasts the audio stream and shows the subsequent classification results underneath as well as a list of the classifications in the last hour/day. This is already available as RPi-code in a github repo (although I have never managed to get this installed). That would be very nice to have as an add-on feature of BirdNET-system which can be switched off, if CPU resources or RAM do not allow
• The notification system for mobile phone does not work for me, simply because the App „pushed“ is not functional on Android systems. There are many complaints in the app ratings stating that the app does crash on startup, and this is also the case with my android phone. So this does not seem to be a solution for Android users in general
• Phone notification: would also be nice to have an option to be notified when a specific species has been heard: would make sense in my example of pygmy owl monitoring!
• In the longterm and for more professional use cases it would be good to be able to swap classification models. One could then use different models for different use cases: NFC (night flight calls), owls, global models, regionally tuned models. However, of course this would need specifically trained models (either trained by the researcher himself or by Stefan Kahl :-))
Please understand these points as an input to make your wonderful system even more wonderful!
Regards, Frank DD4WH
Beta Was this translation helpful? Give feedback.
All reactions