{
  "id": 129273,
  "title": "[Question] about cutmix and mixup(help me....)",
  "url": "/competitions/bengaliai-cv19/discussion/129273",
  "author_name": "",
  "post_date": "2020-02-06T18:48:49.496339200Z",
  "votes": 2,
  "comment_count": 9,
  "views": 0,
  "content": "<p>hi i am trying cutmix + mixup.\nThis is my settings.\n<code>\nimage_size: 128\ncutmix: 0.4\nmixup: 0.4\naugumentation: random_rotate\ncv: 0.96551740474\nlb:0.0615\n</code>\nmy lb is terrible.\nI attach my log.txt.\nAs you can see in the log, My training loss is always lower than val loss.\nIs this normal?</p>",
  "messages": [
    {
      "id": "738602",
      "postDate": "02/06/2020 18:48:49",
      "content": "<p>hi i am trying cutmix + mixup.\nThis is my settings.\n<code>\nimage_size: 128\ncutmix: 0.4\nmixup: 0.4\naugumentation: random_rotate\ncv: 0.96551740474\nlb:0.0615\n</code>\nmy lb is terrible.\nI attach my log.txt.\nAs you can see in the log, My training loss is always lower than val loss.\nIs this normal?</p>",
      "rawMarkdown": "hi i am trying cutmix + mixup.\nThis is my settings.\n```\nimage_size: 128\ncutmix: 0.4\nmixup: 0.4\naugumentation: random_rotate\ncv: 0.96551740474\nlb:0.0615\n```\nmy lb is terrible.\nI attach my log.txt.\nAs you can see in the log, My training loss is always lower than val loss.\nIs this normal?",
      "votes": null
    },
    {
      "id": "738606",
      "postDate": "02/06/2020 18:58:23",
      "content": "<p>That LB score isn't from a bad model, you're not submitting on the full test set. I'm guessing that is a typo based on your LB place?</p>",
      "rawMarkdown": "That LB score isn't from a bad model, you're not submitting on the full test set. I'm guessing that is a typo based on your LB place?",
      "votes": null
    },
    {
      "id": "738671",
      "postDate": "02/06/2020 21:12:17",
      "content": "<p>Yes its normal when using mixup or cutmix to have high training loss. </p>",
      "rawMarkdown": "Yes its normal when using mixup or cutmix to have high training loss.",
      "votes": null
    },
    {
      "id": "738801",
      "postDate": "02/07/2020 03:24:23",
      "content": "<p>It was just a typo...\n Thank you, both of you.</p>",
      "rawMarkdown": "It was just a typo...\n Thank you, both of you.",
      "votes": null
    },
    {
      "id": "738810",
      "postDate": "02/07/2020 03:29:21",
      "content": "<p>Did you forget to load the model weights?😯 </p>",
      "rawMarkdown": "Did you forget to load the model weights?😯",
      "votes": null
    },
    {
      "id": "738812",
      "postDate": "02/07/2020 03:34:36",
      "content": "<p>I forgot to resize the image😂</p>",
      "rawMarkdown": "I forgot to resize the image😂",
      "votes": null
    },
    {
      "id": "739837",
      "postDate": "02/08/2020 13:38:07",
      "content": "<p><code>cutmix: 0.4\nmixup: 0.4</code>\nwhat's the meaning of 0.4?</p>",
      "rawMarkdown": "`cutmix: 0.4\nmixup: 0.4`\nwhat's the meaning of 0.4?",
      "votes": null
    },
    {
      "id": "740347",
      "postDate": "02/09/2020 09:54:47",
      "content": "<p>It is just a ratio.　At the time of training, I decide what percentage to do each.\n40% -&gt; cutmix\n40% -&gt; mixup\n20% -&gt; no mixup and cutmix</p>",
      "rawMarkdown": "It is just a ratio.　At the time of training, I decide what percentage to do each.\n40% -&gt; cutmix\n40% -&gt; mixup\n20% -&gt; no mixup and cutmix",
      "votes": null
    },
    {
      "id": "740348",
      "postDate": "02/09/2020 09:56:55",
      "content": "<p>check out this discussion.\n<a href=\"https://www.kaggle.com/c/bengaliai-cv19/discussion/126504\">mixup/cutmix is all you need</a></p>",
      "rawMarkdown": "check out this discussion.\n[mixup/cutmix is all you need](https://www.kaggle.com/c/bengaliai-cv19/discussion/126504)",
      "votes": null
    },
    {
      "id": "741119",
      "postDate": "02/10/2020 07:50:22",
      "content": "<p>thank you</p>",
      "rawMarkdown": "thank you",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 738606,
      "author_name": "greatgamedota",
      "author_url": "",
      "post_date": "02/06/2020 18:58:23",
      "content": "<p>That LB score isn't from a bad model, you're not submitting on the full test set. I'm guessing that is a typo based on your LB place?</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 738671,
      "author_name": "drhabib",
      "author_url": "",
      "post_date": "02/06/2020 21:12:17",
      "content": "<p>Yes its normal when using mixup or cutmix to have high training loss. </p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 738801,
      "author_name": "kani23",
      "author_url": "",
      "post_date": "02/07/2020 03:24:23",
      "content": "<p>It was just a typo...\n Thank you, both of you.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 738810,
      "author_name": "garybios",
      "author_url": "",
      "post_date": "02/07/2020 03:29:21",
      "content": "<p>Did you forget to load the model weights?😯 </p>",
      "votes": null,
      "replies": [
        {
          "id": 738812,
          "author_name": "kani23",
          "author_url": "",
          "post_date": "02/07/2020 03:34:36",
          "content": "<p>I forgot to resize the image😂</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 739837,
      "author_name": "dldmw579",
      "author_url": "",
      "post_date": "02/08/2020 13:38:07",
      "content": "<p><code>cutmix: 0.4\nmixup: 0.4</code>\nwhat's the meaning of 0.4?</p>",
      "votes": null,
      "replies": [
        {
          "id": 740347,
          "author_name": "kani23",
          "author_url": "",
          "post_date": "02/09/2020 09:54:47",
          "content": "<p>It is just a ratio.　At the time of training, I decide what percentage to do each.\n40% -&gt; cutmix\n40% -&gt; mixup\n20% -&gt; no mixup and cutmix</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 740348,
          "author_name": "kani23",
          "author_url": "",
          "post_date": "02/09/2020 09:56:55",
          "content": "<p>check out this discussion.\n<a href=\"https://www.kaggle.com/c/bengaliai-cv19/discussion/126504\">mixup/cutmix is all you need</a></p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 741119,
          "author_name": "dldmw579",
          "author_url": "",
          "post_date": "02/10/2020 07:50:22",
          "content": "<p>thank you</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "738602": "hi i am trying cutmix + mixup.\nThis is my settings.\n```\nimage_size: 128\ncutmix: 0.4\nmixup: 0.4\naugumentation: random_rotate\ncv: 0.96551740474\nlb:0.0615\n```\nmy lb is terrible.\nI attach my log.txt.\nAs you can see in the log, My training loss is always lower than val loss.\nIs this normal?",
    "738606": "That LB score isn't from a bad model, you're not submitting on the full test set. I'm guessing that is a typo based on your LB place?",
    "738671": "Yes its normal when using mixup or cutmix to have high training loss.",
    "738801": "It was just a typo...\n Thank you, both of you.",
    "738810": "Did you forget to load the model weights?😯",
    "738812": "I forgot to resize the image😂",
    "739837": "`cutmix: 0.4\nmixup: 0.4`\nwhat's the meaning of 0.4?",
    "740347": "It is just a ratio.　At the time of training, I decide what percentage to do each.\n40% -&gt; cutmix\n40% -&gt; mixup\n20% -&gt; no mixup and cutmix",
    "740348": "check out this discussion.\n[mixup/cutmix is all you need](https://www.kaggle.com/c/bengaliai-cv19/discussion/126504)",
    "741119": "thank you"
  },
  "source": "meta"
}