{
  "id": 38017,
  "title": "Pre-trained model weights or CRF post-processing or atrous convolutions/deeplabv3 or PSPNet?",
  "url": "/competitions/carvana-image-masking-challenge/discussion/38017",
  "author_name": "",
  "post_date": "2017-08-13T18:15:04.809137800Z",
  "votes": 1,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Anyone using pre-trained model weights or CRF post-processing or any other model except UNet? Deeplabv3 ? are they worth spending time on for this problem? Any help/advice/tips are appreciated. Thanks.</p>",
  "messages": [
    {
      "id": "213102",
      "postDate": "08/13/2017 18:15:04",
      "content": "<p>Anyone using pre-trained model weights or CRF post-processing or any other model except UNet? Deeplabv3 ? are they worth spending time on for this problem? Any help/advice/tips are appreciated. Thanks.</p>",
      "rawMarkdown": "Anyone using pre-trained model weights or CRF post-processing or any other model except UNet? Deeplabv3 ? are they worth spending time on for this problem? Any help/advice/tips are appreciated. Thanks.",
      "votes": null
    },
    {
      "id": "213130",
      "postDate": "08/13/2017 19:49:33",
      "content": "<p>Has the source to Deeplab v3 been released? All I can find is v2: <a href=\"https://bitbucket.org/aquariusjay/\">https://bitbucket.org/aquariusjay/</a> Also DLv3 is going to be hard to train w/o that multi-card batchnorm implementation.</p>",
      "rawMarkdown": "Has the source to Deeplab v3 been released? All I can find is v2: https://bitbucket.org/aquariusjay/ Also DLv3 is going to be hard to train w/o that multi-card batchnorm implementation.",
      "votes": null
    },
    {
      "id": "213137",
      "postDate": "08/13/2017 20:33:25",
      "content": "<p>Unfortunately it is not. PSPNet is the closest to DeepLabV3 in benchmarks. Which approach are you using?</p>",
      "rawMarkdown": "Unfortunately it is not. PSPNet is the closest to DeepLabV3 in benchmarks. Which approach are you using?",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 213130,
      "author_name": "authman",
      "author_url": "",
      "post_date": "08/13/2017 19:49:33",
      "content": "<p>Has the source to Deeplab v3 been released? All I can find is v2: <a href=\"https://bitbucket.org/aquariusjay/\">https://bitbucket.org/aquariusjay/</a> Also DLv3 is going to be hard to train w/o that multi-card batchnorm implementation.</p>",
      "votes": null,
      "replies": [
        {
          "id": 213137,
          "author_name": "rrqqmm",
          "author_url": "",
          "post_date": "08/13/2017 20:33:25",
          "content": "<p>Unfortunately it is not. PSPNet is the closest to DeepLabV3 in benchmarks. Which approach are you using?</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "213102": "Anyone using pre-trained model weights or CRF post-processing or any other model except UNet? Deeplabv3 ? are they worth spending time on for this problem? Any help/advice/tips are appreciated. Thanks.",
    "213130": "Has the source to Deeplab v3 been released? All I can find is v2: https://bitbucket.org/aquariusjay/ Also DLv3 is going to be hard to train w/o that multi-card batchnorm implementation.",
    "213137": "Unfortunately it is not. PSPNet is the closest to DeepLabV3 in benchmarks. Which approach are you using?"
  },
  "source": "meta"
}