{
  "id": 492752,
  "title": "Clarification on submission format",
  "url": "/competitions/birdclef-2024/discussion/492752",
  "author_name": "",
  "post_date": "2024-04-10T20:48:21.347793Z",
  "votes": 2,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Hi,</p>\n<p>I saw in the submission sample that for a specific sound test file, there are several entries, with only end times specified (5 seconds I assume). Should we for each sound test file split the file in 5 seconds sample and evaluate these splitted samples ?  If there is no bird sound during a period of time, what do we do ? Do we drop the corresponding samples ? </p>\n<p>Thanks,<br>\nStéphane</p>",
  "messages": [
    {
      "id": "2745809",
      "postDate": "04/10/2024 20:48:21",
      "content": "<p>Hi,</p>\n<p>I saw in the submission sample that for a specific sound test file, there are several entries, with only end times specified (5 seconds I assume). Should we for each sound test file split the file in 5 seconds sample and evaluate these splitted samples ?  If there is no bird sound during a period of time, what do we do ? Do we drop the corresponding samples ? </p>\n<p>Thanks,<br>\nStéphane</p>",
      "rawMarkdown": "Hi,\n\nI saw in the submission sample that for a specific sound test file, there are several entries, with only end times specified (5 seconds I assume). Should we for each sound test file split the file in 5 seconds sample and evaluate these splitted samples ?  If there is no bird sound during a period of time, what do we do ? Do we drop the corresponding samples ? \n\nThanks,\nStéphane",
      "votes": null
    },
    {
      "id": "2745829",
      "postDate": "04/10/2024 21:32:06",
      "content": "<p>Hello Stéphane. <br>\nNo, Ideally your model needs to predicts full 0s for those segments, which are actually very common if you listen to the unlabelled soundscapes. For each file you need 48 predictions, named {filename_without_the_extention}_{ending_segment_seconds}. The challenge of the comp is that some segments don't have birds, some have many, and you need to predict them all correctely from the given data.</p>",
      "rawMarkdown": "Hello Stéphane. \nNo, Ideally your model needs to predicts full 0s for those segments, which are actually very common if you listen to the unlabelled soundscapes. For each file you need 48 predictions, named {filename_without_the_extention}_{ending_segment_seconds}. The challenge of the comp is that some segments don't have birds, some have many, and you need to predict them all correctely from the given data.",
      "votes": null
    },
    {
      "id": "2745848",
      "postDate": "04/10/2024 22:00:09",
      "content": "<p>Ok, thanks for the clarification !</p>",
      "rawMarkdown": "Ok, thanks for the clarification !",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2745829,
      "author_name": "janmpia",
      "author_url": "",
      "post_date": "04/10/2024 21:32:06",
      "content": "<p>Hello Stéphane. <br>\nNo, Ideally your model needs to predicts full 0s for those segments, which are actually very common if you listen to the unlabelled soundscapes. For each file you need 48 predictions, named {filename_without_the_extention}_{ending_segment_seconds}. The challenge of the comp is that some segments don't have birds, some have many, and you need to predict them all correctely from the given data.</p>",
      "votes": null,
      "replies": [
        {
          "id": 2745848,
          "author_name": "stephdamico",
          "author_url": "",
          "post_date": "04/10/2024 22:00:09",
          "content": "<p>Ok, thanks for the clarification !</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "2745809": "Hi,\n\nI saw in the submission sample that for a specific sound test file, there are several entries, with only end times specified (5 seconds I assume). Should we for each sound test file split the file in 5 seconds sample and evaluate these splitted samples ?  If there is no bird sound during a period of time, what do we do ? Do we drop the corresponding samples ? \n\nThanks,\nStéphane",
    "2745829": "Hello Stéphane. \nNo, Ideally your model needs to predicts full 0s for those segments, which are actually very common if you listen to the unlabelled soundscapes. For each file you need 48 predictions, named {filename_without_the_extention}_{ending_segment_seconds}. The challenge of the comp is that some segments don't have birds, some have many, and you need to predict them all correctely from the given data.",
    "2745848": "Ok, thanks for the clarification !"
  },
  "source": "meta"
}