{
  "id": 225227,
  "title": "New iNaturalist 2021 Challenge Launched on Kaggle",
  "url": "/competitions/inaturalist-challenge-at-fgvc-2017/discussion/225227",
  "author_name": "macaodha",
  "post_date": "2021-03-11T10:42:31.056000",
  "votes": 0,
  "comment_count": 0,
  "views": 0,
  "content": "<p>Greetings everyone, </p>\n<p>We have just launched a new iNaturalist Challenge on Kaggle:<br>\n<a href=\"https://www.kaggle.com/c/inaturalist-2021\" target=\"_blank\">https://www.kaggle.com/c/inaturalist-2021</a></p>\n<p>This new competition features even more species and images along with additional location meta data. More details below. </p>\n<p>Oisin</p>\n<p>We made a few modifications to the competition this year. Similar to the 2017 competition, we are releasing the species names immediately, instead of obfuscating them. Our reason for obfuscating them in 2018 and 2019 was to make it difficult for competitors to scrape the web (or iNaturalist itself) for additional images. Because we are releasing 2.7M training images and the dataset doesn't necessarily focus on the long tail problem we feel that we can release the species names without worry. This does not mean that scraping is allowed. Please do not scrape for additional data, especially from iNaturalist. Having the species names also makes interpreting validation results easier when examining confusion matrices and accuracy statistics.</p>\n<p>We are also releasing location and date information for each image in the form of latitude, longitude, location_uncertainty, and date values. We have retroactively added this information to the 2017 and 2018 datasets, but this year competitors are able to utilize this information when building models. We hope this motivates competitors to devise interesting solutions to this large scale problem.</p>",
  "messages": [
    {
      "id": 1234526,
      "postDate": "2021-03-11T10:42:31.057Z",
      "content": "<p>Greetings everyone, </p>\n<p>We have just launched a new iNaturalist Challenge on Kaggle:<br>\n<a href=\"https://www.kaggle.com/c/inaturalist-2021\" target=\"_blank\">https://www.kaggle.com/c/inaturalist-2021</a></p>\n<p>This new competition features even more species and images along with additional location meta data. More details below. </p>\n<p>Oisin</p>\n<p>We made a few modifications to the competition this year. Similar to the 2017 competition, we are releasing the species names immediately, instead of obfuscating them. Our reason for obfuscating them in 2018 and 2019 was to make it difficult for competitors to scrape the web (or iNaturalist itself) for additional images. Because we are releasing 2.7M training images and the dataset doesn't necessarily focus on the long tail problem we feel that we can release the species names without worry. This does not mean that scraping is allowed. Please do not scrape for additional data, especially from iNaturalist. Having the species names also makes interpreting validation results easier when examining confusion matrices and accuracy statistics.</p>\n<p>We are also releasing location and date information for each image in the form of latitude, longitude, location_uncertainty, and date values. We have retroactively added this information to the 2017 and 2018 datasets, but this year competitors are able to utilize this information when building models. We hope this motivates competitors to devise interesting solutions to this large scale problem.</p>",
      "rawMarkdown": "Greetings everyone, \n\nWe have just launched a new iNaturalist Challenge on Kaggle:\nhttps://www.kaggle.com/c/inaturalist-2021\n\nThis new competition features even more species and images along with additional location meta data. More details below. \n\nOisin\n\n\nWe made a few modifications to the competition this year. Similar to the 2017 competition, we are releasing the species names immediately, instead of obfuscating them. Our reason for obfuscating them in 2018 and 2019 was to make it difficult for competitors to scrape the web (or iNaturalist itself) for additional images. Because we are releasing 2.7M training images and the dataset doesn't necessarily focus on the long tail problem we feel that we can release the species names without worry. This does not mean that scraping is allowed. Please do not scrape for additional data, especially from iNaturalist. Having the species names also makes interpreting validation results easier when examining confusion matrices and accuracy statistics.\n\nWe are also releasing location and date information for each image in the form of latitude, longitude, location_uncertainty, and date values. We have retroactively added this information to the 2017 and 2018 datasets, but this year competitors are able to utilize this information when building models. We hope this motivates competitors to devise interesting solutions to this large scale problem."
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "1234526": "Greetings everyone, \n\nWe have just launched a new iNaturalist Challenge on Kaggle:\nhttps://www.kaggle.com/c/inaturalist-2021\n\nThis new competition features even more species and images along with additional location meta data. More details below. \n\nOisin\n\n\nWe made a few modifications to the competition this year. Similar to the 2017 competition, we are releasing the species names immediately, instead of obfuscating them. Our reason for obfuscating them in 2018 and 2019 was to make it difficult for competitors to scrape the web (or iNaturalist itself) for additional images. Because we are releasing 2.7M training images and the dataset doesn't necessarily focus on the long tail problem we feel that we can release the species names without worry. This does not mean that scraping is allowed. Please do not scrape for additional data, especially from iNaturalist. Having the species names also makes interpreting validation results easier when examining confusion matrices and accuracy statistics.\n\nWe are also releasing location and date information for each image in the form of latitude, longitude, location_uncertainty, and date values. We have retroactively added this information to the 2017 and 2018 datasets, but this year competitors are able to utilize this information when building models. We hope this motivates competitors to devise interesting solutions to this large scale problem."
  }
}