{
  "id": 171702,
  "title": "Duplicates during ensembling",
  "url": "/competitions/global-wheat-detection/discussion/171702",
  "author_name": "",
  "post_date": "2020-08-02T05:59:04.624063800Z",
  "votes": null,
  "comment_count": 6,
  "views": 0,
  "content": "<p>Does Alex's ensembling kernel that ensembles multiple folds of the model,take care of duplicate boxes. How it takes care?</p>",
  "messages": [
    {
      "id": "954822",
      "postDate": "08/02/2020 05:59:04",
      "content": "<p>Does Alex's ensembling kernel that ensembles multiple folds of the model,take care of duplicate boxes. How it takes care?</p>",
      "rawMarkdown": "Does Alex's ensembling kernel that ensembles multiple folds of the model,take care of duplicate boxes. How it takes care?",
      "votes": null
    },
    {
      "id": "954827",
      "postDate": "08/02/2020 06:08:14",
      "content": "<p>He uses weighted box fusion (WBF) to ensemble multiple predictions. Maybe I don't understand what you're asking</p>",
      "rawMarkdown": "He uses weighted box fusion (WBF) to ensemble multiple predictions. Maybe I don't understand what you're asking",
      "votes": null
    },
    {
      "id": "954838",
      "postDate": "08/02/2020 06:13:27",
      "content": "<p>i see he does ensembling of all the folds .. each fold may give either same boxes  or some additional boxes\nsay fold 1 for an image give 30 boxes, fold2 models give 34 .. out of these there may be some boxes by both how does wbf takes care of merging these into single</p>",
      "rawMarkdown": "i see he does ensembling of all the folds .. each fold may give either same boxes  or some additional boxes\nsay fold 1 for an image give 30 boxes, fold2 models give 34 .. out of these there may be some boxes by both how does wbf takes care of merging these into single",
      "votes": null
    },
    {
      "id": "954855",
      "postDate": "08/02/2020 06:32:42",
      "content": "<p>That is what WBF is for, fusing and ensemble of multiple predictions. You can read about the technical aspects of how it works in the paper here: <a href=\"https://arxiv.org/pdf/1910.13302.pdf\">https://arxiv.org/pdf/1910.13302.pdf</a></p>",
      "rawMarkdown": "That is what WBF is for, fusing and ensemble of multiple predictions. You can read about the technical aspects of how it works in the paper here: https://arxiv.org/pdf/1910.13302.pdf",
      "votes": null
    },
    {
      "id": "954865",
      "postDate": "08/02/2020 06:43:34",
      "content": "<p>thanks so any number of models we ensemble it would fuse them to form a single set for a given images ?</p>",
      "rawMarkdown": "thanks so any number of models we ensemble it would fuse them to form a single set for a given images ?",
      "votes": null
    },
    {
      "id": "954881",
      "postDate": "08/02/2020 07:01:29",
      "content": "<p>Yeah, that's the idea</p>",
      "rawMarkdown": "Yeah, that's the idea",
      "votes": null
    },
    {
      "id": "954902",
      "postDate": "08/02/2020 07:25:44",
      "content": "<p>is there also any kernel doing pseudo labelling with ensembling ?\nor is pseudo labelling for only single model ?</p>",
      "rawMarkdown": "is there also any kernel doing pseudo labelling with ensembling ?\nor is pseudo labelling for only single model ?",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 954827,
      "author_name": "matthewmasters",
      "author_url": "",
      "post_date": "08/02/2020 06:08:14",
      "content": "<p>He uses weighted box fusion (WBF) to ensemble multiple predictions. Maybe I don't understand what you're asking</p>",
      "votes": null,
      "replies": [
        {
          "id": 954838,
          "author_name": "jaideepvalani",
          "author_url": "",
          "post_date": "08/02/2020 06:13:27",
          "content": "<p>i see he does ensembling of all the folds .. each fold may give either same boxes  or some additional boxes\nsay fold 1 for an image give 30 boxes, fold2 models give 34 .. out of these there may be some boxes by both how does wbf takes care of merging these into single</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 954855,
          "author_name": "matthewmasters",
          "author_url": "",
          "post_date": "08/02/2020 06:32:42",
          "content": "<p>That is what WBF is for, fusing and ensemble of multiple predictions. You can read about the technical aspects of how it works in the paper here: <a href=\"https://arxiv.org/pdf/1910.13302.pdf\">https://arxiv.org/pdf/1910.13302.pdf</a></p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 954865,
          "author_name": "jaideepvalani",
          "author_url": "",
          "post_date": "08/02/2020 06:43:34",
          "content": "<p>thanks so any number of models we ensemble it would fuse them to form a single set for a given images ?</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 954881,
          "author_name": "matthewmasters",
          "author_url": "",
          "post_date": "08/02/2020 07:01:29",
          "content": "<p>Yeah, that's the idea</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 954902,
          "author_name": "jaideepvalani",
          "author_url": "",
          "post_date": "08/02/2020 07:25:44",
          "content": "<p>is there also any kernel doing pseudo labelling with ensembling ?\nor is pseudo labelling for only single model ?</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "954822": "Does Alex's ensembling kernel that ensembles multiple folds of the model,take care of duplicate boxes. How it takes care?",
    "954827": "He uses weighted box fusion (WBF) to ensemble multiple predictions. Maybe I don't understand what you're asking",
    "954838": "i see he does ensembling of all the folds .. each fold may give either same boxes  or some additional boxes\nsay fold 1 for an image give 30 boxes, fold2 models give 34 .. out of these there may be some boxes by both how does wbf takes care of merging these into single",
    "954855": "That is what WBF is for, fusing and ensemble of multiple predictions. You can read about the technical aspects of how it works in the paper here: https://arxiv.org/pdf/1910.13302.pdf",
    "954865": "thanks so any number of models we ensemble it would fuse them to form a single set for a given images ?",
    "954881": "Yeah, that's the idea",
    "954902": "is there also any kernel doing pseudo labelling with ensembling ?\nor is pseudo labelling for only single model ?"
  },
  "source": "meta"
}