Thank you everyone for participating and all your fantastic contributions, congratulations to the winners. We are completely blown away by the level of participation and activity through code, forum posts and submissions. We would have never anticipated to see such high scores (we were expecting scores in the 70s) and we are grateful for all the hard work you put into these solutions.
\nWe also honestly never anticipated that BirdNET would perform so well (it showed mediocre results in our real-world experiments on Hawaii despite the fact that it uses vastly more training data), in retrospect, we probably should have investigated the impact on the competition earlier. However, it is great to see that non-BirdNET submissions are on-par and that is really encouraging. Just imagine what your systems could do when using more training data! This will be interesting for us to see in the upcoming weeks.
\nSpeaking of which: Please make sure to leave a write-up, or - even better - submit a working note so that we can understand what your attempt was, can learn about the things that worked and didn't work and can replicate your results while accurately acknowledging your work. As a reminder, there will be two best working note awards, and those prizes will be awarded independent of the final score - all that counts is the quality of your working note. Take a look at this thread to learn more.
\nAgain, thanks everyone for your hard work, and we hope to see you again in 2023 for a new edition of BirdCLEF.
\nBest,
\nStefan
Thank you everyone for participating and all your fantastic contributions, congratulations to the winners. We are completely blown away by the level of participation and activity through code, forum posts and submissions. We would have never anticipated to see such high scores (we were expecting scores in the 70s) and we are grateful for all the hard work you put into these solutions.
\nWe also honestly never anticipated that BirdNET would perform so well (it showed mediocre results in our real-world experiments on Hawaii despite the fact that it uses vastly more training data), in retrospect, we probably should have investigated the impact on the competition earlier. However, it is great to see that non-BirdNET submissions are on-par and that is really encouraging. Just imagine what your systems could do when using more training data! This will be interesting for us to see in the upcoming weeks.
\nSpeaking of which: Please make sure to leave a write-up, or - even better - submit a working note so that we can understand what your attempt was, can learn about the things that worked and didn't work and can replicate your results while accurately acknowledging your work. As a reminder, there will be two best working note awards, and those prizes will be awarded independent of the final score - all that counts is the quality of your working note. Take a look at this thread to learn more.
\nAgain, thanks everyone for your hard work, and we hope to see you again in 2023 for a new edition of BirdCLEF.
\nBest,
\nStefan
@stefankahl I don't understand why you weakened the license of your BirdNET model one day before the competition closed without even posting it here on Kaggle. Also, keeping the license strict until the end would have resulted in a much more fair competition.
", "rawMarkdown": "@stefankahl I don't understand why you weakened the license of your BirdNET model one day before the competition closed without even posting it here on Kaggle. Also, keeping the license strict until the end would have resulted in a much more fair competition.", "votes": 5, "replies": [ { "id": 1801085, "postDate": "2022-05-25T12:21:38.180Z", "content": "Yeah, I get the frustration, I wish we would have known earlier that people were intending to use BirdNET. We had a long discussion about the license after a group contacted us (just one day before the deadline) and decided that the training dataset uses the same license as BirdNET and that we therefore cannot enforce the \"non-commercial\" clause, although teams will receive prize money…
", "rawMarkdown": "Yeah, I get the frustration, I wish we would have known earlier that people were intending to use BirdNET. We had a long discussion about the license after a group contacted us (just one day before the deadline) and decided that the training dataset uses the same license as BirdNET and that we therefore cannot enforce the \"non-commercial\" clause, although teams will receive prize money..." } ] }, { "id": 1800949, "postDate": "2022-05-25T10:10:18.473Z", "content": "Was any of the test data of this competition used to train BirdNET?
", "rawMarkdown": "Was any of the test data of this competition used to train BirdNET?", "votes": 2, "replies": [ { "id": 1801033, "postDate": "2022-05-25T11:19:34.063Z", "content": "Or a follow-up question: how exactly was the process to annotate (parts of) the test data structured? Was BirdNET used in any way to annotate the data?
", "rawMarkdown": "Or a follow-up question: how exactly was the process to annotate (parts of) the test data structured? Was BirdNET used in any way to annotate the data?", "votes": 3 }, { "id": 1801086, "postDate": "2022-05-25T12:22:28.087Z", "content": "No, but BirdNET uses other/additional data sources and thus works better for some of the target species (not all though, judging from the submissions).
", "rawMarkdown": "No, but BirdNET uses other/additional data sources and thus works better for some of the target species (not all though, judging from the submissions)." } ] }, { "id": 1801659, "postDate": "2022-05-26T02:38:01.077Z", "content": "Just wanna say thank you for being an awesome host. Birds are my favorite competition, and I will try again next year.
", "rawMarkdown": "Just wanna say thank you for being an awesome host. Birds are my favorite competition, and I will try again next year.", "votes": 2, "replies": [ { "id": 1802061, "postDate": "2022-05-26T12:10:36.877Z", "content": "Thanks :) We'll be back in 2023 with a new edition.
", "rawMarkdown": "Thanks :) We'll be back in 2023 with a new edition." } ] }, { "id": 1801525, "postDate": "2022-05-25T20:28:59.517Z", "rawMarkdown": "", "isDeleted": true } ], "comments": [ { "id": 1801080, "author_name": "Markus Vogelbacher", "author_url": "", "post_date": "2022-05-25T12:16:58.300000", "content": "@stefankahl I don't understand why you weakened the license of your BirdNET model one day before the competition closed without even posting it here on Kaggle. Also, keeping the license strict until the end would have resulted in a much more fair competition.
", "votes": 5, "replies": [ { "id": 1801085, "author_name": "Stefan Kahl", "author_url": "", "post_date": "2022-05-25T12:21:38.180000", "content": "Yeah, I get the frustration, I wish we would have known earlier that people were intending to use BirdNET. We had a long discussion about the license after a group contacted us (just one day before the deadline) and decided that the training dataset uses the same license as BirdNET and that we therefore cannot enforce the \"non-commercial\" clause, although teams will receive prize money…
", "votes": 0, "replies": [] } ] }, { "id": 1800949, "author_name": "Gilles Vandewiele", "author_url": "", "post_date": "2022-05-25T10:10:18.473000", "content": "Was any of the test data of this competition used to train BirdNET?
", "votes": 2, "replies": [ { "id": 1801033, "author_name": "Gilles Vandewiele", "author_url": "", "post_date": "2022-05-25T11:19:34.063000", "content": "Or a follow-up question: how exactly was the process to annotate (parts of) the test data structured? Was BirdNET used in any way to annotate the data?
", "votes": 3, "replies": [] }, { "id": 1801086, "author_name": "Stefan Kahl", "author_url": "", "post_date": "2022-05-25T12:22:28.087000", "content": "No, but BirdNET uses other/additional data sources and thus works better for some of the target species (not all though, judging from the submissions).
", "votes": 0, "replies": [] } ] }, { "id": 1801659, "author_name": "yukiya", "author_url": "", "post_date": "2022-05-26T02:38:01.077000", "content": "Just wanna say thank you for being an awesome host. Birds are my favorite competition, and I will try again next year.
", "votes": 2, "replies": [ { "id": 1802061, "author_name": "Stefan Kahl", "author_url": "", "post_date": "2022-05-26T12:10:36.877000", "content": "Thanks :) We'll be back in 2023 with a new edition.
", "votes": 0, "replies": [] } ] }, { "id": 1801525, "author_name": "", "author_url": "", "post_date": "2022-05-25T20:28:59.517000", "content": "", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "1800751": "Thank you everyone for participating and all your fantastic contributions, congratulations to the winners. We are completely blown away by the level of participation and activity through code, forum posts and submissions. We would have never anticipated to see such high scores (we were expecting scores in the 70s) and we are grateful for all the hard work you put into these solutions.\n\nWe also honestly never anticipated that [BirdNET](https://github.com/kahst/BirdNET-Analyzer) would perform so well (it showed mediocre results in our real-world experiments on Hawaii despite the fact that it uses vastly more training data), in retrospect, we probably should have investigated the impact on the competition earlier. However, it is great to see that non-BirdNET submissions are on-par and that is really encouraging. Just imagine what your systems could do when using more training data! This will be interesting for us to see in the upcoming weeks.\n\nSpeaking of which: Please make sure to leave a write-up, or - even better - submit a working note so that we can understand what your attempt was, can learn about the things that worked and didn't work and can replicate your results while accurately acknowledging your work. As a reminder, there will be two best working note awards, and those prizes will be awarded independent of the final score - all that counts is the quality of your working note. Take a look at [this thread](https://www.kaggle.com/competitions/birdclef-2022/discussion/307744) to learn more.\n\nAgain, thanks everyone for your hard work, and we hope to see you again in 2023 for a new edition of BirdCLEF.\n\nBest,\nStefan", "1801080": "@stefankahl I don't understand why you weakened the license of your BirdNET model one day before the competition closed without even posting it here on Kaggle. Also, keeping the license strict until the end would have resulted in a much more fair competition.", "1800949": "Was any of the test data of this competition used to train BirdNET?", "1801659": "Just wanna say thank you for being an awesome host. Birds are my favorite competition, and I will try again next year.", "1801525": "" } }