{
  "id": 50680,
  "title": "Improve rpn, rcnn or detection sub network with attention ",
  "url": "/competitions/data-science-bowl-2018/discussion/50680",
  "author_name": "",
  "post_date": "2018-02-28T02:30:28.193601900Z",
  "votes": 10,
  "comment_count": 5,
  "views": 0,
  "content": "<p>You note that the ground truth mask for each object is given. This can be used to generate attention map to improve object detection by suppression background noise feature during training.</p>\n\n<p>An example is here <a href=\"https://www.arxiv-vanity.com/papers/1711.07246/\">https://www.arxiv-vanity.com/papers/1711.07246/</a>, which use semantic segmentation loss as attention loss</p>\n\n<p>This is also a natural way to merge semantic segmentation( which has about 0.98 accuracy here) with object detection </p>",
  "messages": [
    {
      "id": "288691",
      "postDate": "02/28/2018 02:30:28",
      "content": "<p>You note that the ground truth mask for each object is given. This can be used to generate attention map to improve object detection by suppression background noise feature during training.</p>\n\n<p>An example is here <a href=\"https://www.arxiv-vanity.com/papers/1711.07246/\">https://www.arxiv-vanity.com/papers/1711.07246/</a>, which use semantic segmentation loss as attention loss</p>\n\n<p>This is also a natural way to merge semantic segmentation( which has about 0.98 accuracy here) with object detection </p>",
      "rawMarkdown": "You note that the ground truth mask for each object is given. This can be used to generate attention map to improve object detection by suppression background noise feature during training.\n\n\nAn example is here https://www.arxiv-vanity.com/papers/1711.07246/, which use semantic segmentation loss as attention loss\n\n\nThis is also a natural way to merge semantic segmentation( which has about 0.98 accuracy here) with object detection",
      "votes": null
    },
    {
      "id": "288832",
      "postDate": "02/28/2018 09:42:01",
      "content": "<p>can u explain  a little what they are doing?</p>",
      "rawMarkdown": "can u explain  a little what they are doing?",
      "votes": null
    },
    {
      "id": "288843",
      "postDate": "02/28/2018 10:11:15",
      "content": "<p>&gt;&gt;can u explain a little what they are doing?</p>\n\n<p><img src=\"https://storage.googleapis.com/kaggle-forum-message-attachments/288843/8652/attention.png\" alt=\"enter image description here\">\nas attached</p>\n\n<p><img src=\"https://storage.googleapis.com/kaggle-forum-message-attachments/288843/8653/attention2.png\" alt=\"enter image description here\"></p>",
      "rawMarkdown": "&gt;&gt;can u explain a little what they are doing?\n\n\n  ![enter image description here][1]\nas attached\n\n  ![enter image description here][2]\n\n\n  [1]: https://kaggle2.blob.core.windows.net/forum-message-attachments/288843/8652/attention.png\n  [2]: https://kaggle2.blob.core.windows.net/forum-message-attachments/288843/8653/attention2.png",
      "votes": null
    },
    {
      "id": "293294",
      "postDate": "03/09/2018 15:42:14",
      "content": "<p>example of attention map</p>\n\n<p><img src=\"https://storage.googleapis.com/kaggle-forum-message-attachments/293294/8720/mask.png\" alt=\"enter image description here\"></p>\n\n<p><img src=\"https://storage.googleapis.com/kaggle-forum-message-attachments/293294/8721/cut_mask.png\" alt=\"enter image description here\"></p>",
      "rawMarkdown": "example of attention map\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/293294/8720/mask.png\n  [2]: https://kaggle2.blob.core.windows.net/forum-message-attachments/293294/8721/cut_mask.png",
      "votes": null
    },
    {
      "id": "293394",
      "postDate": "03/09/2018 19:20:06",
      "content": "<blockquote>\n  <p>semantic segmentation( which has about 0.98 accuracy here)</p>\n</blockquote>\n\n<p>Which metric do you mean? IoU? Asking because IoU with semantic segmentation is far from even 0.9 for me.</p>",
      "rawMarkdown": "&gt; semantic segmentation( which has about 0.98 accuracy here)\n\nWhich metric do you mean? IoU? Asking because IoU with semantic segmentation is far from even 0.9 for me.",
      "votes": null
    },
    {
      "id": "293510",
      "postDate": "03/10/2018 02:39:03",
      "content": "<p>I am refer to per pixel classifications </p>",
      "rawMarkdown": "I am refer to per pixel classifications",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 288832,
      "author_name": "aviwol",
      "author_url": "",
      "post_date": "02/28/2018 09:42:01",
      "content": "<p>can u explain  a little what they are doing?</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 288843,
      "author_name": "hengck23",
      "author_url": "",
      "post_date": "02/28/2018 10:11:15",
      "content": "<p>&gt;&gt;can u explain a little what they are doing?</p>\n\n<p><img src=\"https://storage.googleapis.com/kaggle-forum-message-attachments/288843/8652/attention.png\" alt=\"enter image description here\">\nas attached</p>\n\n<p><img src=\"https://storage.googleapis.com/kaggle-forum-message-attachments/288843/8653/attention2.png\" alt=\"enter image description here\"></p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 293294,
      "author_name": "hengck23",
      "author_url": "",
      "post_date": "03/09/2018 15:42:14",
      "content": "<p>example of attention map</p>\n\n<p><img src=\"https://storage.googleapis.com/kaggle-forum-message-attachments/293294/8720/mask.png\" alt=\"enter image description here\"></p>\n\n<p><img src=\"https://storage.googleapis.com/kaggle-forum-message-attachments/293294/8721/cut_mask.png\" alt=\"enter image description here\"></p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 293394,
      "author_name": "lopuhin",
      "author_url": "",
      "post_date": "03/09/2018 19:20:06",
      "content": "<blockquote>\n  <p>semantic segmentation( which has about 0.98 accuracy here)</p>\n</blockquote>\n\n<p>Which metric do you mean? IoU? Asking because IoU with semantic segmentation is far from even 0.9 for me.</p>",
      "votes": null,
      "replies": [
        {
          "id": 293510,
          "author_name": "hengck23",
          "author_url": "",
          "post_date": "03/10/2018 02:39:03",
          "content": "<p>I am refer to per pixel classifications </p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "288691": "You note that the ground truth mask for each object is given. This can be used to generate attention map to improve object detection by suppression background noise feature during training.\n\n\nAn example is here https://www.arxiv-vanity.com/papers/1711.07246/, which use semantic segmentation loss as attention loss\n\n\nThis is also a natural way to merge semantic segmentation( which has about 0.98 accuracy here) with object detection",
    "288832": "can u explain  a little what they are doing?",
    "288843": "&gt;&gt;can u explain a little what they are doing?\n\n\n  ![enter image description here][1]\nas attached\n\n  ![enter image description here][2]\n\n\n  [1]: https://kaggle2.blob.core.windows.net/forum-message-attachments/288843/8652/attention.png\n  [2]: https://kaggle2.blob.core.windows.net/forum-message-attachments/288843/8653/attention2.png",
    "293294": "example of attention map\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/293294/8720/mask.png\n  [2]: https://kaggle2.blob.core.windows.net/forum-message-attachments/293294/8721/cut_mask.png",
    "293394": "&gt; semantic segmentation( which has about 0.98 accuracy here)\n\nWhich metric do you mean? IoU? Asking because IoU with semantic segmentation is far from even 0.9 for me.",
    "293510": "I am refer to per pixel classifications"
  },
  "source": "meta"
}