Hi Sean!

Wanted to just mention that Arbimon, Rainforest Connection's ecoacoustic platform, has a number of projects in Peru (here, here, & others, if you search by keyword 'peru'). We have some existing CNNs for that region (mostly from Ecuador & Brazil, but there is likely species overlap). Do feel free to DM me here or email me (carly@rfcx.org) and I'm happy to talk about collaborating! 

-Carly   

Also tagging @NickGardner who works on a similar project! (detecting birds from audio in Peruvian flooded forests)

Interesting! 

Hi Sean, sounds like an excellent project. Definitely talk with the Arbimon folks! As @carlybatist said, I am working with birds in the Peruvian Amazon, but in Loreto. Definitely would like to hear more about your project. As for this labelling issue here, definitely looks like an error. I have not used this dataset, now I'm curious.  To be honest, some questionable labelling in that file in general. Bounding boxes can be very subjective... 

Hi Sean!

I was the one who annotated these recordings and this is definitely a mistake- that song is a Hauxwell's Thrush; it's annotated correctly with 15622 but not with 15616 (that's the same species as 15620). I'm currently in the field and don't have the ability to fix it right now, but should be able to later this month. If you have more of these, please send them to me (wah63@cornell.edu) and I'll correct any errors and upload a version 2. Same for you @NickGardner - I'm very happy to chat about any labeling decisions and correct mistakes or inconsistencies where applicable. Some were intentional/tossup decisions because of the subjective nature of manual bounding boxes, as you note, and some of them are obvious mistakes like the example Sean found, which can happen for a ton of reasons, despite my efforts to reduce or eliminate them. For context, I did the data collection, annotation, etc of this project essentially by myself when I was an undergrad, and so I guarantee there are some other similar errors in there, and, well, there is a reason that there haven't been many similar datasets made in this system! Especially at the time I did these (2019), there just wasn't an established standard for many of these decisions, and I just did the best I could. My main goal was just getting *something* from these systems out there, fully transparent and free, so that people like you could have more to work with. Ironically, these sort of human errors are exactly why the tools you're working on are so important, even though the fear of committing them probably contributes to the scarcity of similar datasets. Hopefully we will eventually be able to eliminate this kind of thing with better automated processes, but in the meantime, yes, please send me everything you found and we can correct them.

Thanks so much for using my dataset and looking at it closely enough to find mistakes - together we can make it better and more useful for future users :)

Best,

Alec