{
  "id": 91767,
  "title": "Convert Data format to COCO?",
  "url": "/competitions/imaterialist-fashion-2019-FGVC6/discussion/91767",
  "author_name": "",
  "post_date": "2019-05-08T20:28:18.503204700Z",
  "votes": null,
  "comment_count": 3,
  "views": 0,
  "content": "<p>We are considering converting the image style to COCO style to train Mask R-cnn for the task, but we noticed that there are not bbox information provided. So in this case, should we generate the bbox ourselves from the mask information? Any ideas? Thanks!</p>",
  "messages": [
    {
      "id": "528881",
      "postDate": "05/08/2019 20:28:18",
      "content": "<p>We are considering converting the image style to COCO style to train Mask R-cnn for the task, but we noticed that there are not bbox information provided. So in this case, should we generate the bbox ourselves from the mask information? Any ideas? Thanks!</p>",
      "rawMarkdown": "We are considering converting the image style to COCO style to train Mask R-cnn for the task, but we noticed that there are not bbox information provided. So in this case, should we generate the bbox ourselves from the mask information? Any ideas? Thanks!",
      "votes": null
    },
    {
      "id": "529425",
      "postDate": "05/09/2019 21:37:20",
      "content": "<p>yep</p>",
      "rawMarkdown": "yep",
      "votes": null
    },
    {
      "id": "532901",
      "postDate": "05/18/2019 02:27:41",
      "content": "<p>Hi Joyce, if you are still looking for solutions, you can check out pycocotools below: \n<a href=\"https://github.com/cocodataset/cocoapi/tree/master/PythonAPI/pycocotools\">https://github.com/cocodataset/cocoapi/tree/master/PythonAPI/pycocotools</a></p>",
      "rawMarkdown": "Hi Joyce, if you are still looking for solutions, you can check out pycocotools below: \nhttps://github.com/cocodataset/cocoapi/tree/master/PythonAPI/pycocotools",
      "votes": null
    },
    {
      "id": "534553",
      "postDate": "05/21/2019 13:00:49",
      "content": "<p>Hi maybe some interesting infos in this kernel : <a href=\"https://www.kaggle.com/voglinio/from-masks-to-bounding-boxes\">https://www.kaggle.com/voglinio/from-masks-to-bounding-boxes</a> :) </p>",
      "rawMarkdown": "Hi maybe some interesting infos in this kernel : https://www.kaggle.com/voglinio/from-masks-to-bounding-boxes :)",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 529425,
      "author_name": "alexanderliao",
      "author_url": "",
      "post_date": "05/09/2019 21:37:20",
      "content": "<p>yep</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 532901,
      "author_name": "makeitworkjml",
      "author_url": "",
      "post_date": "05/18/2019 02:27:41",
      "content": "<p>Hi Joyce, if you are still looking for solutions, you can check out pycocotools below: \n<a href=\"https://github.com/cocodataset/cocoapi/tree/master/PythonAPI/pycocotools\">https://github.com/cocodataset/cocoapi/tree/master/PythonAPI/pycocotools</a></p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 534553,
      "author_name": "pasc1996",
      "author_url": "",
      "post_date": "05/21/2019 13:00:49",
      "content": "<p>Hi maybe some interesting infos in this kernel : <a href=\"https://www.kaggle.com/voglinio/from-masks-to-bounding-boxes\">https://www.kaggle.com/voglinio/from-masks-to-bounding-boxes</a> :) </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "528881": "We are considering converting the image style to COCO style to train Mask R-cnn for the task, but we noticed that there are not bbox information provided. So in this case, should we generate the bbox ourselves from the mask information? Any ideas? Thanks!",
    "529425": "yep",
    "532901": "Hi Joyce, if you are still looking for solutions, you can check out pycocotools below: \nhttps://github.com/cocodataset/cocoapi/tree/master/PythonAPI/pycocotools",
    "534553": "Hi maybe some interesting infos in this kernel : https://www.kaggle.com/voglinio/from-masks-to-bounding-boxes :)"
  },
  "source": "meta"
}