{
  "id": 229171,
  "title": "Only single species identified per image",
  "url": "/competitions/iwildcam2021-fgvc8/discussion/229171",
  "author_name": "",
  "post_date": "2021-03-28T17:54:13.642346200Z",
  "votes": 3,
  "comment_count": 5,
  "views": 0,
  "content": "<p>In the training set, there appears to be only a single species identified per image. </p>\n<pre><code>BASE_DIR = '/kaggle/input/iwildcam2021-fgvc8'\n\nwith open(f'{BASE_DIR}/metadata/iwildcam2021_train_annotations.json') as f:\n    metadata = json.load(f)\n\nannotations = pd.DataFrame(metadata['annotations'])\nannotations['image_id'].duplicated().any()\n&gt;&gt; False\n</code></pre>\n<p>Is it really the case that there were only a single species identified per image?</p>\n<p>I found a seq_id that definitely violates this assumption. seq_id 3001cbf6-7d42-11eb-8fb5-0242ac1c0002 definitely has a zebra and giraffe in the same image but only a single species is listed per image. </p>",
  "messages": [
    {
      "id": "1255353",
      "postDate": "03/28/2021 17:54:13",
      "content": "<p>In the training set, there appears to be only a single species identified per image. </p>\n<pre><code>BASE_DIR = '/kaggle/input/iwildcam2021-fgvc8'\n\nwith open(f'{BASE_DIR}/metadata/iwildcam2021_train_annotations.json') as f:\n    metadata = json.load(f)\n\nannotations = pd.DataFrame(metadata['annotations'])\nannotations['image_id'].duplicated().any()\n&gt;&gt; False\n</code></pre>\n<p>Is it really the case that there were only a single species identified per image?</p>\n<p>I found a seq_id that definitely violates this assumption. seq_id 3001cbf6-7d42-11eb-8fb5-0242ac1c0002 definitely has a zebra and giraffe in the same image but only a single species is listed per image. </p>",
      "rawMarkdown": "In the training set, there appears to be only a single species identified per image. \n\n```\nBASE_DIR = '/kaggle/input/iwildcam2021-fgvc8'\n\nwith open(f'{BASE_DIR}/metadata/iwildcam2021_train_annotations.json') as f:\n    metadata = json.load(f)\n\nannotations = pd.DataFrame(metadata['annotations'])\nannotations['image_id'].duplicated().any()\n>> False\n```\n\nIs it really the case that there were only a single species identified per image?\n\nI found a seq_id that definitely violates this assumption. seq_id 3001cbf6-7d42-11eb-8fb5-0242ac1c0002 definitely has a zebra and giraffe in the same image but only a single species is listed per image.",
      "votes": null
    },
    {
      "id": "1256116",
      "postDate": "03/29/2021 14:56:59",
      "content": "<p>We provide the labels for the training set as they were provided to us by the Wildlife Conservation Society, and have not manually re-labeled the training data to catch errors that the ecologists may have made. We did go through and re-label the test data for counts, and (we believe) found and correctly labeled all counts, including those with multiple species, in the test sequences.</p>",
      "rawMarkdown": "We provide the labels for the training set as they were provided to us by the Wildlife Conservation Society, and have not manually re-labeled the training data to catch errors that the ecologists may have made. We did go through and re-label the test data for counts, and (we believe) found and correctly labeled all counts, including those with multiple species, in the test sequences.",
      "votes": null
    },
    {
      "id": "1260033",
      "postDate": "04/01/2021 19:30:01",
      "content": "<p>Can we relabel the training data ourselves? </p>",
      "rawMarkdown": "Can we relabel the training data ourselves?",
      "votes": null
    },
    {
      "id": "1260038",
      "postDate": "04/01/2021 19:35:44",
      "content": "<p>Yes absolutely!! We just ask that you share any annotations you collect to keep things fair (this is mentioned in the rules)</p>",
      "rawMarkdown": "Yes absolutely!! We just ask that you share any annotations you collect to keep things fair (this is mentioned in the rules)",
      "votes": null
    },
    {
      "id": "1271440",
      "postDate": "04/12/2021 15:52:58",
      "content": "<p>thank you! of course :) what's the best way to share them? just post a link to a google drive? </p>",
      "rawMarkdown": "thank you! of course :) what's the best way to share them? just post a link to a google drive?",
      "votes": null
    },
    {
      "id": "1271447",
      "postDate": "04/12/2021 16:00:19",
      "content": "<p>Yes, that works for me!</p>",
      "rawMarkdown": "Yes, that works for me!",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1256116,
      "author_name": "sbeery",
      "author_url": "",
      "post_date": "03/29/2021 14:56:59",
      "content": "<p>We provide the labels for the training set as they were provided to us by the Wildlife Conservation Society, and have not manually re-labeled the training data to catch errors that the ecologists may have made. We did go through and re-label the test data for counts, and (we believe) found and correctly labeled all counts, including those with multiple species, in the test sequences.</p>",
      "votes": null,
      "replies": [
        {
          "id": 1260033,
          "author_name": "psicobloc",
          "author_url": "",
          "post_date": "04/01/2021 19:30:01",
          "content": "<p>Can we relabel the training data ourselves? </p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1260038,
          "author_name": "sbeery",
          "author_url": "",
          "post_date": "04/01/2021 19:35:44",
          "content": "<p>Yes absolutely!! We just ask that you share any annotations you collect to keep things fair (this is mentioned in the rules)</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1271440,
          "author_name": "psicobloc",
          "author_url": "",
          "post_date": "04/12/2021 15:52:58",
          "content": "<p>thank you! of course :) what's the best way to share them? just post a link to a google drive? </p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1271447,
          "author_name": "sbeery",
          "author_url": "",
          "post_date": "04/12/2021 16:00:19",
          "content": "<p>Yes, that works for me!</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1255353": "In the training set, there appears to be only a single species identified per image. \n\n```\nBASE_DIR = '/kaggle/input/iwildcam2021-fgvc8'\n\nwith open(f'{BASE_DIR}/metadata/iwildcam2021_train_annotations.json') as f:\n    metadata = json.load(f)\n\nannotations = pd.DataFrame(metadata['annotations'])\nannotations['image_id'].duplicated().any()\n>> False\n```\n\nIs it really the case that there were only a single species identified per image?\n\nI found a seq_id that definitely violates this assumption. seq_id 3001cbf6-7d42-11eb-8fb5-0242ac1c0002 definitely has a zebra and giraffe in the same image but only a single species is listed per image.",
    "1256116": "We provide the labels for the training set as they were provided to us by the Wildlife Conservation Society, and have not manually re-labeled the training data to catch errors that the ecologists may have made. We did go through and re-label the test data for counts, and (we believe) found and correctly labeled all counts, including those with multiple species, in the test sequences.",
    "1260033": "Can we relabel the training data ourselves?",
    "1260038": "Yes absolutely!! We just ask that you share any annotations you collect to keep things fair (this is mentioned in the rules)",
    "1271440": "thank you! of course :) what's the best way to share them? just post a link to a google drive?",
    "1271447": "Yes, that works for me!"
  },
  "source": "meta"
}