{
  "id": 90710,
  "title": "Question about submission",
  "url": "/competitions/imaterialist-fashion-2019-FGVC6/discussion/90710",
  "author_name": "",
  "post_date": "2019-04-26T07:35:21.427105100Z",
  "votes": null,
  "comment_count": 4,
  "views": 0,
  "content": "<p>Just for clarification.</p>\n\n<p>Our goal is to predict the segmentation of both the whole body objects, and the different small parts of the objects. So for each image, we may have multiple lines in the submission file.\nAnd the attributes are not actually used for evaluation, so we can actually ignore it if we do not want to use them. Or, do we need to predict all the attributes correctly to earn the point of that object?</p>\n\n<p>Am I understanding it correctly?</p>",
  "messages": [
    {
      "id": "523412",
      "postDate": "04/26/2019 07:35:21",
      "content": "<p>Just for clarification.</p>\n\n<p>Our goal is to predict the segmentation of both the whole body objects, and the different small parts of the objects. So for each image, we may have multiple lines in the submission file.\nAnd the attributes are not actually used for evaluation, so we can actually ignore it if we do not want to use them. Or, do we need to predict all the attributes correctly to earn the point of that object?</p>\n\n<p>Am I understanding it correctly?</p>",
      "rawMarkdown": "Just for clarification.\n\nOur goal is to predict the segmentation of both the whole body objects, and the different small parts of the objects. So for each image, we may have multiple lines in the submission file.\nAnd the attributes are not actually used for evaluation, so we can actually ignore it if we do not want to use them. Or, do we need to predict all the attributes correctly to earn the point of that object?\n\nAm I understanding it correctly?",
      "votes": null
    },
    {
      "id": "523503",
      "postDate": "04/26/2019 12:28:13",
      "content": "<p><strong>\"So for each image, we may have multiple lines in the submission file.\"</strong>: this is correct. There may be multiple masks in one image.</p>\n\n<p><strong>\"the attributes are not actually used for evaluation\"</strong>:  attributes <strong>are</strong> used for evaluation (if there is any). At <a href=\"https://www.kaggle.com/c/imaterialist-fashion-2019-FGVC6/overview/evaluation\">the evaluation page</a>, \"It is important to distinguish both fine-grained classes of the object as well as object instances. If the mask is correct but the class associated with the mask is wrong then it is considered a false positive.\".</p>\n\n<p>Hope it helps!</p>",
      "rawMarkdown": "**\"So for each image, we may have multiple lines in the submission file.\"**: this is correct. There may be multiple masks in one image.\n\n**\"the attributes are not actually used for evaluation\"**:  attributes **are** used for evaluation (if there is any). At [the evaluation page](https://www.kaggle.com/c/imaterialist-fashion-2019-FGVC6/overview/evaluation), \"It is important to distinguish both fine-grained classes of the object as well as object instances. If the mask is correct but the class associated with the mask is wrong then it is considered a false positive.\".\n\nHope it helps!",
      "votes": null
    },
    {
      "id": "523559",
      "postDate": "04/26/2019 14:13:07",
      "content": "<p>Thanks for the reply.\nSo we will have to predict every single attribute of an object correctly in order to get the point?</p>\n\n<p>That seems a little bit challenging for objects with many attributes.</p>",
      "rawMarkdown": "Thanks for the reply.\nSo we will have to predict every single attribute of an object correctly in order to get the point?\n\nThat seems a little bit challenging for objects with many attributes.",
      "votes": null
    },
    {
      "id": "523607",
      "postDate": "04/26/2019 15:55:06",
      "content": "<p>yep, that is the goal and what's fun about this competition :)</p>",
      "rawMarkdown": "yep, that is the goal and what's fun about this competition :)",
      "votes": null
    },
    {
      "id": "525622",
      "postDate": "05/01/2019 12:15:24",
      "content": "<p>So for images that have multiple segmentation masks how are they represented in the EncodedPixels.</p>",
      "rawMarkdown": "So for images that have multiple segmentation masks how are they represented in the EncodedPixels.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 523503,
      "author_name": "makeitworkjml",
      "author_url": "",
      "post_date": "04/26/2019 12:28:13",
      "content": "<p><strong>\"So for each image, we may have multiple lines in the submission file.\"</strong>: this is correct. There may be multiple masks in one image.</p>\n\n<p><strong>\"the attributes are not actually used for evaluation\"</strong>:  attributes <strong>are</strong> used for evaluation (if there is any). At <a href=\"https://www.kaggle.com/c/imaterialist-fashion-2019-FGVC6/overview/evaluation\">the evaluation page</a>, \"It is important to distinguish both fine-grained classes of the object as well as object instances. If the mask is correct but the class associated with the mask is wrong then it is considered a false positive.\".</p>\n\n<p>Hope it helps!</p>",
      "votes": null,
      "replies": [
        {
          "id": 523559,
          "author_name": "lzhbrian",
          "author_url": "",
          "post_date": "04/26/2019 14:13:07",
          "content": "<p>Thanks for the reply.\nSo we will have to predict every single attribute of an object correctly in order to get the point?</p>\n\n<p>That seems a little bit challenging for objects with many attributes.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 523607,
          "author_name": "makeitworkjml",
          "author_url": "",
          "post_date": "04/26/2019 15:55:06",
          "content": "<p>yep, that is the goal and what's fun about this competition :)</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 525622,
          "author_name": "kushaj",
          "author_url": "",
          "post_date": "05/01/2019 12:15:24",
          "content": "<p>So for images that have multiple segmentation masks how are they represented in the EncodedPixels.</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "523412": "Just for clarification.\n\nOur goal is to predict the segmentation of both the whole body objects, and the different small parts of the objects. So for each image, we may have multiple lines in the submission file.\nAnd the attributes are not actually used for evaluation, so we can actually ignore it if we do not want to use them. Or, do we need to predict all the attributes correctly to earn the point of that object?\n\nAm I understanding it correctly?",
    "523503": "**\"So for each image, we may have multiple lines in the submission file.\"**: this is correct. There may be multiple masks in one image.\n\n**\"the attributes are not actually used for evaluation\"**:  attributes **are** used for evaluation (if there is any). At [the evaluation page](https://www.kaggle.com/c/imaterialist-fashion-2019-FGVC6/overview/evaluation), \"It is important to distinguish both fine-grained classes of the object as well as object instances. If the mask is correct but the class associated with the mask is wrong then it is considered a false positive.\".\n\nHope it helps!",
    "523559": "Thanks for the reply.\nSo we will have to predict every single attribute of an object correctly in order to get the point?\n\nThat seems a little bit challenging for objects with many attributes.",
    "523607": "yep, that is the goal and what's fun about this competition :)",
    "525622": "So for images that have multiple segmentation masks how are they represented in the EncodedPixels."
  },
  "source": "meta"
}