Hello and welcome to the 2023 BirdCLEF competition!
\nBirdCLEF is a unique opportunity for participants to showcase their skills and knowledge while contributing to the conservation efforts of avian species worldwide. This year's competition is especially important for conservation efforts in Africa, where the impact of climate change, habitat loss, and poaching is threatening many bird species.
\nAs hosts of the competition, we will try to be as active and responsive as possible to assist you in your endeavor. And all without giving away any secrets about the test data, so don't bother asking :)
\nIn this thread, I will give all the hosts the opportunity to introduce themselves. I will be the first to do so.
\nI am a research associate within the K. Lisa Yang Center for Conservation Bioacoustics at the Cornell Lab of Ornithology and Chemnitz University of Technology, focusing on the development of advanced machine learning models for automatic detection and identification of bird species in large audio collections. Furthermore, I am the technology lead for the BirdNET project and have been organizing the BirdCLEF Challenge since 2018. Feel free to ask me anything related to Deep Learning for bioacoustics, I might be able to help you.
", "messages": [ { "id": 2171959, "postDate": "2023-03-07T08:06:28.847Z", "content": "Hello and welcome to the 2023 BirdCLEF competition!
\nBirdCLEF is a unique opportunity for participants to showcase their skills and knowledge while contributing to the conservation efforts of avian species worldwide. This year's competition is especially important for conservation efforts in Africa, where the impact of climate change, habitat loss, and poaching is threatening many bird species.
\nAs hosts of the competition, we will try to be as active and responsive as possible to assist you in your endeavor. And all without giving away any secrets about the test data, so don't bother asking :)
\nIn this thread, I will give all the hosts the opportunity to introduce themselves. I will be the first to do so.
\nI am a research associate within the K. Lisa Yang Center for Conservation Bioacoustics at the Cornell Lab of Ornithology and Chemnitz University of Technology, focusing on the development of advanced machine learning models for automatic detection and identification of bird species in large audio collections. Furthermore, I am the technology lead for the BirdNET project and have been organizing the BirdCLEF Challenge since 2018. Feel free to ask me anything related to Deep Learning for bioacoustics, I might be able to help you.
", "rawMarkdown": "Hello and welcome to the 2023 BirdCLEF competition!\n\nBirdCLEF is a unique opportunity for participants to showcase their skills and knowledge while contributing to the conservation efforts of avian species worldwide. This year's competition is especially important for conservation efforts in Africa, where the impact of climate change, habitat loss, and poaching is threatening many bird species.\n\nAs hosts of the competition, we will try to be as active and responsive as possible to assist you in your endeavor. And all without giving away any secrets about the test data, so don't bother asking :)\n\nIn this thread, I will give all the hosts the opportunity to introduce themselves. I will be the first to do so.\n\nI am a research associate within the [K. Lisa Yang Center for Conservation Bioacoustics](https://www.birds.cornell.edu/ccb/) at the [Cornell Lab of Ornithology](https://www.birds.cornell.edu/home) and [Chemnitz University of Technology](https://www.tu-chemnitz.de/index.html.en), focusing on the development of advanced machine learning models for automatic detection and identification of bird species in large audio collections. Furthermore, I am the technology lead for the [BirdNET project](https://birdnet.cornell.edu/) and have been organizing the BirdCLEF Challenge since 2018. Feel free to ask me anything related to Deep Learning for bioacoustics, I might be able to help you.", "votes": 29 }, { "id": 2172849, "postDate": "2023-03-07T21:15:10.610Z", "content": "Hi, ya'll!
\nI'm a research software engineer at Google Brain, and have been co-hosting the BirdCLEF Kaggle competitions for a while now. My group is interested in organizing the sounds of the natural world and making them useful for conservation purposes. Or, put another way, finding ways to solve the basic ML problems such that ecologists can focus on answering ecological questions, instead of annotating data and training classifiers. You can read about some of my prior work on audio separation for birdsong here.
\nHappy hacking!
", "rawMarkdown": "Hi, ya'll!\n\nI'm a research software engineer at Google Brain, and have been co-hosting the BirdCLEF Kaggle competitions for a while now. My group is interested in organizing the sounds of the natural world and making them useful for conservation purposes. Or, put another way, finding ways to solve the basic ML problems such that ecologists can focus on answering ecological questions, instead of annotating data and training classifiers. You can read about some of my prior work on [audio separation for birdsong here](https://ai.googleblog.com/2022/01/separating-birdsong-in-wild-for.html).\n\nHappy hacking!", "votes": 9, "replies": [ { "id": 2181721, "postDate": "2023-03-14T18:00:55.200Z", "content": "Hi, Tom!
\nThank you so much for hosting this competition, My name is Dominic Imbuga, I perform data science at Savannaspace an AI, Cloud and Mobile development startup in Nairobi, Kenya . I'm very exited and thrilled that you are helping solve the birds problem in my country. I'm a big bird watcher and an avid data scientist.
\nThis competition is personal to me as a data scientist and I will give my best to help solve this problem.
\nCheers !
", "rawMarkdown": "Hi, Tom!\n\nThank you so much for hosting this competition, My name is Dominic Imbuga, I perform data science at Savannaspace an AI, Cloud and Mobile development startup in Nairobi, Kenya . I'm very exited and thrilled that you are helping solve the birds problem in my country. I'm a big bird watcher and an avid data scientist.\n\nThis competition is personal to me as a data scientist and I will give my best to help solve this problem.\n\nCheers !", "votes": 4, "replies": [ { "id": 2184983, "postDate": "2023-03-16T18:11:53.627Z", "content": "Karibu sana, Dominic!
\nFeel free to contribute recordings to Xeno-Canto when you have the chance. There are many under-represented species in East Africa - even common species can be undersampled, simply because people don't think to record them. It's also incredibly helpful to get some vocalizations from different places, so we can get a sense of individual and geographic variation.
\nAs Nathan Pieplow likes to say, it's pretty hard to take a really novel picture of a bird, but very easy to record some audio that is unknown to science. You also don't need expensive equipment: Cell phone recordings can be fine, and the cheaper Zoom microphones (like the H1) are great.
", "rawMarkdown": "Karibu sana, Dominic!\n\nFeel free to contribute recordings to Xeno-Canto when you have the chance. There are many under-represented species in East Africa - even common species can be undersampled, simply because people don't think to record them. It's also incredibly helpful to get some vocalizations from different places, so we can get a sense of individual and geographic variation.\n\nAs Nathan Pieplow likes to say, it's pretty hard to take a really novel picture of a bird, but very easy to record some audio that is unknown to science. You also don't need expensive equipment: Cell phone recordings can be fine, and the cheaper Zoom microphones (like the H1) are great.", "votes": 2 } ] } ] }, { "id": 2172880, "postDate": "2023-03-07T22:17:38.137Z", "content": "Hi everyone,
\nI am the director of the K. Lisa Yang Center for Conservation Bioacoustics at the Cornell Lab of Ornithology. The mission of our center is to collect and interpret sounds in nature by developing and applying innovative conservation technologies across ecologically-relevant scales to inspire and inform the conservation of wildlife and habitats.
\nOver the past several years, machine learning has transformed the field of bioacoustics and helped our community to do a better job conserving and protecting wildlife and their habitats across the globe. This transformation was, at least in part, sparked by competitions like this one, and I am thrilled that the BirdCLEF 2023 competition is now live. I want to thank everyone who worked very hard behind the scenes to make this competition possible (especially our friends at Google and Kaggle). It truly takes a village!
\nBest of luck, everyone. I am excited to see your solutions to detect and classify Eastern African birds acoustically.
\nMay the sound be with you ;)
\nHolger
", "rawMarkdown": "Hi everyone,\n\nI am the director of the [K. Lisa Yang Center for Conservation Bioacoustics](https://www.birds.cornell.edu/ccb/) at the [Cornell Lab of Ornithology](https://www.birds.cornell.edu/home/). The mission of our center is to collect and interpret sounds in nature by developing and applying innovative conservation technologies across ecologically-relevant scales to inspire and inform the conservation of wildlife and habitats. \n\nOver the past several years, machine learning has transformed the field of bioacoustics and helped our community to do a better job conserving and protecting wildlife and their habitats across the globe. This transformation was, at least in part, sparked by competitions like this one, and I am thrilled that the BirdCLEF 2023 competition is now live. I want to thank everyone who worked very hard behind the scenes to make this competition possible (especially our friends at Google and Kaggle). It truly takes a village! \n\nBest of luck, everyone. I am excited to see your solutions to detect and classify Eastern African birds acoustically.\n\nMay the sound be with you ;)\n\nHolger", "votes": 5, "replies": [ { "id": 2181722, "postDate": "2023-03-14T18:01:12.027Z", "content": "Hi, Holger !
\nThank you so much for hosting this competition, My name is Dominic Imbuga, I perform data science at Savannaspace an AI, Cloud and Mobile development startup in Nairobi, Kenya . I'm very exited and thrilled that you are helping solve the birds problem in my country. I'm a big bird watcher and an avid data scientist.
\nThis competition is personal to me as a data scientist and I will give my best to help solve this problem.
\nCheers !
", "rawMarkdown": "Hi, Holger !\n\nThank you so much for hosting this competition, My name is Dominic Imbuga, I perform data science at Savannaspace an AI, Cloud and Mobile development startup in Nairobi, Kenya . I'm very exited and thrilled that you are helping solve the birds problem in my country. I'm a big bird watcher and an avid data scientist.\n\nThis competition is personal to me as a data scientist and I will give my best to help solve this problem.\n\nCheers !", "votes": 1, "replies": [ { "id": 2185393, "postDate": "2023-03-17T02:52:10.920Z", "content": "You are most welcome. Thanks for participating and best of luck!
\nHolger
", "rawMarkdown": "You are most welcome. Thanks for participating and best of luck!\n\nHolger" } ] } ] }, { "id": 2209952, "postDate": "2023-04-05T03:40:17.390Z", "content": "Hi, could you clarify we expect there is no benefit to looking for more on xenocanto.org? Is it strictly prohibited?
No, it's not prohibited, it's just not encouraged, which means that we are trying to take some load off the XC servers. We included every eligible recording from XC (e.g., according to license type) in the dataset, so only very few species do have additional XC data online that can be used in this competition. However, if you decide to look for more recordings, make sure to follow the XC API guidelines and rules: https://xeno-canto.org/explore/api
", "rawMarkdown": "No, it's not prohibited, it's just not encouraged, which means that we are trying to take some load off the XC servers. We included every eligible recording from XC (e.g., according to license type) in the dataset, so only very few species do have additional XC data online that can be used in this competition. However, if you decide to look for more recordings, make sure to follow the XC API guidelines and rules: [https://xeno-canto.org/explore/api](https://xeno-canto.org/explore/api)", "votes": 3, "replies": [ { "id": 2215980, "postDate": "2023-04-09T17:29:39.467Z", "content": "@stefankahl regarding eligible recording from XC (e.g., according to license type)
\nAre we allowed to use only by-nc-sa or it is possible to use
by-sa or byby-nc-ndThe \"nd\" is the tricky part and a legal gray area, that's why we removed those from the dataset. Strictly speaking, extracting chunks from an XC recording is a derivative and therefore not permitted by the license.
", "rawMarkdown": "The \"nd\" is the tricky part and a legal gray area, that's why we removed those from the dataset. Strictly speaking, extracting chunks from an XC recording is a derivative and therefore not permitted by the license. ", "votes": 1 }, { "id": 2218967, "postDate": "2023-04-12T07:00:21.420Z", "content": "Hm
\nDo you mean that by-nc-nd originally comes from by-nc-sa and are simply sub-chunks of original recordings?
Although what about by-sa or by ?
And maybe you have a link with Docs about these licenses (it will be interesting to read)
\nThanks in advance!
", "rawMarkdown": "Hm\nDo you mean that `by-nc-nd` originally comes from `by-nc-sa` and are simply sub-chunks of original recordings?\n\nAlthough what about `by-sa` or `by` ?\n\nAnd maybe you have a link with Docs about these licenses (it will be interesting to read)\n\nThanks in advance! " }, { "id": 2219010, "postDate": "2023-04-12T07:41:45.383Z", "content": "No, \"nd\" means \"no derivative\" and restricts the use of the files as they have to remain unaltered. That's why we decided to not include those files in the training data. All other licenses are ok to use, if you want to know more, you can check the Creative Commons license type website here.
", "rawMarkdown": "No, \"nd\" means \"no derivative\" and restricts the use of the files as they have to remain unaltered. That's why we decided to not include those files in the training data. All other licenses are ok to use, if you want to know more, you can check the Creative Commons license type website [here](https://creativecommons.org/licenses/).", "votes": 2 }, { "id": 2219102, "postDate": "2023-04-12T09:51:32.970Z", "content": "Got it! Thanks!
", "rawMarkdown": "Got it! Thanks!" } ] } ] }, { "id": 2184223, "postDate": "2023-03-16T08:18:02.023Z", "content": "Does this competition allow hand labeling of training data?
", "rawMarkdown": "Does this competition allow hand labeling of training data?", "votes": 1, "replies": [ { "id": 2185760, "postDate": "2023-03-17T09:11:55.737Z", "content": "Sure. Ideally, you then share this data with all participants :)
", "rawMarkdown": "Sure. Ideally, you then share this data with all participants :)", "votes": 3 } ] }, { "id": 2185395, "postDate": "2023-03-17T02:54:18.283Z", "content": "Sure. Hand labeling is allowed. If you do so, we'd appreciate it if you would make the labels publicly available after the competition. Thanks!
", "rawMarkdown": "Sure. Hand labeling is allowed. If you do so, we'd appreciate it if you would make the labels publicly available after the competition. Thanks!", "votes": 2, "replies": [ { "id": 2185403, "postDate": "2023-03-17T03:18:37.500Z", "content": "Thank you.
", "rawMarkdown": "Thank you." } ] }, { "id": 2267999, "postDate": "2023-05-21T11:35:38.953Z", "content": "Hello, I am having a persistent submission score error and I can't understand the origin. Here is the link to my post \"https://www.kaggle.com/competitions/birdclef-2023/discussion/411863\", I hope you can help me. Thanks a lot :)
", "rawMarkdown": "Hello, I am having a persistent submission score error and I can't understand the origin. Here is the link to my post \"https://www.kaggle.com/competitions/birdclef-2023/discussion/411863\", I hope you can help me. Thanks a lot :)" }, { "id": 2263314, "postDate": "2023-05-17T13:59:05.720Z", "content": "Hi !, Is saving pre-processed audio data in form of numpy array and then uploading it and using as an input allowed ?
", "rawMarkdown": "Hi !, Is saving pre-processed audio data in form of numpy array and then uploading it and using as an input allowed ?" }, { "id": 2243333, "postDate": "2023-05-02T20:30:53.947Z", "content": "Hi Holgher, Stefan and Tom,
\nI'm new to Kaggle, many thanks to you and all participants for this opportunity.
\nI'm late in this competition and maybe the question has already been asked by someone else, in case I apologize. In train_metadata.csv there are 3 interesting columns, primary_label, secondary_labels and rating. I don't understand if what I'm about to ask is the subject of the competition but is it possible to know the rule for assigning the probability targets to the submission data for each of the 5s frames of an audio file ? In other words, in a bird file as the primary_label says, its 5s frames can have a distribution of probability targets (that sum to 1) based also on the secondary_labels and therefore which birds are present in that frame ? But if a file has no secondary_labels, that rule fixes the target to be the same for all the 5s frames (with probability 1 for a bird only) ? Finally, what is the role of the rating ?
Hi, Gin!
\nThe columns you mentioned are in the training data only, which is gathered from Xeno-Canto. The primary/secondary labels for XC are applied at the file level. The test data is hand-annotated soundscape data, which has bounding boxes for individual vocalizations. Navigating this difference between train and test is one of the many challenges in this space. :)
", "rawMarkdown": "Hi, Gin!\n\nThe columns you mentioned are in the training data only, which is gathered from Xeno-Canto. The primary/secondary labels for XC are applied at the file level. The test data is hand-annotated soundscape data, which has bounding boxes for individual vocalizations. Navigating this difference between train and test is one of the many challenges in this space. :)", "votes": 1, "replies": [ { "id": 2243820, "postDate": "2023-05-03T07:31:44.483Z", "content": "Thanks Tom, your answer is very helpful ! Sorry to bother you again, to be sure I understand, each 5 s frame in the submission data is attributed to a single species of bird (therefore only one probability = 1 in its target) i.e. I cannot have more probabilities for that 5 s frame which add up to 1 in its target. Is right ?
", "rawMarkdown": "Thanks Tom, your answer is very helpful ! Sorry to bother you again, to be sure I understand, each 5 s frame in the submission data is attributed to a single species of bird (therefore only one probability = 1 in its target) i.e. I cannot have more probabilities for that 5 s frame which add up to 1 in its target. Is right ? " }, { "id": 2249455, "postDate": "2023-05-07T18:10:29.723Z", "content": "Unfortunately birds are very rude and often speak over one another. The test task is a multi-class, multi-label problem - you can have multiple birds present and annotated in a single window. This /can/ be handled by treating each species as an independent binary classification problem, for example, instead of using a softmax activation on output probabilities. But many strategies are possible!
", "rawMarkdown": "Unfortunately birds are very rude and often speak over one another. The test task is a multi-class, multi-label problem - you can have multiple birds present and annotated in a single window. This /can/ be handled by treating each species as an independent binary classification problem, for example, instead of using a softmax activation on output probabilities. But many strategies are possible!", "votes": 2 }, { "id": 2256478, "postDate": "2023-05-12T14:11:24.493Z", "content": "Thank you very much Tom for your valuable and detailed information !
", "rawMarkdown": "Thank you very much Tom for your valuable and detailed information !" } ] } ] }, { "id": 2222730, "postDate": "2023-04-15T14:01:32.067Z", "content": "Hi, I am new to kaggle and confused by the external data license issue mentioned above. It seems that CC-BY-ND licensed data is not allowed. However, practically this only applies to prize winner, because of the \"WINNERS OBLIGATIONS\" in the rule, right? Does everyone has to submit their source data and related licenses?
", "rawMarkdown": "Hi, I am new to kaggle and confused by the external data license issue mentioned above. It seems that CC-BY-ND licensed data is not allowed. However, practically this only applies to prize winner, because of the \"WINNERS OBLIGATIONS\" in the rule, right? Does everyone has to submit their source data and related licenses?", "replies": [ { "id": 2223627, "postDate": "2023-04-16T13:14:22.467Z", "content": "No, license restrictions apply to everyone, not just winners. However, CC-BY-ND is in a bit of a gray area, which is why we didn't include recordings of Xeno-canto with that license in the dataset. This does not mean that the license is not allowed per se, we just wanted to make sure that we did not include anything that could be an issue.
", "rawMarkdown": "No, license restrictions apply to everyone, not just winners. However, CC-BY-ND is in a bit of a gray area, which is why we didn't include recordings of Xeno-canto with that license in the dataset. This does not mean that the license is not allowed per se, we just wanted to make sure that we did not include anything that could be an issue.", "votes": 1 } ] }, { "id": 2212811, "postDate": "2023-04-07T05:37:34.793Z", "content": "Hi, thank you for hosting the competition! I'm really learning and having fun throughout the process.
\nI'm trying to download the Northeastern United States bird dataset on zenodo, but zenodo is extremely slow for the past few days and estimates it to take 24 hours to download. Is there any other way to download this dataset?
", "rawMarkdown": "Hi, thank you for hosting the competition! I'm really learning and having fun throughout the process.\n\nI'm trying to download the [Northeastern United States bird dataset](https://zenodo.org/record/7079380) on zenodo, but zenodo is extremely slow for the past few days and estimates it to take 24 hours to download. Is there any other way to download this dataset?", "replies": [ { "id": 2212872, "postDate": "2023-04-07T06:24:20.583Z", "content": "Hmmm, not sure why Zenodo is so slow. Maybe try again later? Unfortunately, there's no other way to download the data.
", "rawMarkdown": "Hmmm, not sure why Zenodo is so slow. Maybe try again later? Unfortunately, there's no other way to download the data.", "votes": 1 } ] }, { "id": 2210081, "postDate": "2023-04-05T05:37:39.590Z", "content": "Hi Stefan and Holger,
\nI'm struggling with the submission scoring error as do other participants (described in this thread https://www.kaggle.com/competitions/birdclef-2023/discussion/399506). Is there are any place where we can seek assistance to resolve this?
", "rawMarkdown": "Hi Stefan and Holger,\n\nI'm struggling with the submission scoring error as do other participants (described in this thread https://www.kaggle.com/competitions/birdclef-2023/discussion/399506). Is there are any place where we can seek assistance to resolve this?", "replies": [ { "id": 2210728, "postDate": "2023-04-05T15:26:12.503Z", "content": "Hi, Jedrek -
\nWe strongly recommend re-using the code from the sample submission for creating your submissions. It's difficult to get the submission format right when starting from scratch, and re-using the example code will allow you to focus more on the modeling problem.
But I have checked that the generated CSV files are exactly the same, so what could possibly be the problem?
", "rawMarkdown": "But I have checked that the generated CSV files are exactly the same, so what could possibly be the problem?" }, { "id": 2210932, "postDate": "2023-04-05T18:03:43.707Z", "content": "By reading the original post you shared, I think you might miss the rows for the hidden test set. You can try to read the sample_submission and modify its values, instead of creating a new pd.DataFrame.
", "rawMarkdown": "By reading the original post you shared, I think you might miss the rows for the hidden test set. You can try to read the sample_submission and modify its values, instead of creating a new pd.DataFrame.", "votes": 1 }, { "id": 2211635, "postDate": "2023-04-06T06:52:37.800Z", "content": "Yes, this was precisely the problem, so you have to read in the sample dataframe provided, here is the crucial piece of code:
\ndf= pd.read_csv(\"/kaggle/input/birdclef-2023/sample_submission.csv\")\ndf[classes] = df[classes].astype(np.float32)\ndf[classes] = 0.0038\ndf.to_csv(\"submission.csv\", index=False)\nHi, ya'll!
\nI'm a research software engineer at Google Brain, and have been co-hosting the BirdCLEF Kaggle competitions for a while now. My group is interested in organizing the sounds of the natural world and making them useful for conservation purposes. Or, put another way, finding ways to solve the basic ML problems such that ecologists can focus on answering ecological questions, instead of annotating data and training classifiers. You can read about some of my prior work on audio separation for birdsong here.
\nHappy hacking!
", "votes": 9, "replies": [ { "id": 2181721, "author_name": "Dominic Imbuga", "author_url": "", "post_date": "2023-03-14T18:00:55.200000", "content": "Hi, Tom!
\nThank you so much for hosting this competition, My name is Dominic Imbuga, I perform data science at Savannaspace an AI, Cloud and Mobile development startup in Nairobi, Kenya . I'm very exited and thrilled that you are helping solve the birds problem in my country. I'm a big bird watcher and an avid data scientist.
\nThis competition is personal to me as a data scientist and I will give my best to help solve this problem.
\nCheers !
", "votes": 4, "replies": [ { "id": 2184983, "author_name": "Tom Denton", "author_url": "", "post_date": "2023-03-16T18:11:53.627000", "content": "Karibu sana, Dominic!
\nFeel free to contribute recordings to Xeno-Canto when you have the chance. There are many under-represented species in East Africa - even common species can be undersampled, simply because people don't think to record them. It's also incredibly helpful to get some vocalizations from different places, so we can get a sense of individual and geographic variation.
\nAs Nathan Pieplow likes to say, it's pretty hard to take a really novel picture of a bird, but very easy to record some audio that is unknown to science. You also don't need expensive equipment: Cell phone recordings can be fine, and the cheaper Zoom microphones (like the H1) are great.
", "votes": 2, "replies": [] } ] } ] }, { "id": 2172880, "author_name": "Holger Klinck", "author_url": "", "post_date": "2023-03-07T22:17:38.137000", "content": "Hi everyone,
\nI am the director of the K. Lisa Yang Center for Conservation Bioacoustics at the Cornell Lab of Ornithology. The mission of our center is to collect and interpret sounds in nature by developing and applying innovative conservation technologies across ecologically-relevant scales to inspire and inform the conservation of wildlife and habitats.
\nOver the past several years, machine learning has transformed the field of bioacoustics and helped our community to do a better job conserving and protecting wildlife and their habitats across the globe. This transformation was, at least in part, sparked by competitions like this one, and I am thrilled that the BirdCLEF 2023 competition is now live. I want to thank everyone who worked very hard behind the scenes to make this competition possible (especially our friends at Google and Kaggle). It truly takes a village!
\nBest of luck, everyone. I am excited to see your solutions to detect and classify Eastern African birds acoustically.
\nMay the sound be with you ;)
\nHolger
", "votes": 5, "replies": [ { "id": 2181722, "author_name": "Dominic Imbuga", "author_url": "", "post_date": "2023-03-14T18:01:12.027000", "content": "Hi, Holger !
\nThank you so much for hosting this competition, My name is Dominic Imbuga, I perform data science at Savannaspace an AI, Cloud and Mobile development startup in Nairobi, Kenya . I'm very exited and thrilled that you are helping solve the birds problem in my country. I'm a big bird watcher and an avid data scientist.
\nThis competition is personal to me as a data scientist and I will give my best to help solve this problem.
\nCheers !
", "votes": 1, "replies": [ { "id": 2185393, "author_name": "Holger Klinck", "author_url": "", "post_date": "2023-03-17T02:52:10.920000", "content": "You are most welcome. Thanks for participating and best of luck!
\nHolger
", "votes": 0, "replies": [] } ] } ] }, { "id": 2209952, "author_name": "Zhongkai Shangguan", "author_url": "", "post_date": "2023-04-05T03:40:17.390000", "content": "Hi, could you clarify we expect there is no benefit to looking for more on xenocanto.org? Is it strictly prohibited?
No, it's not prohibited, it's just not encouraged, which means that we are trying to take some load off the XC servers. We included every eligible recording from XC (e.g., according to license type) in the dataset, so only very few species do have additional XC data online that can be used in this competition. However, if you decide to look for more recordings, make sure to follow the XC API guidelines and rules: https://xeno-canto.org/explore/api
", "votes": 3, "replies": [ { "id": 2215980, "author_name": "Volodymyr", "author_url": "", "post_date": "2023-04-09T17:29:39.467000", "content": "@stefankahl regarding eligible recording from XC (e.g., according to license type)
\nAre we allowed to use only by-nc-sa or it is possible to use
by-sa or byby-nc-ndThe \"nd\" is the tricky part and a legal gray area, that's why we removed those from the dataset. Strictly speaking, extracting chunks from an XC recording is a derivative and therefore not permitted by the license.
", "votes": 1, "replies": [] }, { "id": 2218967, "author_name": "Volodymyr", "author_url": "", "post_date": "2023-04-12T07:00:21.420000", "content": "Hm
\nDo you mean that by-nc-nd originally comes from by-nc-sa and are simply sub-chunks of original recordings?
Although what about by-sa or by ?
And maybe you have a link with Docs about these licenses (it will be interesting to read)
\nThanks in advance!
", "votes": 0, "replies": [] }, { "id": 2219010, "author_name": "Stefan Kahl", "author_url": "", "post_date": "2023-04-12T07:41:45.383000", "content": "No, \"nd\" means \"no derivative\" and restricts the use of the files as they have to remain unaltered. That's why we decided to not include those files in the training data. All other licenses are ok to use, if you want to know more, you can check the Creative Commons license type website here.
", "votes": 2, "replies": [] }, { "id": 2219102, "author_name": "Volodymyr", "author_url": "", "post_date": "2023-04-12T09:51:32.970000", "content": "Got it! Thanks!
", "votes": 0, "replies": [] } ] } ] }, { "id": 2184223, "author_name": "shinmura0", "author_url": "", "post_date": "2023-03-16T08:18:02.023000", "content": "Does this competition allow hand labeling of training data?
", "votes": 1, "replies": [ { "id": 2185760, "author_name": "Stefan Kahl", "author_url": "", "post_date": "2023-03-17T09:11:55.737000", "content": "Sure. Ideally, you then share this data with all participants :)
", "votes": 3, "replies": [] } ] }, { "id": 2185395, "author_name": "Holger Klinck", "author_url": "", "post_date": "2023-03-17T02:54:18.283000", "content": "Sure. Hand labeling is allowed. If you do so, we'd appreciate it if you would make the labels publicly available after the competition. Thanks!
", "votes": 2, "replies": [ { "id": 2185403, "author_name": "shinmura0", "author_url": "", "post_date": "2023-03-17T03:18:37.500000", "content": "Thank you.
", "votes": 0, "replies": [] } ] }, { "id": 2267999, "author_name": "fares ernez", "author_url": "", "post_date": "2023-05-21T11:35:38.953000", "content": "Hello, I am having a persistent submission score error and I can't understand the origin. Here is the link to my post \"https://www.kaggle.com/competitions/birdclef-2023/discussion/411863\", I hope you can help me. Thanks a lot :)
", "votes": 0, "replies": [] }, { "id": 2263314, "author_name": "Kethupio", "author_url": "", "post_date": "2023-05-17T13:59:05.720000", "content": "Hi !, Is saving pre-processed audio data in form of numpy array and then uploading it and using as an input allowed ?
", "votes": 0, "replies": [] }, { "id": 2243333, "author_name": "Gin", "author_url": "", "post_date": "2023-05-02T20:30:53.947000", "content": "Hi Holgher, Stefan and Tom,
\nI'm new to Kaggle, many thanks to you and all participants for this opportunity.
\nI'm late in this competition and maybe the question has already been asked by someone else, in case I apologize. In train_metadata.csv there are 3 interesting columns, primary_label, secondary_labels and rating. I don't understand if what I'm about to ask is the subject of the competition but is it possible to know the rule for assigning the probability targets to the submission data for each of the 5s frames of an audio file ? In other words, in a bird file as the primary_label says, its 5s frames can have a distribution of probability targets (that sum to 1) based also on the secondary_labels and therefore which birds are present in that frame ? But if a file has no secondary_labels, that rule fixes the target to be the same for all the 5s frames (with probability 1 for a bird only) ? Finally, what is the role of the rating ?
Hi, Gin!
\nThe columns you mentioned are in the training data only, which is gathered from Xeno-Canto. The primary/secondary labels for XC are applied at the file level. The test data is hand-annotated soundscape data, which has bounding boxes for individual vocalizations. Navigating this difference between train and test is one of the many challenges in this space. :)
", "votes": 1, "replies": [ { "id": 2243820, "author_name": "Gin", "author_url": "", "post_date": "2023-05-03T07:31:44.483000", "content": "Thanks Tom, your answer is very helpful ! Sorry to bother you again, to be sure I understand, each 5 s frame in the submission data is attributed to a single species of bird (therefore only one probability = 1 in its target) i.e. I cannot have more probabilities for that 5 s frame which add up to 1 in its target. Is right ?
", "votes": 0, "replies": [] }, { "id": 2249455, "author_name": "Tom Denton", "author_url": "", "post_date": "2023-05-07T18:10:29.723000", "content": "Unfortunately birds are very rude and often speak over one another. The test task is a multi-class, multi-label problem - you can have multiple birds present and annotated in a single window. This /can/ be handled by treating each species as an independent binary classification problem, for example, instead of using a softmax activation on output probabilities. But many strategies are possible!
", "votes": 2, "replies": [] }, { "id": 2256478, "author_name": "Gin", "author_url": "", "post_date": "2023-05-12T14:11:24.493000", "content": "Thank you very much Tom for your valuable and detailed information !
", "votes": 0, "replies": [] } ] } ] }, { "id": 2222730, "author_name": "Aphysict", "author_url": "", "post_date": "2023-04-15T14:01:32.067000", "content": "Hi, I am new to kaggle and confused by the external data license issue mentioned above. It seems that CC-BY-ND licensed data is not allowed. However, practically this only applies to prize winner, because of the \"WINNERS OBLIGATIONS\" in the rule, right? Does everyone has to submit their source data and related licenses?
", "votes": 0, "replies": [ { "id": 2223627, "author_name": "Stefan Kahl", "author_url": "", "post_date": "2023-04-16T13:14:22.467000", "content": "No, license restrictions apply to everyone, not just winners. However, CC-BY-ND is in a bit of a gray area, which is why we didn't include recordings of Xeno-canto with that license in the dataset. This does not mean that the license is not allowed per se, we just wanted to make sure that we did not include anything that could be an issue.
", "votes": 1, "replies": [] } ] }, { "id": 2212811, "author_name": "lhanhsin", "author_url": "", "post_date": "2023-04-07T05:37:34.793000", "content": "Hi, thank you for hosting the competition! I'm really learning and having fun throughout the process.
\nI'm trying to download the Northeastern United States bird dataset on zenodo, but zenodo is extremely slow for the past few days and estimates it to take 24 hours to download. Is there any other way to download this dataset?
", "votes": 0, "replies": [ { "id": 2212872, "author_name": "Stefan Kahl", "author_url": "", "post_date": "2023-04-07T06:24:20.583000", "content": "Hmmm, not sure why Zenodo is so slow. Maybe try again later? Unfortunately, there's no other way to download the data.
", "votes": 1, "replies": [] } ] }, { "id": 2210081, "author_name": "Jędrek", "author_url": "", "post_date": "2023-04-05T05:37:39.590000", "content": "Hi Stefan and Holger,
\nI'm struggling with the submission scoring error as do other participants (described in this thread https://www.kaggle.com/competitions/birdclef-2023/discussion/399506). Is there are any place where we can seek assistance to resolve this?
", "votes": 0, "replies": [ { "id": 2210728, "author_name": "Tom Denton", "author_url": "", "post_date": "2023-04-05T15:26:12.503000", "content": "Hi, Jedrek -
\nWe strongly recommend re-using the code from the sample submission for creating your submissions. It's difficult to get the submission format right when starting from scratch, and re-using the example code will allow you to focus more on the modeling problem.
But I have checked that the generated CSV files are exactly the same, so what could possibly be the problem?
", "votes": 0, "replies": [] }, { "id": 2210932, "author_name": "Zhongkai Shangguan", "author_url": "", "post_date": "2023-04-05T18:03:43.707000", "content": "By reading the original post you shared, I think you might miss the rows for the hidden test set. You can try to read the sample_submission and modify its values, instead of creating a new pd.DataFrame.
", "votes": 1, "replies": [] }, { "id": 2211635, "author_name": "Jędrek", "author_url": "", "post_date": "2023-04-06T06:52:37.800000", "content": "Yes, this was precisely the problem, so you have to read in the sample dataframe provided, here is the crucial piece of code:
\ndf= pd.read_csv(\"/kaggle/input/birdclef-2023/sample_submission.csv\")\ndf[classes] = df[classes].astype(np.float32)\ndf[classes] = 0.0038\ndf.to_csv(\"submission.csv\", index=False)\n