Hi, I am not sure what does end_time mean here why does one soundscape_id has 3 rows with same soundscape_id and different end time? Should be split one sound to duration of 5 seconds each? Need some clarification on this part please.
", "messages": [ { "id": 2173137, "postDate": "2023-03-08T05:52:43.040Z", "content": "Hi, I am not sure what does end_time mean here why does one soundscape_id has 3 rows with same soundscape_id and different end time? Should be split one sound to duration of 5 seconds each? Need some clarification on this part please.
", "rawMarkdown": "Hi, I am not sure what does end_time mean here why does one soundscape_id has 3 rows with same soundscape_id and different end time? Should be split one sound to duration of 5 seconds each? Need some clarification on this part please.", "votes": 12 }, { "id": 2173251, "postDate": "2023-03-08T08:18:46.480Z", "content": "Yes, soundscapes need to be split into 5-second chunks and the submissions file contains predictions for all species across all 5-second chunks. This way, soundscape_12345_15 denotes the audio chunk between timestamps 00:00:10 and 00:00:15.
", "rawMarkdown": "Yes, soundscapes need to be split into 5-second chunks and the submissions file contains predictions for all species across all 5-second chunks. This way, soundscape_12345_15 denotes the audio chunk between timestamps 00:00:10 and 00:00:15.", "votes": 5, "replies": [ { "id": 2174254, "postDate": "2023-03-09T02:08:03.400Z", "rawMarkdown": "", "isDeleted": true }, { "id": 2174916, "postDate": "2023-03-09T14:08:18.037Z", "content": "hi, so Is this a multi-label task?
\nAs we can see, the test audio is 10 minutes long which we need to predict every 5-second chunks. So is there only one kind of bird in each 5-second chunks, or there could be more than one kind of bird?
\nIt leads to there should be only one max value in the submission row, or more than one max value to predict.
Yes, it is a multi-label task. There might be more than one bird per 5-second segment. You need to provide scores for each species for each segment, which will then be treated as a ranked list to compute the ranking metric cmAP.
", "rawMarkdown": "Yes, it is a multi-label task. There might be more than one bird per 5-second segment. You need to provide scores for each species for each segment, which will then be treated as a ranked list to compute the ranking metric cmAP.", "votes": 4 }, { "id": 2175570, "postDate": "2023-03-10T00:51:43.300Z", "content": "Thx! As u said, for test data, there might be more than one kind of birds. But for training data, there should be only one kind of bird in each ogg file cause there is only one label for each ogg file, is this right?
", "rawMarkdown": "Thx! As u said, for test data, there might be more than one kind of birds. But for training data, there should be only one kind of bird in each ogg file cause there is only one label for each ogg file, is this right?" }, { "id": 2175576, "postDate": "2023-03-10T01:04:00.883Z", "content": "One of the hard parts of bioacoustics is that there's really very little 'clean' data - we keep inviting the birds to the recording studio, but they never return our emails…
\nSome Xeno-Canto files are pretty clean, but others may have many birds vocalizing; /sometimes/ these are recorded in the 'background species' field in the metadata, if the audio uploader filled it out (and they knew what the background species actually were).
\nTypically the target species will be the loudest/clearest in a given recording, however.
", "rawMarkdown": "One of the hard parts of bioacoustics is that there's really very little 'clean' data - we keep inviting the birds to the recording studio, but they never return our emails... \n\nSome Xeno-Canto files are pretty clean, but others may have many birds vocalizing; /sometimes/ these are recorded in the 'background species' field in the metadata, if the audio uploader filled it out (and they knew what the background species actually were). \n\nTypically the target species will be the loudest/clearest in a given recording, however.", "votes": 4 }, { "id": 2175605, "postDate": "2023-03-10T02:11:51.300Z", "content": "Got it, very clear explanation, thx again!
", "rawMarkdown": "Got it, very clear explanation, thx again!" } ] }, { "id": 2175103, "postDate": "2023-03-09T16:39:31.100Z", "content": "If this is the case then the test should have 120 rows for each soundscape? Each test soundscape duration is 600 seconds, is this correct? Then the test set should have 24000 rows because they are 200 test soundscapes.
\nThe sample submission only have the first 3 chunks just for faster inference? Actually this test soundscape has 120 chunks?
\nIs this correct?, thanks for the help
\nAlso the train data is not multi-label task, it is just multi-class classification task, therefore for each observation we can only have 1 bird? then the model trained with this data will not be aligned with the test data, it will be hard to align the CV metric with the LB.
", "rawMarkdown": "If this is the case then the test should have 120 rows for each soundscape? Each test soundscape duration is 600 seconds, is this correct? Then the test set should have 24000 rows because they are 200 test soundscapes.\n\nThe sample submission only have the first 3 chunks just for faster inference? Actually this test soundscape has 120 chunks?\n\nIs this correct?, thanks for the help\n\nAlso the train data is not multi-label task, it is just multi-class classification task, therefore for each observation we can only have 1 bird? then the model trained with this data will not be aligned with the test data, it will be hard to align the CV metric with the LB.", "votes": 1, "replies": [ { "id": 2175170, "postDate": "2023-03-09T17:27:03.707Z", "content": "The copy of the sample submission that you can download was truncated to just the first few rows.
", "rawMarkdown": "The copy of the sample submission that you can download was truncated to just the first few rows.", "votes": 1 }, { "id": 2176722, "postDate": "2023-03-10T21:26:19.243Z", "content": "The same questions occurred to me: Is this a multi-label or multi-class problem? Each train sample seems to be associated with a single target bird (primary_label), but what exactly does the secondary_label column signify?
", "rawMarkdown": "The same questions occurred to me: Is this a multi-label or multi-class problem? Each train sample seems to be associated with a single target bird (primary_label), but what exactly does the secondary_label column signify?", "votes": 2 } ] }, { "id": 2179349, "postDate": "2023-03-13T04:59:40.403Z", "content": "\n\nYes, soundscapes need to be split into 5-second chunks and the submissions file contains predictions for all species across all 5-second chunks. This way, soundscape_12345_15 denotes the audio chunk between timestamps 00:00:10 and 00:00:15.
\n
Thank You for clarifying my doubt.
", "rawMarkdown": "> Yes, soundscapes need to be split into 5-second chunks and the submissions file contains predictions for all species across all 5-second chunks. This way, soundscape_12345_15 denotes the audio chunk between timestamps 00:00:10 and 00:00:15.\n\nThank You for clarifying my doubt." }, { "id": 2218415, "postDate": "2023-04-11T16:27:47.990Z", "content": "Hi @stefankahl
\nI could not find this information in the competition's dataset description. I was mainly assuming that the audio recordings need to be split into 5s recordings just based on the submission format and various other submissions. If this information can be added into the main page, it would be extremely beneficial. Please correct me if I am wrong. Thanks.
I see, I'll make sure we add this info to the evaluation section. Thanks.
", "rawMarkdown": "I see, I'll make sure we add this info to the evaluation section. Thanks." } ] } ] }, { "id": 2180542, "postDate": "2023-03-13T22:52:35.637Z", "content": "sounds end
", "rawMarkdown": "sounds end" }, { "id": 2173883, "postDate": "2023-03-08T17:41:51.753Z", "content": "@prateekcoder You only need to read row_id from sample submission files. This files contains all chunks you need to predict.
", "rawMarkdown": "@prateekcoder You only need to read row_id from sample submission files. This files contains all chunks you need to predict.", "replies": [ { "id": 2175101, "postDate": "2023-03-09T16:39:11.747Z", "rawMarkdown": "", "isDeleted": true } ] } ], "comments": [ { "id": 2173251, "author_name": "Stefan Kahl", "author_url": "", "post_date": "2023-03-08T08:18:46.480000", "content": "Yes, soundscapes need to be split into 5-second chunks and the submissions file contains predictions for all species across all 5-second chunks. This way, soundscape_12345_15 denotes the audio chunk between timestamps 00:00:10 and 00:00:15.
", "votes": 5, "replies": [ { "id": 2174254, "author_name": "", "author_url": "", "post_date": "2023-03-09T02:08:03.400000", "content": "", "votes": 0, "replies": [] }, { "id": 2174916, "author_name": "Mr.Fire", "author_url": "", "post_date": "2023-03-09T14:08:18.037000", "content": "hi, so Is this a multi-label task?
\nAs we can see, the test audio is 10 minutes long which we need to predict every 5-second chunks. So is there only one kind of bird in each 5-second chunks, or there could be more than one kind of bird?
\nIt leads to there should be only one max value in the submission row, or more than one max value to predict.
Yes, it is a multi-label task. There might be more than one bird per 5-second segment. You need to provide scores for each species for each segment, which will then be treated as a ranked list to compute the ranking metric cmAP.
", "votes": 4, "replies": [] }, { "id": 2175570, "author_name": "Mr.Fire", "author_url": "", "post_date": "2023-03-10T00:51:43.300000", "content": "Thx! As u said, for test data, there might be more than one kind of birds. But for training data, there should be only one kind of bird in each ogg file cause there is only one label for each ogg file, is this right?
", "votes": 0, "replies": [] }, { "id": 2175576, "author_name": "Tom Denton", "author_url": "", "post_date": "2023-03-10T01:04:00.883000", "content": "One of the hard parts of bioacoustics is that there's really very little 'clean' data - we keep inviting the birds to the recording studio, but they never return our emails…
\nSome Xeno-Canto files are pretty clean, but others may have many birds vocalizing; /sometimes/ these are recorded in the 'background species' field in the metadata, if the audio uploader filled it out (and they knew what the background species actually were).
\nTypically the target species will be the loudest/clearest in a given recording, however.
", "votes": 4, "replies": [] }, { "id": 2175605, "author_name": "Mr.Fire", "author_url": "", "post_date": "2023-03-10T02:11:51.300000", "content": "Got it, very clear explanation, thx again!
", "votes": 0, "replies": [] } ] }, { "id": 2175103, "author_name": "Martin Kovacevic Buvinic", "author_url": "", "post_date": "2023-03-09T16:39:31.100000", "content": "If this is the case then the test should have 120 rows for each soundscape? Each test soundscape duration is 600 seconds, is this correct? Then the test set should have 24000 rows because they are 200 test soundscapes.
\nThe sample submission only have the first 3 chunks just for faster inference? Actually this test soundscape has 120 chunks?
\nIs this correct?, thanks for the help
\nAlso the train data is not multi-label task, it is just multi-class classification task, therefore for each observation we can only have 1 bird? then the model trained with this data will not be aligned with the test data, it will be hard to align the CV metric with the LB.
", "votes": 1, "replies": [ { "id": 2175170, "author_name": "Sohier Dane", "author_url": "", "post_date": "2023-03-09T17:27:03.707000", "content": "The copy of the sample submission that you can download was truncated to just the first few rows.
", "votes": 1, "replies": [] }, { "id": 2176722, "author_name": "David J. Slate", "author_url": "", "post_date": "2023-03-10T21:26:19.243000", "content": "The same questions occurred to me: Is this a multi-label or multi-class problem? Each train sample seems to be associated with a single target bird (primary_label), but what exactly does the secondary_label column signify?
", "votes": 2, "replies": [] } ] }, { "id": 2179349, "author_name": "Prateek", "author_url": "", "post_date": "2023-03-13T04:59:40.403000", "content": "\n\nYes, soundscapes need to be split into 5-second chunks and the submissions file contains predictions for all species across all 5-second chunks. This way, soundscape_12345_15 denotes the audio chunk between timestamps 00:00:10 and 00:00:15.
\n
Thank You for clarifying my doubt.
", "votes": 0, "replies": [] }, { "id": 2218415, "author_name": "Abhishek Varshney", "author_url": "", "post_date": "2023-04-11T16:27:47.990000", "content": "Hi @stefankahl
\nI could not find this information in the competition's dataset description. I was mainly assuming that the audio recordings need to be split into 5s recordings just based on the submission format and various other submissions. If this information can be added into the main page, it would be extremely beneficial. Please correct me if I am wrong. Thanks.
I see, I'll make sure we add this info to the evaluation section. Thanks.
", "votes": 0, "replies": [] } ] } ] }, { "id": 2180542, "author_name": "Sibi Vishtan", "author_url": "", "post_date": "2023-03-13T22:52:35.637000", "content": "sounds end
", "votes": 0, "replies": [] }, { "id": 2173883, "author_name": "Anh Bui", "author_url": "", "post_date": "2023-03-08T17:41:51.753000", "content": "@prateekcoder You only need to read row_id from sample submission files. This files contains all chunks you need to predict.
", "votes": 0, "replies": [ { "id": 2175101, "author_name": "", "author_url": "", "post_date": "2023-03-09T16:39:11.747000", "content": "", "votes": 0, "replies": [] } ] } ], "raw_markdown_by_id": { "2173137": "Hi, I am not sure what does end_time mean here why does one soundscape_id has 3 rows with same soundscape_id and different end time? Should be split one sound to duration of 5 seconds each? Need some clarification on this part please.", "2173251": "Yes, soundscapes need to be split into 5-second chunks and the submissions file contains predictions for all species across all 5-second chunks. This way, soundscape_12345_15 denotes the audio chunk between timestamps 00:00:10 and 00:00:15.", "2180542": "sounds end", "2173883": "@prateekcoder You only need to read row_id from sample submission files. This files contains all chunks you need to predict." } }