{
  "id": 299356,
  "title": "track id for train dataset!",
  "url": "/competitions/tensorflow-great-barrier-reef/discussion/299356",
  "author_name": "",
  "post_date": "2022-01-07T14:52:05.759092Z",
  "votes": 21,
  "comment_count": 6,
  "views": 0,
  "content": "<p><a href=\"https://www.kaggle.com/hengck23/track-id-for-tensorflow-great-barrier-reef-dataset\" target=\"_blank\">https://www.kaggle.com/hengck23/track-id-for-tensorflow-great-barrier-reef-dataset</a></p>\n<p>i attached the track object id for each of the COTS object in each sequence.<br>\n</p>\n<p>all video sequences are updated</p>\n<pre><code>number of unique COTS object in each video sequence:\n\nvideo_0\n    996   4\n    8399  13\n    35305 2\n    40258 7\n    45015 2\n    45518 3\n    53708 24\n    59337 3\n\nvideo_1\n    8503  66\n    15827 2\n    17665 3\n    18048 3\n    29424 0\n    44160 0\n    60510 4\n    60754 47\n\n\nvideo_2 \n    22643 46\n    26651 1\n    29859 3\n    37114 0\n</code></pre>",
  "messages": [
    {
      "id": "1641568",
      "postDate": "01/07/2022 14:52:05",
      "content": "<p><a href=\"https://www.kaggle.com/hengck23/track-id-for-tensorflow-great-barrier-reef-dataset\" target=\"_blank\">https://www.kaggle.com/hengck23/track-id-for-tensorflow-great-barrier-reef-dataset</a></p>\n<p>i attached the track object id for each of the COTS object in each sequence.<br>\n</p>\n<p>all video sequences are updated</p>\n<pre><code>number of unique COTS object in each video sequence:\n\nvideo_0\n    996   4\n    8399  13\n    35305 2\n    40258 7\n    45015 2\n    45518 3\n    53708 24\n    59337 3\n\nvideo_1\n    8503  66\n    15827 2\n    17665 3\n    18048 3\n    29424 0\n    44160 0\n    60510 4\n    60754 47\n\n\nvideo_2 \n    22643 46\n    26651 1\n    29859 3\n    37114 0\n</code></pre>",
      "rawMarkdown": "https://www.kaggle.com/hengck23/track-id-for-tensorflow-great-barrier-reef-dataset\n\ni attached the track object id for each of the COTS object in each sequence.\n~~Current, I have finished for video\\_0 only. The rest of the videos will be updated soon~~\n\nall video sequences are updated\n\n```\nnumber of unique COTS object in each video sequence:\n\nvideo_0\n\t996   4\n\t8399  13\n\t35305 2\n\t40258 7\n\t45015 2\n\t45518 3\n\t53708 24\n\t59337 3\n\nvideo_1\n\t8503  66\n\t15827 2\n\t17665 3\n\t18048 3\n\t29424 0\n\t44160 0\n\t60510 4\n\t60754 47\n\n\nvideo_2 \n\t22643 46\n\t26651 1\n\t29859 3\n\t37114 0\n\n```",
      "votes": null
    },
    {
      "id": "1641578",
      "postDate": "01/07/2022 14:59:21",
      "content": "<p>Great work! Thanks for sharing!😍</p>",
      "rawMarkdown": "Great work! Thanks for sharing!😍",
      "votes": null
    },
    {
      "id": "1641588",
      "postDate": "01/07/2022 15:04:38",
      "content": "<p>WOW!!! Thank you for sharing.👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍</p>",
      "rawMarkdown": "WOW!!! Thank you for sharing.👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍",
      "votes": null
    },
    {
      "id": "1641858",
      "postDate": "01/07/2022 19:07:56",
      "content": "<p>Already mentioned :) <a href=\"https://www.kaggle.com/c/tensorflow-great-barrier-reef/discussion/296721\" target=\"_blank\">https://www.kaggle.com/c/tensorflow-great-barrier-reef/discussion/296721</a> it gives about 0.01 to score</p>",
      "rawMarkdown": "Already mentioned :) https://www.kaggle.com/c/tensorflow-great-barrier-reef/discussion/296721 it gives about 0.01 to score",
      "votes": null
    },
    {
      "id": "1641862",
      "postDate": "01/07/2022 19:12:20",
      "content": "<p>Check the Łukasz`s link. It can help you to check the current state of knowledge</p>",
      "rawMarkdown": "Check the Łukasz`s link. It can help you to check the current state of knowledge",
      "votes": null
    },
    {
      "id": "1642076",
      "postDate": "01/08/2022 02:53:36",
      "content": "<p>I have done something similar using a tracking library to extract an id for each COT, I'll run my data by your dataset as a quick check. (I didn't release it since my code was not clean lol!)</p>",
      "rawMarkdown": "I have done something similar using a tracking library to extract an id for each COT, I'll run my data by your dataset as a quick check. (I didn't release it since my code was not clean lol!)",
      "votes": null
    },
    {
      "id": "1643391",
      "postDate": "01/09/2022 11:11:53",
      "content": "<p>i note that actually, it is not a tracking problem. we had a moving camera and static target objects.<br>\nthe target objects undergo almost similar image global motion. </p>\n<p>in fact, one can easily do this:</p>\n<pre><code>while 1:\n    image = get_new_frame\n    bbox_det, score_det = do_detection(net1, image)\n\n   global_motion = estimation_motion(net2, image, prev_image)\n   if  global_motion is None: #new seq\n\n      bbox, score = bbox_det, score_det\n      prev_bbox, prev_score = bbox, score \n      pass\n   else:\n      gx, gy = global_motion \n      # depending on your deep model, you can train more accurate motion prediction that is\n      #  pixel location-aware and also output scale change\n\n       bbox_track = prev_bbox + [gx, gy]\n       score_track = prev_score\n\n      bbox, score =  post_process(bbox_det, score_det ,bbox_track, score_track)\n      prev_bbox, prev_score = bbox, score \n\n  prev_image = image\n</code></pre>",
      "rawMarkdown": "i note that actually, it is not a tracking problem. we had a moving camera and static target objects.\nthe target objects undergo almost similar image global motion. \n\nin fact, one can easily do this:\n\n```\n\nwhile 1:\n    image = get_new_frame\n    bbox_det, score_det = do_detection(net1, image)\n\n   global_motion = estimation_motion(net2, image, prev_image)\n   if  global_motion is None: #new seq\n\n      bbox, score = bbox_det, score_det\n      prev_bbox, prev_score = bbox, score \n      pass\n   else:\n      gx, gy = global_motion \n      # depending on your deep model, you can train more accurate motion prediction that is\n      #  pixel location-aware and also output scale change\n   \n       bbox_track = prev_bbox + [gx, gy]\n       score_track = prev_score\n\n      bbox, score =  post_process(bbox_det, score_det ,bbox_track, score_track)\n      prev_bbox, prev_score = bbox, score \n \n  prev_image = image\n```",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1641578,
      "author_name": "shengzheliu",
      "author_url": "",
      "post_date": "01/07/2022 14:59:21",
      "content": "<p>Great work! Thanks for sharing!😍</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1641588,
      "author_name": "shigengtian",
      "author_url": "",
      "post_date": "01/07/2022 15:04:38",
      "content": "<p>WOW!!! Thank you for sharing.👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1641858,
      "author_name": "lukaszborecki",
      "author_url": "",
      "post_date": "01/07/2022 19:07:56",
      "content": "<p>Already mentioned :) <a href=\"https://www.kaggle.com/c/tensorflow-great-barrier-reef/discussion/296721\" target=\"_blank\">https://www.kaggle.com/c/tensorflow-great-barrier-reef/discussion/296721</a> it gives about 0.01 to score</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1641862,
      "author_name": "marcinstasko",
      "author_url": "",
      "post_date": "01/07/2022 19:12:20",
      "content": "<p>Check the Łukasz`s link. It can help you to check the current state of knowledge</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1642076,
      "author_name": "outwrest",
      "author_url": "",
      "post_date": "01/08/2022 02:53:36",
      "content": "<p>I have done something similar using a tracking library to extract an id for each COT, I'll run my data by your dataset as a quick check. (I didn't release it since my code was not clean lol!)</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1643391,
      "author_name": "hengck23",
      "author_url": "",
      "post_date": "01/09/2022 11:11:53",
      "content": "<p>i note that actually, it is not a tracking problem. we had a moving camera and static target objects.<br>\nthe target objects undergo almost similar image global motion. </p>\n<p>in fact, one can easily do this:</p>\n<pre><code>while 1:\n    image = get_new_frame\n    bbox_det, score_det = do_detection(net1, image)\n\n   global_motion = estimation_motion(net2, image, prev_image)\n   if  global_motion is None: #new seq\n\n      bbox, score = bbox_det, score_det\n      prev_bbox, prev_score = bbox, score \n      pass\n   else:\n      gx, gy = global_motion \n      # depending on your deep model, you can train more accurate motion prediction that is\n      #  pixel location-aware and also output scale change\n\n       bbox_track = prev_bbox + [gx, gy]\n       score_track = prev_score\n\n      bbox, score =  post_process(bbox_det, score_det ,bbox_track, score_track)\n      prev_bbox, prev_score = bbox, score \n\n  prev_image = image\n</code></pre>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1641568": "https://www.kaggle.com/hengck23/track-id-for-tensorflow-great-barrier-reef-dataset\n\ni attached the track object id for each of the COTS object in each sequence.\n~~Current, I have finished for video\\_0 only. The rest of the videos will be updated soon~~\n\nall video sequences are updated\n\n```\nnumber of unique COTS object in each video sequence:\n\nvideo_0\n\t996   4\n\t8399  13\n\t35305 2\n\t40258 7\n\t45015 2\n\t45518 3\n\t53708 24\n\t59337 3\n\nvideo_1\n\t8503  66\n\t15827 2\n\t17665 3\n\t18048 3\n\t29424 0\n\t44160 0\n\t60510 4\n\t60754 47\n\n\nvideo_2 \n\t22643 46\n\t26651 1\n\t29859 3\n\t37114 0\n\n```",
    "1641578": "Great work! Thanks for sharing!😍",
    "1641588": "WOW!!! Thank you for sharing.👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍👍",
    "1641858": "Already mentioned :) https://www.kaggle.com/c/tensorflow-great-barrier-reef/discussion/296721 it gives about 0.01 to score",
    "1641862": "Check the Łukasz`s link. It can help you to check the current state of knowledge",
    "1642076": "I have done something similar using a tracking library to extract an id for each COT, I'll run my data by your dataset as a quick check. (I didn't release it since my code was not clean lol!)",
    "1643391": "i note that actually, it is not a tracking problem. we had a moving camera and static target objects.\nthe target objects undergo almost similar image global motion. \n\nin fact, one can easily do this:\n\n```\n\nwhile 1:\n    image = get_new_frame\n    bbox_det, score_det = do_detection(net1, image)\n\n   global_motion = estimation_motion(net2, image, prev_image)\n   if  global_motion is None: #new seq\n\n      bbox, score = bbox_det, score_det\n      prev_bbox, prev_score = bbox, score \n      pass\n   else:\n      gx, gy = global_motion \n      # depending on your deep model, you can train more accurate motion prediction that is\n      #  pixel location-aware and also output scale change\n   \n       bbox_track = prev_bbox + [gx, gy]\n       score_track = prev_score\n\n      bbox, score =  post_process(bbox_det, score_det ,bbox_track, score_track)\n      prev_bbox, prev_score = bbox, score \n \n  prev_image = image\n```"
  },
  "source": "meta"
}