{
  "id": 305663,
  "title": "Has anyone else tried a multi-object tracker like Bytetrack or deepsort?",
  "url": "/competitions/tensorflow-great-barrier-reef/discussion/305663",
  "author_name": "Sugar0",
  "post_date": "2022-02-06T12:02:12.054000",
  "votes": 6,
  "comment_count": 9,
  "views": 0,
  "content": "<p>I tried to use <a href=\"https://github.com/ifzhang/ByteTrack\" target=\"_blank\">Bytetracker</a>, which is a state-of-the-art tracker in <a href=\"https://paperswithcode.com/sota/multi-object-tracking-on-mot20-1?p=bytetrack-multi-object-tracking-by-1\" target=\"_blank\">the field of multi-object tracking</a> because trackers seem to be effective in detecting COTS.</p>\n<p>However, the results were not good.</p>\n<p>Has anyone else tried a multi-object tracker like Bytetrack or deepsort?</p>",
  "messages": [
    {
      "id": 1678271,
      "postDate": "2022-02-06T12:02:12.053Z",
      "content": "<p>I tried to use <a href=\"https://github.com/ifzhang/ByteTrack\" target=\"_blank\">Bytetracker</a>, which is a state-of-the-art tracker in <a href=\"https://paperswithcode.com/sota/multi-object-tracking-on-mot20-1?p=bytetrack-multi-object-tracking-by-1\" target=\"_blank\">the field of multi-object tracking</a> because trackers seem to be effective in detecting COTS.</p>\n<p>However, the results were not good.</p>\n<p>Has anyone else tried a multi-object tracker like Bytetrack or deepsort?</p>",
      "rawMarkdown": "I tried to use [Bytetracker](https://github.com/ifzhang/ByteTrack), which is a state-of-the-art tracker in [the field of multi-object tracking](https://paperswithcode.com/sota/multi-object-tracking-on-mot20-1?p=bytetrack-multi-object-tracking-by-1) because trackers seem to be effective in detecting COTS.\n\nHowever, the results were not good.\n\nHas anyone else tried a multi-object tracker like Bytetrack or deepsort?",
      "votes": 6
    },
    {
      "id": 1678432,
      "postDate": "2022-02-06T14:25:07.910Z",
      "content": "<p>I have not completely given up trying, but so far I have not been able to show significant improvements with deepsort or bytetracker. </p>",
      "rawMarkdown": "I have not completely given up trying, but so far I have not been able to show significant improvements with deepsort or bytetracker. ",
      "votes": 3,
      "replies": [
        {
          "id": 1678535,
          "postDate": "2022-02-06T15:42:48.303Z",
          "content": "<p>Thank you for your comment!<br>\nI don't know why, but the COTS position that the Kalman filter infers seems to be wrong….</p>",
          "rawMarkdown": "Thank you for your comment!\nI don't know why, but the COTS position that the Kalman filter infers seems to be wrong...."
        },
        {
          "id": 1678819,
          "postDate": "2022-02-06T20:00:14.730Z",
          "content": "<p>If you make a video of the predicted position on a val sequence, you will see that it is mostly wrong and tracking is not very well but it is just enough to be useful in this competition.</p>",
          "rawMarkdown": "If you make a video of the predicted position on a val sequence, you will see that it is mostly wrong and tracking is not very well but it is just enough to be useful in this competition.",
          "votes": 1
        },
        {
          "id": 1679077,
          "postDate": "2022-02-07T00:39:24.137Z",
          "content": "<p>Yes, I made a video of the predicted position on a val sequence.<br>\nDo you mean that bytetracker or deepsort can't track well in val sequences, but can track well in LB sequences?</p>",
          "rawMarkdown": "Yes, I made a video of the predicted position on a val sequence.\nDo you mean that bytetracker or deepsort can't track well in val sequences, but can track well in LB sequences?"
        },
        {
          "id": 1679082,
          "postDate": "2022-02-07T01:00:09.453Z",
          "content": "<p>I've only explored deep sort a little bit, but tracking is a hard problem with this dataset. I made videos and it maybe needs a lot of fine-tuning (or another solution). The popular tracking solution with norfair is not good too, but it is enough to work well.  </p>",
          "rawMarkdown": "I've only explored deep sort a little bit, but tracking is a hard problem with this dataset. I made videos and it maybe needs a lot of fine-tuning (or another solution). The popular tracking solution with norfair is not good too, but it is enough to work well.  "
        },
        {
          "id": 1679088,
          "postDate": "2022-02-07T01:17:23.607Z",
          "content": "<p>Thank you for your reply. I'll try some more.<br>\nI think that bytetrack can detect COTS with low confidence score.</p>",
          "rawMarkdown": "Thank you for your reply. I'll try some more.\nI think that bytetrack can detect COTS with low confidence score."
        }
      ]
    },
    {
      "id": 1678458,
      "postDate": "2022-02-06T14:42:06.303Z",
      "content": "<p>Yes. I tried both ….. in December …. Think that solution provided by <a href=\"https://www.kaggle.com/parapapapam\" target=\"_blank\">@parapapapam</a> is really great and enough. </p>",
      "rawMarkdown": "Yes. I tried both ..... in December .... Think that solution provided by @parapapapam is really great and enough. ",
      "votes": 2,
      "replies": [
        {
          "id": 1678560,
          "postDate": "2022-02-06T16:13:29.957Z",
          "content": "<p>I appreciate your comment!</p>",
          "rawMarkdown": "I appreciate your comment!"
        }
      ]
    },
    {
      "id": 1783693,
      "postDate": "2022-05-10T15:44:50.600Z",
      "rawMarkdown": "",
      "isDeleted": true
    }
  ],
  "comments": [
    {
      "id": 1678432,
      "author_name": "danjafish",
      "author_url": "",
      "post_date": "2022-02-06T14:25:07.910000",
      "content": "<p>I have not completely given up trying, but so far I have not been able to show significant improvements with deepsort or bytetracker. </p>",
      "votes": 3,
      "replies": [
        {
          "id": 1678535,
          "author_name": "Sugar0",
          "author_url": "",
          "post_date": "2022-02-06T15:42:48.303000",
          "content": "<p>Thank you for your comment!<br>\nI don't know why, but the COTS position that the Kalman filter infers seems to be wrong….</p>",
          "votes": 0,
          "replies": []
        },
        {
          "id": 1678819,
          "author_name": "outwrest",
          "author_url": "",
          "post_date": "2022-02-06T20:00:14.730000",
          "content": "<p>If you make a video of the predicted position on a val sequence, you will see that it is mostly wrong and tracking is not very well but it is just enough to be useful in this competition.</p>",
          "votes": 1,
          "replies": []
        },
        {
          "id": 1679077,
          "author_name": "Sugar0",
          "author_url": "",
          "post_date": "2022-02-07T00:39:24.137000",
          "content": "<p>Yes, I made a video of the predicted position on a val sequence.<br>\nDo you mean that bytetracker or deepsort can't track well in val sequences, but can track well in LB sequences?</p>",
          "votes": 0,
          "replies": []
        },
        {
          "id": 1679082,
          "author_name": "outwrest",
          "author_url": "",
          "post_date": "2022-02-07T01:00:09.453000",
          "content": "<p>I've only explored deep sort a little bit, but tracking is a hard problem with this dataset. I made videos and it maybe needs a lot of fine-tuning (or another solution). The popular tracking solution with norfair is not good too, but it is enough to work well.  </p>",
          "votes": 0,
          "replies": []
        },
        {
          "id": 1679088,
          "author_name": "Sugar0",
          "author_url": "",
          "post_date": "2022-02-07T01:17:23.607000",
          "content": "<p>Thank you for your reply. I'll try some more.<br>\nI think that bytetrack can detect COTS with low confidence score.</p>",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 1678458,
      "author_name": "Remek Kinas",
      "author_url": "",
      "post_date": "2022-02-06T14:42:06.303000",
      "content": "<p>Yes. I tried both ….. in December …. Think that solution provided by <a href=\"https://www.kaggle.com/parapapapam\" target=\"_blank\">@parapapapam</a> is really great and enough. </p>",
      "votes": 2,
      "replies": [
        {
          "id": 1678560,
          "author_name": "Sugar0",
          "author_url": "",
          "post_date": "2022-02-06T16:13:29.957000",
          "content": "<p>I appreciate your comment!</p>",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 1783693,
      "author_name": "",
      "author_url": "",
      "post_date": "2022-05-10T15:44:50.600000",
      "content": "",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1678271": "I tried to use [Bytetracker](https://github.com/ifzhang/ByteTrack), which is a state-of-the-art tracker in [the field of multi-object tracking](https://paperswithcode.com/sota/multi-object-tracking-on-mot20-1?p=bytetrack-multi-object-tracking-by-1) because trackers seem to be effective in detecting COTS.\n\nHowever, the results were not good.\n\nHas anyone else tried a multi-object tracker like Bytetrack or deepsort?",
    "1678432": "I have not completely given up trying, but so far I have not been able to show significant improvements with deepsort or bytetracker. ",
    "1678458": "Yes. I tried both ..... in December .... Think that solution provided by @parapapapam is really great and enough. ",
    "1783693": ""
  }
}