{
  "id": 62776,
  "title": "Rotated Object Detection Boxes for Segmentation?",
  "url": "/competitions/airbus-ship-detection/discussion/62776",
  "author_name": "",
  "post_date": "2018-08-07T02:39:00.475610300Z",
  "votes": 4,
  "comment_count": 5,
  "views": 0,
  "content": "<p>As far as I have seen, the segmentation masks are rectangles in the images. That made me think that you could use object detectors like YOLO, but also add a \"rotation\" output into the mix. That way you could get a box from the images that is rotated so it fits the ship better.\nFor those of you who are more experienced in computer vision or have invested more time into this competition, do you think this is a good idea?</p>",
  "messages": [
    {
      "id": "367072",
      "postDate": "08/07/2018 02:39:00",
      "content": "<p>As far as I have seen, the segmentation masks are rectangles in the images. That made me think that you could use object detectors like YOLO, but also add a \"rotation\" output into the mix. That way you could get a box from the images that is rotated so it fits the ship better.\nFor those of you who are more experienced in computer vision or have invested more time into this competition, do you think this is a good idea?</p>",
      "rawMarkdown": "As far as I have seen, the segmentation masks are rectangles in the images. That made me think that you could use object detectors like YOLO, but also add a \"rotation\" output into the mix. That way you could get a box from the images that is rotated so it fits the ship better.\nFor those of you who are more experienced in computer vision or have invested more time into this competition, do you think this is a good idea?",
      "votes": null
    },
    {
      "id": "367506",
      "postDate": "08/07/2018 23:27:57",
      "content": "<p>I think someone has done that: \"Learning a Rotation Invariant Detector with Rotatable Bounding Box\" <a href=\"https://arxiv.org/abs/1711.09405\">https://arxiv.org/abs/1711.09405</a></p>",
      "rawMarkdown": "I think someone has done that: \"Learning a Rotation Invariant Detector with Rotatable Bounding Box\" https://arxiv.org/abs/1711.09405",
      "votes": null
    },
    {
      "id": "367519",
      "postDate": "08/08/2018 00:05:52",
      "content": "<p>I looked at it. It seems very promising! They used it for satellite imagery too, but they didn't seem to use YOLO, which I will try to use it with because of speed, which is also another aspect of this competition. They said it outperformed rotation non-invariant object detectors. I think I will give this method a try, first with Faster-RCNN, which they used and then try a modified YOLO detector.</p>",
      "rawMarkdown": "I looked at it. It seems very promising! They used it for satellite imagery too, but they didn't seem to use YOLO, which I will try to use it with because of speed, which is also another aspect of this competition. They said it outperformed rotation non-invariant object detectors. I think I will give this method a try, first with Faster-RCNN, which they used and then try a modified YOLO detector.",
      "votes": null
    },
    {
      "id": "452117",
      "postDate": "01/08/2019 08:15:25",
      "content": "<p>There also method like DRBox the to this, does there anyone have further thought on this?</p>",
      "rawMarkdown": "There also method like DRBox the to this, does there anyone have further thought on this?",
      "votes": null
    },
    {
      "id": "455273",
      "postDate": "01/13/2019 13:01:57",
      "content": "<p>Hi Jintian,</p>\n\n<p>My solution is motivated by DRBox. Please check that !<br>\n<a href=\"https://www.kaggle.com/c/airbus-ship-detection/discussion/71875\">https://www.kaggle.com/c/airbus-ship-detection/discussion/71875</a></p>",
      "rawMarkdown": "Hi Jintian,\n\nMy solution is motivated by DRBox. Please check that !<br>\nhttps://www.kaggle.com/c/airbus-ship-detection/discussion/71875",
      "votes": null
    },
    {
      "id": "1021908",
      "postDate": "09/22/2020 07:48:21",
      "content": "<p>Did anyone found any solution to this?</p>",
      "rawMarkdown": "Did anyone found any solution to this?",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1021908,
      "author_name": "ahsanmukhtar",
      "author_url": "",
      "post_date": "09/22/2020 07:48:21",
      "content": "<p>Did anyone found any solution to this?</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 367506,
      "author_name": "waskita",
      "author_url": "",
      "post_date": "08/07/2018 23:27:57",
      "content": "<p>I think someone has done that: \"Learning a Rotation Invariant Detector with Rotatable Bounding Box\" <a href=\"https://arxiv.org/abs/1711.09405\">https://arxiv.org/abs/1711.09405</a></p>",
      "votes": null,
      "replies": [
        {
          "id": 367519,
          "author_name": "arpandhatt",
          "author_url": "",
          "post_date": "08/08/2018 00:05:52",
          "content": "<p>I looked at it. It seems very promising! They used it for satellite imagery too, but they didn't seem to use YOLO, which I will try to use it with because of speed, which is also another aspect of this competition. They said it outperformed rotation non-invariant object detectors. I think I will give this method a try, first with Faster-RCNN, which they used and then try a modified YOLO detector.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 452117,
      "author_name": "jintian123",
      "author_url": "",
      "post_date": "01/08/2019 08:15:25",
      "content": "<p>There also method like DRBox the to this, does there anyone have further thought on this?</p>",
      "votes": null,
      "replies": [
        {
          "id": 455273,
          "author_name": "toshik",
          "author_url": "",
          "post_date": "01/13/2019 13:01:57",
          "content": "<p>Hi Jintian,</p>\n\n<p>My solution is motivated by DRBox. Please check that !<br>\n<a href=\"https://www.kaggle.com/c/airbus-ship-detection/discussion/71875\">https://www.kaggle.com/c/airbus-ship-detection/discussion/71875</a></p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "367072": "As far as I have seen, the segmentation masks are rectangles in the images. That made me think that you could use object detectors like YOLO, but also add a \"rotation\" output into the mix. That way you could get a box from the images that is rotated so it fits the ship better.\nFor those of you who are more experienced in computer vision or have invested more time into this competition, do you think this is a good idea?",
    "367506": "I think someone has done that: \"Learning a Rotation Invariant Detector with Rotatable Bounding Box\" https://arxiv.org/abs/1711.09405",
    "367519": "I looked at it. It seems very promising! They used it for satellite imagery too, but they didn't seem to use YOLO, which I will try to use it with because of speed, which is also another aspect of this competition. They said it outperformed rotation non-invariant object detectors. I think I will give this method a try, first with Faster-RCNN, which they used and then try a modified YOLO detector.",
    "452117": "There also method like DRBox the to this, does there anyone have further thought on this?",
    "455273": "Hi Jintian,\n\nMy solution is motivated by DRBox. Please check that !<br>\nhttps://www.kaggle.com/c/airbus-ship-detection/discussion/71875",
    "1021908": "Did anyone found any solution to this?"
  },
  "source": "meta"
}