{
  "id": 39948,
  "title": "smaller car may gives higher error?",
  "url": "/competitions/carvana-image-masking-challenge/discussion/39948",
  "author_name": "",
  "post_date": "2017-09-24T15:51:44.293193100Z",
  "votes": 9,
  "comment_count": 4,
  "views": 0,
  "content": "<p>I can use a low resolution unet512 to predict the bounding box of the test images. Then I sort the results according to the width of the box. As expected, those 'outlier' with smaller box size has higher error.</p>\n\n<p>My plan is to fine-tune existing model for these easy detectable difficult case.</p>\n\n<p><img src=\"https://storage.googleapis.com/kaggle-forum-message-attachments/223985/7371/image1.png\" alt=\"enter image description here\" title=\"\"></p>\n\n<p><img src=\"https://storage.googleapis.com/kaggle-forum-message-attachments/223985/7372/image2.png\" alt=\"enter image description here\" title=\"\"></p>",
  "messages": [
    {
      "id": "223985",
      "postDate": "09/24/2017 15:51:44",
      "content": "<p>I can use a low resolution unet512 to predict the bounding box of the test images. Then I sort the results according to the width of the box. As expected, those 'outlier' with smaller box size has higher error.</p>\n\n<p>My plan is to fine-tune existing model for these easy detectable difficult case.</p>\n\n<p><img src=\"https://storage.googleapis.com/kaggle-forum-message-attachments/223985/7371/image1.png\" alt=\"enter image description here\" title=\"\"></p>\n\n<p><img src=\"https://storage.googleapis.com/kaggle-forum-message-attachments/223985/7372/image2.png\" alt=\"enter image description here\" title=\"\"></p>",
      "rawMarkdown": "I can use a low resolution unet512 to predict the bounding box of the test images. Then I sort the results according to the width of the box. As expected, those 'outlier' with smaller box size has higher error.\n\nMy plan is to fine-tune existing model for these easy detectable difficult case.\n\n ![enter image description here][1]\n\n ![enter image description here][2]\n\n\n  [1]: https://kaggle2.blob.core.windows.net/forum-message-attachments/223985/7371/image1.png\n  [2]: https://kaggle2.blob.core.windows.net/forum-message-attachments/223985/7372/image2.png",
      "votes": null
    },
    {
      "id": "224706",
      "postDate": "09/27/2017 11:09:12",
      "content": "<p>correcting small car errors:</p>\n\n<p><img src=\"https://storage.googleapis.com/kaggle-forum-message-attachments/224706/7396/Slide1.jpg\" alt=\"enter image description here\" title=\"\">\n  <img src=\"https://storage.googleapis.com/kaggle-forum-message-attachments/224706/7397/Slide2.jpg\" alt=\"enter image description here\" title=\"\">\n  <img src=\"https://storage.googleapis.com/kaggle-forum-message-attachments/224706/7395/Slide3.jpg\" alt=\"enter image description here\" title=\"\"></p>",
      "rawMarkdown": "correcting small car errors:\n\n  ![enter image description here][1]\n  ![enter image description here][2]\n  ![enter image description here][3]\n\n\n  [1]: https://kaggle2.blob.core.windows.net/forum-message-attachments/224706/7396/Slide1.jpg\n  [2]: https://kaggle2.blob.core.windows.net/forum-message-attachments/224706/7397/Slide2.jpg\n  [3]: https://kaggle2.blob.core.windows.net/forum-message-attachments/224706/7395/Slide3.jpg",
      "votes": null
    },
    {
      "id": "224711",
      "postDate": "09/27/2017 12:04:04",
      "content": "<p>correcting express van problem:</p>\n\n<p>It is easy to find-tune a current model for special car type if you can generate correct train data. This is one way in which meta-data can be used.</p>\n\n<p><img src=\"https://storage.googleapis.com/kaggle-forum-message-attachments/224711/7398/Slide1.jpg\" alt=\"enter image description here\" title=\"\"></p>\n\n<p><img src=\"https://storage.googleapis.com/kaggle-forum-message-attachments/224711/7400/Slide2.jpg\" alt=\"enter image description here\" title=\"\"></p>\n\n<p><img src=\"https://storage.googleapis.com/kaggle-forum-message-attachments/224711/7399/Slide3.jpg\" alt=\"enter image description here\" title=\"\"></p>",
      "rawMarkdown": "correcting express van problem:\n\nIt is easy to find-tune a current model for special car type if you can generate correct train data. This is one way in which meta-data can be used.\n\n\n   ![enter image description here][1]\n\n\n   ![enter image description here][2]\n\n\n   ![enter image description here][3]\n\n\n  [1]: https://kaggle2.blob.core.windows.net/forum-message-attachments/224711/7398/Slide1.jpg\n  [2]: https://kaggle2.blob.core.windows.net/forum-message-attachments/224711/7400/Slide2.jpg\n  [3]: https://kaggle2.blob.core.windows.net/forum-message-attachments/224711/7399/Slide3.jpg",
      "votes": null
    },
    {
      "id": "224786",
      "postDate": "09/27/2017 16:57:40",
      "content": "<p>some augmentation that works:</p>\n\n<p><img src=\"https://storage.googleapis.com/kaggle-forum-message-attachments/224786/7405/Slide5.jpg\" alt=\"enter image description here\" title=\"\"></p>",
      "rawMarkdown": "some augmentation that works:\n\n  ![enter image description here][1]\n\n\n  [1]: https://kaggle2.blob.core.windows.net/forum-message-attachments/224786/7405/Slide5.jpg",
      "votes": null
    },
    {
      "id": "224817",
      "postDate": "09/27/2017 18:19:04",
      "content": "<p>Not bad! Really creative idea. How much does it add on LB or CV?</p>",
      "rawMarkdown": "Not bad! Really creative idea. How much does it add on LB or CV?",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 224706,
      "author_name": "hengck23",
      "author_url": "",
      "post_date": "09/27/2017 11:09:12",
      "content": "<p>correcting small car errors:</p>\n\n<p><img src=\"https://storage.googleapis.com/kaggle-forum-message-attachments/224706/7396/Slide1.jpg\" alt=\"enter image description here\" title=\"\">\n  <img src=\"https://storage.googleapis.com/kaggle-forum-message-attachments/224706/7397/Slide2.jpg\" alt=\"enter image description here\" title=\"\">\n  <img src=\"https://storage.googleapis.com/kaggle-forum-message-attachments/224706/7395/Slide3.jpg\" alt=\"enter image description here\" title=\"\"></p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 224711,
      "author_name": "hengck23",
      "author_url": "",
      "post_date": "09/27/2017 12:04:04",
      "content": "<p>correcting express van problem:</p>\n\n<p>It is easy to find-tune a current model for special car type if you can generate correct train data. This is one way in which meta-data can be used.</p>\n\n<p><img src=\"https://storage.googleapis.com/kaggle-forum-message-attachments/224711/7398/Slide1.jpg\" alt=\"enter image description here\" title=\"\"></p>\n\n<p><img src=\"https://storage.googleapis.com/kaggle-forum-message-attachments/224711/7400/Slide2.jpg\" alt=\"enter image description here\" title=\"\"></p>\n\n<p><img src=\"https://storage.googleapis.com/kaggle-forum-message-attachments/224711/7399/Slide3.jpg\" alt=\"enter image description here\" title=\"\"></p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 224786,
      "author_name": "hengck23",
      "author_url": "",
      "post_date": "09/27/2017 16:57:40",
      "content": "<p>some augmentation that works:</p>\n\n<p><img src=\"https://storage.googleapis.com/kaggle-forum-message-attachments/224786/7405/Slide5.jpg\" alt=\"enter image description here\" title=\"\"></p>",
      "votes": null,
      "replies": [
        {
          "id": 224817,
          "author_name": "ceperaang",
          "author_url": "",
          "post_date": "09/27/2017 18:19:04",
          "content": "<p>Not bad! Really creative idea. How much does it add on LB or CV?</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "223985": "I can use a low resolution unet512 to predict the bounding box of the test images. Then I sort the results according to the width of the box. As expected, those 'outlier' with smaller box size has higher error.\n\nMy plan is to fine-tune existing model for these easy detectable difficult case.\n\n ![enter image description here][1]\n\n ![enter image description here][2]\n\n\n  [1]: https://kaggle2.blob.core.windows.net/forum-message-attachments/223985/7371/image1.png\n  [2]: https://kaggle2.blob.core.windows.net/forum-message-attachments/223985/7372/image2.png",
    "224706": "correcting small car errors:\n\n  ![enter image description here][1]\n  ![enter image description here][2]\n  ![enter image description here][3]\n\n\n  [1]: https://kaggle2.blob.core.windows.net/forum-message-attachments/224706/7396/Slide1.jpg\n  [2]: https://kaggle2.blob.core.windows.net/forum-message-attachments/224706/7397/Slide2.jpg\n  [3]: https://kaggle2.blob.core.windows.net/forum-message-attachments/224706/7395/Slide3.jpg",
    "224711": "correcting express van problem:\n\nIt is easy to find-tune a current model for special car type if you can generate correct train data. This is one way in which meta-data can be used.\n\n\n   ![enter image description here][1]\n\n\n   ![enter image description here][2]\n\n\n   ![enter image description here][3]\n\n\n  [1]: https://kaggle2.blob.core.windows.net/forum-message-attachments/224711/7398/Slide1.jpg\n  [2]: https://kaggle2.blob.core.windows.net/forum-message-attachments/224711/7400/Slide2.jpg\n  [3]: https://kaggle2.blob.core.windows.net/forum-message-attachments/224711/7399/Slide3.jpg",
    "224786": "some augmentation that works:\n\n  ![enter image description here][1]\n\n\n  [1]: https://kaggle2.blob.core.windows.net/forum-message-attachments/224786/7405/Slide5.jpg",
    "224817": "Not bad! Really creative idea. How much does it add on LB or CV?"
  },
  "source": "meta"
}