{
  "id": 294834,
  "title": "Has anyone tried semi-training? Does that improve your score?",
  "url": "/competitions/sartorius-cell-instance-segmentation/discussion/294834",
  "author_name": "",
  "post_date": "2021-12-13T06:40:17.945636600Z",
  "votes": 5,
  "comment_count": 3,
  "views": 0,
  "content": "<p>We are given train-semi-supervised data. I'm wondering if anyone use it during your training, and are there any improvement after adding that data?</p>",
  "messages": [
    {
      "id": "1616094",
      "postDate": "12/13/2021 06:40:17",
      "content": "<p>We are given train-semi-supervised data. I'm wondering if anyone use it during your training, and are there any improvement after adding that data?</p>",
      "rawMarkdown": "We are given train-semi-supervised data. I'm wondering if anyone use it during your training, and are there any improvement after adding that data?",
      "votes": null
    },
    {
      "id": "1617598",
      "postDate": "12/14/2021 07:38:43",
      "content": "<p>semi-training can get 0.308 from 0.291 (based on torchvision)<br>\nbut it hard to integrate with detectron2 for me</p>",
      "rawMarkdown": "semi-training can get 0.308 from 0.291 (based on torchvision)\nbut it hard to integrate with detectron2 for me",
      "votes": null
    },
    {
      "id": "1618398",
      "postDate": "12/15/2021 00:28:15",
      "content": "<p>hey did you use some data distillation techniques or did you directly train after pseudo-labelling all the given SSL data? </p>",
      "rawMarkdown": "hey did you use some data distillation techniques or did you directly train after pseudo-labelling all the given SSL data?",
      "votes": null
    },
    {
      "id": "1618462",
      "postDate": "12/15/2021 02:37:56",
      "content": "<p>data distillation</p>",
      "rawMarkdown": "data distillation",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1617598,
      "author_name": "sqqqqy",
      "author_url": "",
      "post_date": "12/14/2021 07:38:43",
      "content": "<p>semi-training can get 0.308 from 0.291 (based on torchvision)<br>\nbut it hard to integrate with detectron2 for me</p>",
      "votes": null,
      "replies": [
        {
          "id": 1618398,
          "author_name": "ferlockx",
          "author_url": "",
          "post_date": "12/15/2021 00:28:15",
          "content": "<p>hey did you use some data distillation techniques or did you directly train after pseudo-labelling all the given SSL data? </p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1618462,
          "author_name": "sqqqqy",
          "author_url": "",
          "post_date": "12/15/2021 02:37:56",
          "content": "<p>data distillation</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1616094": "We are given train-semi-supervised data. I'm wondering if anyone use it during your training, and are there any improvement after adding that data?",
    "1617598": "semi-training can get 0.308 from 0.291 (based on torchvision)\nbut it hard to integrate with detectron2 for me",
    "1618398": "hey did you use some data distillation techniques or did you directly train after pseudo-labelling all the given SSL data?",
    "1618462": "data distillation"
  },
  "source": "meta"
}