{
  "id": 95130,
  "title": "My approach (almost)",
  "url": "/competitions/imaterialist-fashion-2019-FGVC6/discussion/95130",
  "author_name": "",
  "post_date": "2019-06-10T07:59:58.161854Z",
  "votes": 2,
  "comment_count": 5,
  "views": 0,
  "content": "<p>I have shared my approach to this problem here. \nI invested quite a lot of time in this competition but was unable to get a good score.\nThe approach can give you 0.17+</p>\n\n<p><a href=\"https://www.kaggle.com/abhishek/mask-rcnn-using-torchvision-0-17\">https://www.kaggle.com/abhishek/mask-rcnn-using-torchvision-0-17</a></p>",
  "messages": [
    {
      "id": "549027",
      "postDate": "06/10/2019 07:59:58",
      "content": "<p>I have shared my approach to this problem here. \nI invested quite a lot of time in this competition but was unable to get a good score.\nThe approach can give you 0.17+</p>\n\n<p><a href=\"https://www.kaggle.com/abhishek/mask-rcnn-using-torchvision-0-17\">https://www.kaggle.com/abhishek/mask-rcnn-using-torchvision-0-17</a></p>",
      "rawMarkdown": "I have shared my approach to this problem here. \nI invested quite a lot of time in this competition but was unable to get a good score.\nThe approach can give you 0.17+\n\nhttps://www.kaggle.com/abhishek/mask-rcnn-using-torchvision-0-17",
      "votes": null
    },
    {
      "id": "549118",
      "postDate": "06/10/2019 10:07:15",
      "content": "<p>Thanks for sharing! Have you tried using images of arbitrary size instead of fixed (H,W) during training/inference? I wonder if preserving aspect ratio can increase model accuracy.</p>",
      "rawMarkdown": "Thanks for sharing! Have you tried using images of arbitrary size instead of fixed (H,W) during training/inference? I wonder if preserving aspect ratio can increase model accuracy.",
      "votes": null
    },
    {
      "id": "549195",
      "postDate": "06/10/2019 12:02:36",
      "content": "<p>No. I didnt try that and unfortunately, no time to try it now :)</p>",
      "rawMarkdown": "No. I didnt try that and unfortunately, no time to try it now :)",
      "votes": null
    },
    {
      "id": "549201",
      "postDate": "06/10/2019 12:06:58",
      "content": "<p>Why do you share just 12 hours before the deadline? Why not a week ago or after the deadline?</p>",
      "rawMarkdown": "Why do you share just 12 hours before the deadline? Why not a week ago or after the deadline?",
      "votes": null
    },
    {
      "id": "549204",
      "postDate": "06/10/2019 12:09:11",
      "content": "<p>doesnt matter. this kernel is not capable of producing something good in just 12 hours.</p>",
      "rawMarkdown": "doesnt matter. this kernel is not capable of producing something good in just 12 hours.",
      "votes": null
    },
    {
      "id": "549956",
      "postDate": "06/11/2019 07:08:40",
      "content": "<p>Could you explain how multiple GPUs are enabled in the code? I'm new to pytorch and trying to learn something from your great code. Thanks!</p>",
      "rawMarkdown": "Could you explain how multiple GPUs are enabled in the code? I'm new to pytorch and trying to learn something from your great code. Thanks!",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 549118,
      "author_name": "devvindan",
      "author_url": "",
      "post_date": "06/10/2019 10:07:15",
      "content": "<p>Thanks for sharing! Have you tried using images of arbitrary size instead of fixed (H,W) during training/inference? I wonder if preserving aspect ratio can increase model accuracy.</p>",
      "votes": null,
      "replies": [
        {
          "id": 549195,
          "author_name": "abhishek",
          "author_url": "",
          "post_date": "06/10/2019 12:02:36",
          "content": "<p>No. I didnt try that and unfortunately, no time to try it now :)</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 549201,
      "author_name": "aerdem4",
      "author_url": "",
      "post_date": "06/10/2019 12:06:58",
      "content": "<p>Why do you share just 12 hours before the deadline? Why not a week ago or after the deadline?</p>",
      "votes": null,
      "replies": [
        {
          "id": 549204,
          "author_name": "abhishek",
          "author_url": "",
          "post_date": "06/10/2019 12:09:11",
          "content": "<p>doesnt matter. this kernel is not capable of producing something good in just 12 hours.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 549956,
      "author_name": "lucaskg",
      "author_url": "",
      "post_date": "06/11/2019 07:08:40",
      "content": "<p>Could you explain how multiple GPUs are enabled in the code? I'm new to pytorch and trying to learn something from your great code. Thanks!</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "549027": "I have shared my approach to this problem here. \nI invested quite a lot of time in this competition but was unable to get a good score.\nThe approach can give you 0.17+\n\nhttps://www.kaggle.com/abhishek/mask-rcnn-using-torchvision-0-17",
    "549118": "Thanks for sharing! Have you tried using images of arbitrary size instead of fixed (H,W) during training/inference? I wonder if preserving aspect ratio can increase model accuracy.",
    "549195": "No. I didnt try that and unfortunately, no time to try it now :)",
    "549201": "Why do you share just 12 hours before the deadline? Why not a week ago or after the deadline?",
    "549204": "doesnt matter. this kernel is not capable of producing something good in just 12 hours.",
    "549956": "Could you explain how multiple GPUs are enabled in the code? I'm new to pytorch and trying to learn something from your great code. Thanks!"
  },
  "source": "meta"
}