{
  "id": 304338,
  "title": "motion vector for free",
  "url": "/competitions/tensorflow-great-barrier-reef/discussion/304338",
  "author_name": "",
  "post_date": "2022-01-31T20:57:30.316681500Z",
  "votes": 12,
  "comment_count": 1,
  "views": 0,
  "content": "<p>because reef images are highly textured, this is what you get when you have a moving average of the frames<br>\n<img src=\"https://i.ibb.co/9g9h6X6/Selection-003.png\" alt=\"https://i.ibb.co/9g9h6X6/Selection-003.png\"></p>\n<p>the bbox<br>\n<img src=\"https://i.ibb.co/b7cwnQd/Selection-007.png\" alt=\"https://i.ibb.co/b7cwnQd/Selection-007.png\"></p>\n<p>the box are labelled by software … note the smooth trajectory (cx,cy) and smooth change in size(w,h). Anyone familiar with video editing software ccan try motion tracking</p>\n<p>i have a feeling that frame interpolation algo will work well (increase frame rate = more data?) and maybe also joint temporal spatial interpolation?</p>\n<p>i cannot find work that does underwater image restoration using video. maybe this can be for future research. it may be easier to recovery transmission map, etc using video instead of single image</p>",
  "messages": [
    {
      "id": "1670765",
      "postDate": "01/31/2022 20:57:30",
      "content": "<p>because reef images are highly textured, this is what you get when you have a moving average of the frames<br>\n<img src=\"https://i.ibb.co/9g9h6X6/Selection-003.png\" alt=\"https://i.ibb.co/9g9h6X6/Selection-003.png\"></p>\n<p>the bbox<br>\n<img src=\"https://i.ibb.co/b7cwnQd/Selection-007.png\" alt=\"https://i.ibb.co/b7cwnQd/Selection-007.png\"></p>\n<p>the box are labelled by software … note the smooth trajectory (cx,cy) and smooth change in size(w,h). Anyone familiar with video editing software ccan try motion tracking</p>\n<p>i have a feeling that frame interpolation algo will work well (increase frame rate = more data?) and maybe also joint temporal spatial interpolation?</p>\n<p>i cannot find work that does underwater image restoration using video. maybe this can be for future research. it may be easier to recovery transmission map, etc using video instead of single image</p>",
      "rawMarkdown": "because reef images are highly textured, this is what you get when you have a moving average of the frames\n![https://i.ibb.co/9g9h6X6/Selection-003.png](https://i.ibb.co/9g9h6X6/Selection-003.png)\n\n\nthe bbox\n![https://i.ibb.co/b7cwnQd/Selection-007.png](https://i.ibb.co/b7cwnQd/Selection-007.png)\n\nthe box are labelled by software ... note the smooth trajectory (cx,cy) and smooth change in size(w,h). Anyone familiar with video editing software ccan try motion tracking\n\n\ni have a feeling that frame interpolation algo will work well (increase frame rate = more data?) and maybe also joint temporal spatial interpolation?\n\ni cannot find work that does underwater image restoration using video. maybe this can be for future research. it may be easier to recovery transmission map, etc using video instead of single image",
      "votes": null
    },
    {
      "id": "1670812",
      "postDate": "01/31/2022 22:01:28",
      "content": "<p>i mark the center point of the bbox. note that it looks like results of feature tracking (stick to the same part of the object)<br>\nfor a few frames</p>\n<p>some of the corner points behaves smiliarly</p>\n<p><img src=\"https://i.ibb.co/tB1FMMd/Selection-020.png\" alt=\"https://i.ibb.co/tB1FMMd/Selection-020.png\"></p>",
      "rawMarkdown": "i mark the center point of the bbox. note that it looks like results of feature tracking (stick to the same part of the object)\nfor a few frames\n\nsome of the corner points behaves smiliarly\n\n\n![https://i.ibb.co/tB1FMMd/Selection-020.png](https://i.ibb.co/tB1FMMd/Selection-020.png)",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1670812,
      "author_name": "hengck23",
      "author_url": "",
      "post_date": "01/31/2022 22:01:28",
      "content": "<p>i mark the center point of the bbox. note that it looks like results of feature tracking (stick to the same part of the object)<br>\nfor a few frames</p>\n<p>some of the corner points behaves smiliarly</p>\n<p><img src=\"https://i.ibb.co/tB1FMMd/Selection-020.png\" alt=\"https://i.ibb.co/tB1FMMd/Selection-020.png\"></p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1670765": "because reef images are highly textured, this is what you get when you have a moving average of the frames\n![https://i.ibb.co/9g9h6X6/Selection-003.png](https://i.ibb.co/9g9h6X6/Selection-003.png)\n\n\nthe bbox\n![https://i.ibb.co/b7cwnQd/Selection-007.png](https://i.ibb.co/b7cwnQd/Selection-007.png)\n\nthe box are labelled by software ... note the smooth trajectory (cx,cy) and smooth change in size(w,h). Anyone familiar with video editing software ccan try motion tracking\n\n\ni have a feeling that frame interpolation algo will work well (increase frame rate = more data?) and maybe also joint temporal spatial interpolation?\n\ni cannot find work that does underwater image restoration using video. maybe this can be for future research. it may be easier to recovery transmission map, etc using video instead of single image",
    "1670812": "i mark the center point of the bbox. note that it looks like results of feature tracking (stick to the same part of the object)\nfor a few frames\n\nsome of the corner points behaves smiliarly\n\n\n![https://i.ibb.co/tB1FMMd/Selection-020.png](https://i.ibb.co/tB1FMMd/Selection-020.png)"
  },
  "source": "meta"
}