{
  "id": 109519,
  "title": "Download Only Segmentation Dataset",
  "url": "/competitions/open-images-2019-instance-segmentation/discussion/109519",
  "author_name": "",
  "post_date": "2019-09-19T19:09:37.585117100Z",
  "votes": null,
  "comment_count": 2,
  "views": 0,
  "content": "<p>@benenson @akuznetsa or anyone else who knows.</p>\n\n<p>On this website: <a href=\"https://storage.googleapis.com/openimages/web/factsfigures.html\">https://storage.googleapis.com/openimages/web/factsfigures.html</a>\nIt states there are 944,037 training images for segmentation. </p>\n\n<p>But on this website: <a href=\"https://github.com/cvdfoundation/open-images-dataset#download-images-with-bounding-boxes-annotations\">https://github.com/cvdfoundation/open-images-dataset#download-images-with-bounding-boxes-annotations</a>\nThere are 1,743,042 train images for bounding boxes. </p>\n\n<p>How do I download only the 944,037 segmentation train images instead of the full 1,743,042 images for bounding boxes?</p>\n\n<p>Thank you.</p>",
  "messages": [
    {
      "id": "630125",
      "postDate": "09/19/2019 19:09:37",
      "content": "<p>@benenson @akuznetsa or anyone else who knows.</p>\n\n<p>On this website: <a href=\"https://storage.googleapis.com/openimages/web/factsfigures.html\">https://storage.googleapis.com/openimages/web/factsfigures.html</a>\nIt states there are 944,037 training images for segmentation. </p>\n\n<p>But on this website: <a href=\"https://github.com/cvdfoundation/open-images-dataset#download-images-with-bounding-boxes-annotations\">https://github.com/cvdfoundation/open-images-dataset#download-images-with-bounding-boxes-annotations</a>\nThere are 1,743,042 train images for bounding boxes. </p>\n\n<p>How do I download only the 944,037 segmentation train images instead of the full 1,743,042 images for bounding boxes?</p>\n\n<p>Thank you.</p>",
      "rawMarkdown": "benenson @akuznetsa or anyone else who knows.\n\nOn this website: https://storage.googleapis.com/openimages/web/factsfigures.html\nIt states there are 944,037 training images for segmentation. \n\nBut on this website: https://github.com/cvdfoundation/open-images-dataset#download-images-with-bounding-boxes-annotations\nThere are 1,743,042 train images for bounding boxes. \n\nHow do I download only the 944,037 segmentation train images instead of the full 1,743,042 images for bounding boxes?\n\nThank you.",
      "votes": null
    },
    {
      "id": "631925",
      "postDate": "09/23/2019 00:43:57",
      "content": "<p>I don't think there's a way to avoid downloading all 1.7M images, however you can train on only the ones with instance segmentation annotations by looking at the unique ImageIds in train-annotations-object-segmentation.csv</p>",
      "rawMarkdown": "I don't think there's a way to avoid downloading all 1.7M images, however you can train on only the ones with instance segmentation annotations by looking at the unique ImageIds in train-annotations-object-segmentation.csv",
      "votes": null
    },
    {
      "id": "634557",
      "postDate": "09/26/2019 12:41:36",
      "content": "<p>As <a href=\"/dfan97\">@dfan97</a> indicates we do not provide a specific subset for segmentation only.</p>\n\n<p>The logic is that:\na) factor 2x makes no big difference for participants.\nb) we expect the overlap of bounding box users and segmentation masks users to be high.\nE.g. for training instance segmentation models one can use the box annotations too.</p>\n\n<p>I hope you managed to sort out the files download and handling.</p>",
      "rawMarkdown": "As @dfan97 indicates we do not provide a specific subset for segmentation only.\n\nThe logic is that:\na) factor 2x makes no big difference for participants.\nb) we expect the overlap of bounding box users and segmentation masks users to be high.\nE.g. for training instance segmentation models one can use the box annotations too.\n\nI hope you managed to sort out the files download and handling.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 631925,
      "author_name": "dfan97",
      "author_url": "",
      "post_date": "09/23/2019 00:43:57",
      "content": "<p>I don't think there's a way to avoid downloading all 1.7M images, however you can train on only the ones with instance segmentation annotations by looking at the unique ImageIds in train-annotations-object-segmentation.csv</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 634557,
      "author_name": "benenson",
      "author_url": "",
      "post_date": "09/26/2019 12:41:36",
      "content": "<p>As <a href=\"/dfan97\">@dfan97</a> indicates we do not provide a specific subset for segmentation only.</p>\n\n<p>The logic is that:\na) factor 2x makes no big difference for participants.\nb) we expect the overlap of bounding box users and segmentation masks users to be high.\nE.g. for training instance segmentation models one can use the box annotations too.</p>\n\n<p>I hope you managed to sort out the files download and handling.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "630125": "benenson @akuznetsa or anyone else who knows.\n\nOn this website: https://storage.googleapis.com/openimages/web/factsfigures.html\nIt states there are 944,037 training images for segmentation. \n\nBut on this website: https://github.com/cvdfoundation/open-images-dataset#download-images-with-bounding-boxes-annotations\nThere are 1,743,042 train images for bounding boxes. \n\nHow do I download only the 944,037 segmentation train images instead of the full 1,743,042 images for bounding boxes?\n\nThank you.",
    "631925": "I don't think there's a way to avoid downloading all 1.7M images, however you can train on only the ones with instance segmentation annotations by looking at the unique ImageIds in train-annotations-object-segmentation.csv",
    "634557": "As @dfan97 indicates we do not provide a specific subset for segmentation only.\n\nThe logic is that:\na) factor 2x makes no big difference for participants.\nb) we expect the overlap of bounding box users and segmentation masks users to be high.\nE.g. for training instance segmentation models one can use the box annotations too.\n\nI hope you managed to sort out the files download and handling."
  },
  "source": "meta"
}