{
  "id": 96133,
  "title": "Which  one will be better 1) fine tuning or 2) Training From Scratch ?",
  "url": "/competitions/open-images-2019-object-detection/discussion/96133",
  "author_name": "",
  "post_date": "2019-06-18T08:33:29.963808600Z",
  "votes": 1,
  "comment_count": 9,
  "views": 0,
  "content": "<p>As the given dataset is very large and has enough testing and training data,\nUsing fine tuning on pre trained model will lead to overfit ?\nwill it good to finetune it or should we train it from scratch ? </p>",
  "messages": [
    {
      "id": "554959",
      "postDate": "06/18/2019 08:33:29",
      "content": "<p>As the given dataset is very large and has enough testing and training data,\nUsing fine tuning on pre trained model will lead to overfit ?\nwill it good to finetune it or should we train it from scratch ? </p>",
      "rawMarkdown": "As the given dataset is very large and has enough testing and training data,\nUsing fine tuning on pre trained model will lead to overfit ?\nwill it good to finetune it or should we train it from scratch ?",
      "votes": null
    },
    {
      "id": "554986",
      "postDate": "06/18/2019 09:27:31",
      "content": "<p>i would prefer pretrained models than training from scratch</p>",
      "rawMarkdown": "i would prefer pretrained models than training from scratch",
      "votes": null
    },
    {
      "id": "555004",
      "postDate": "06/18/2019 09:53:44",
      "content": "<p>ohk thank you very much for your opinion!!\nBut dont you think that this much amount of larger data will lead to overfit!!!!!</p>",
      "rawMarkdown": "ohk thank you very much for your opinion!!\nBut dont you think that this much amount of larger data will lead to overfit!!!!!",
      "votes": null
    },
    {
      "id": "555005",
      "postDate": "06/18/2019 09:54:01",
      "content": "<p>Ad if not ,\nCan you explain me how?</p>",
      "rawMarkdown": "Ad if not ,\nCan you explain me how?",
      "votes": null
    },
    {
      "id": "555036",
      "postDate": "06/18/2019 10:55:24",
      "content": "<p>It's hard to overfit with such a huge amount of data</p>",
      "rawMarkdown": "It's hard to overfit with such a huge amount of data",
      "votes": null
    },
    {
      "id": "555047",
      "postDate": "06/18/2019 11:15:59",
      "content": "<p>the more the data the better the generalization...so lesser overfitting. smaller number of samples leads to overfitting in which model tries to \"remember\" the patterns</p>",
      "rawMarkdown": "the more the data the better the generalization...so lesser overfitting. smaller number of samples leads to overfitting in which model tries to \"remember\" the patterns",
      "votes": null
    },
    {
      "id": "556408",
      "postDate": "06/20/2019 07:58:39",
      "content": "<p>In my opinion its better to fine tune then to train from scratch</p>",
      "rawMarkdown": "In my opinion its better to fine tune then to train from scratch",
      "votes": null
    },
    {
      "id": "557854",
      "postDate": "06/21/2019 14:22:29",
      "content": "<p>Thank you very much for your help\ni appreciate that</p>",
      "rawMarkdown": "Thank you very much for your help\ni appreciate that",
      "votes": null
    },
    {
      "id": "557855",
      "postDate": "06/21/2019 14:22:50",
      "content": "<p>Thanks a lot\nI appreciate your help</p>",
      "rawMarkdown": "Thanks a lot\nI appreciate your help",
      "votes": null
    },
    {
      "id": "557857",
      "postDate": "06/21/2019 14:23:14",
      "content": "<p>Thank you very much your help\nI appreciate your help</p>",
      "rawMarkdown": "Thank you very much your help\nI appreciate your help",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 554986,
      "author_name": "bakkuuuu",
      "author_url": "",
      "post_date": "06/18/2019 09:27:31",
      "content": "<p>i would prefer pretrained models than training from scratch</p>",
      "votes": null,
      "replies": [
        {
          "id": 555004,
          "author_name": "siddhrath",
          "author_url": "",
          "post_date": "06/18/2019 09:53:44",
          "content": "<p>ohk thank you very much for your opinion!!\nBut dont you think that this much amount of larger data will lead to overfit!!!!!</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 555005,
          "author_name": "siddhrath",
          "author_url": "",
          "post_date": "06/18/2019 09:54:01",
          "content": "<p>Ad if not ,\nCan you explain me how?</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 555047,
          "author_name": "bakkuuuu",
          "author_url": "",
          "post_date": "06/18/2019 11:15:59",
          "content": "<p>the more the data the better the generalization...so lesser overfitting. smaller number of samples leads to overfitting in which model tries to \"remember\" the patterns</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 557855,
          "author_name": "siddhrath",
          "author_url": "",
          "post_date": "06/21/2019 14:22:50",
          "content": "<p>Thanks a lot\nI appreciate your help</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 555036,
      "author_name": "dsvolkov",
      "author_url": "",
      "post_date": "06/18/2019 10:55:24",
      "content": "<p>It's hard to overfit with such a huge amount of data</p>",
      "votes": null,
      "replies": [
        {
          "id": 557857,
          "author_name": "siddhrath",
          "author_url": "",
          "post_date": "06/21/2019 14:23:14",
          "content": "<p>Thank you very much your help\nI appreciate your help</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 556408,
      "author_name": "krishnakatyal",
      "author_url": "",
      "post_date": "06/20/2019 07:58:39",
      "content": "<p>In my opinion its better to fine tune then to train from scratch</p>",
      "votes": null,
      "replies": [
        {
          "id": 557854,
          "author_name": "siddhrath",
          "author_url": "",
          "post_date": "06/21/2019 14:22:29",
          "content": "<p>Thank you very much for your help\ni appreciate that</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "554959": "As the given dataset is very large and has enough testing and training data,\nUsing fine tuning on pre trained model will lead to overfit ?\nwill it good to finetune it or should we train it from scratch ?",
    "554986": "i would prefer pretrained models than training from scratch",
    "555004": "ohk thank you very much for your opinion!!\nBut dont you think that this much amount of larger data will lead to overfit!!!!!",
    "555005": "Ad if not ,\nCan you explain me how?",
    "555036": "It's hard to overfit with such a huge amount of data",
    "555047": "the more the data the better the generalization...so lesser overfitting. smaller number of samples leads to overfitting in which model tries to \"remember\" the patterns",
    "556408": "In my opinion its better to fine tune then to train from scratch",
    "557854": "Thank you very much for your help\ni appreciate that",
    "557855": "Thanks a lot\nI appreciate your help",
    "557857": "Thank you very much your help\nI appreciate your help"
  },
  "source": "meta"
}