{
  "id": 97850,
  "title": "Can we use image bounding boxes to crop images",
  "url": "/competitions/generative-dog-images/discussion/97850",
  "author_name": "",
  "post_date": "2019-06-29T08:42:07.981252100Z",
  "votes": 17,
  "comment_count": 8,
  "views": 0,
  "content": "<p>I saw the images, the images are not cropped and contains all the glitter around it. That will surely confuse our model.</p>\n\n<p>So can we use the official annotations (bounding boxes) from stanford site to crop the images ?</p>",
  "messages": [
    {
      "id": "564290",
      "postDate": "06/29/2019 08:42:07",
      "content": "<p>I saw the images, the images are not cropped and contains all the glitter around it. That will surely confuse our model.</p>\n\n<p>So can we use the official annotations (bounding boxes) from stanford site to crop the images ?</p>",
      "rawMarkdown": "I saw the images, the images are not cropped and contains all the glitter around it. That will surely confuse our model.\n\nSo can we use the official annotations (bounding boxes) from stanford site to crop the images ?",
      "votes": null
    },
    {
      "id": "564395",
      "postDate": "06/29/2019 11:10:01",
      "content": "<p>And if not, can we train our own as part of our model.</p>",
      "rawMarkdown": "And if not, can we train our own as part of our model.",
      "votes": null
    },
    {
      "id": "564626",
      "postDate": "06/29/2019 18:05:45",
      "content": "<p><a href=\"/wendykan\">@wendykan</a> </p>",
      "rawMarkdown": "wendykan",
      "votes": null
    },
    {
      "id": "565981",
      "postDate": "07/01/2019 16:41:59",
      "content": "<p>Good idea. That is reasonable. I'll add those annotations in. </p>",
      "rawMarkdown": "Good idea. That is reasonable. I'll add those annotations in.",
      "votes": null
    },
    {
      "id": "566095",
      "postDate": "07/01/2019 19:37:09",
      "content": "<p>Thank you. I think using the annotations will improve the quality of our generated dog images.</p>",
      "rawMarkdown": "Thank you. I think using the annotations will improve the quality of our generated dog images.",
      "votes": null
    },
    {
      "id": "566124",
      "postDate": "07/01/2019 20:40:09",
      "content": "<p>Thanks a lot, it will surely make it much easier to train the models. \nEdit: I am thinking about how that would change the metric. \nIf now, all dogs are centered in our training set, so will the test set look like . A discriminator would very easily detect that. </p>",
      "rawMarkdown": "Thanks a lot, it will surely make it much easier to train the models. \nEdit: I am thinking about how that would change the metric. \nIf now, all dogs are centered in our training set, so will the test set look like . A discriminator would very easily detect that.",
      "votes": null
    },
    {
      "id": "566884",
      "postDate": "07/02/2019 18:25:06",
      "content": "<p>Will annotations be available for private images?</p>",
      "rawMarkdown": "Will annotations be available for private images?",
      "votes": null
    },
    {
      "id": "566930",
      "postDate": "07/02/2019 19:37:50",
      "content": "<p>There are no private images for your code to access. This is not a two-stage competition :) </p>",
      "rawMarkdown": "There are no private images for your code to access. This is not a two-stage competition :)",
      "votes": null
    },
    {
      "id": "566964",
      "postDate": "07/02/2019 21:01:35",
      "content": "<p>Thanks. I guess private images include public, in order to decrease a memorization score. </p>",
      "rawMarkdown": "Thanks. I guess private images include public, in order to decrease a memorization score.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 564395,
      "author_name": "robga",
      "author_url": "",
      "post_date": "06/29/2019 11:10:01",
      "content": "<p>And if not, can we train our own as part of our model.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 564626,
      "author_name": "atmyre",
      "author_url": "",
      "post_date": "06/29/2019 18:05:45",
      "content": "<p><a href=\"/wendykan\">@wendykan</a> </p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 565981,
      "author_name": "wendykan",
      "author_url": "",
      "post_date": "07/01/2019 16:41:59",
      "content": "<p>Good idea. That is reasonable. I'll add those annotations in. </p>",
      "votes": null,
      "replies": [
        {
          "id": 566095,
          "author_name": "cdeotte",
          "author_url": "",
          "post_date": "07/01/2019 19:37:09",
          "content": "<p>Thank you. I think using the annotations will improve the quality of our generated dog images.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 566124,
          "author_name": "ilu000",
          "author_url": "",
          "post_date": "07/01/2019 20:40:09",
          "content": "<p>Thanks a lot, it will surely make it much easier to train the models. \nEdit: I am thinking about how that would change the metric. \nIf now, all dogs are centered in our training set, so will the test set look like . A discriminator would very easily detect that. </p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 566884,
          "author_name": "sorokin",
          "author_url": "",
          "post_date": "07/02/2019 18:25:06",
          "content": "<p>Will annotations be available for private images?</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 566930,
          "author_name": "wendykan",
          "author_url": "",
          "post_date": "07/02/2019 19:37:50",
          "content": "<p>There are no private images for your code to access. This is not a two-stage competition :) </p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 566964,
          "author_name": "sorokin",
          "author_url": "",
          "post_date": "07/02/2019 21:01:35",
          "content": "<p>Thanks. I guess private images include public, in order to decrease a memorization score. </p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "564290": "I saw the images, the images are not cropped and contains all the glitter around it. That will surely confuse our model.\n\nSo can we use the official annotations (bounding boxes) from stanford site to crop the images ?",
    "564395": "And if not, can we train our own as part of our model.",
    "564626": "wendykan",
    "565981": "Good idea. That is reasonable. I'll add those annotations in.",
    "566095": "Thank you. I think using the annotations will improve the quality of our generated dog images.",
    "566124": "Thanks a lot, it will surely make it much easier to train the models. \nEdit: I am thinking about how that would change the metric. \nIf now, all dogs are centered in our training set, so will the test set look like . A discriminator would very easily detect that.",
    "566884": "Will annotations be available for private images?",
    "566930": "There are no private images for your code to access. This is not a two-stage competition :)",
    "566964": "Thanks. I guess private images include public, in order to decrease a memorization score."
  },
  "source": "meta"
}