{
  "id": 223384,
  "title": "Will ViT(Vision Transformer) survive?",
  "url": "/competitions/ranzcr-clip-catheter-line-classification/discussion/223384",
  "author_name": "",
  "post_date": "2021-03-03T15:26:25.512752300Z",
  "votes": 4,
  "comment_count": 9,
  "views": 0,
  "content": "<p>I joined late after the cassava competition.<br>\n( Maybe it's too late. :) )</p>\n<p>After testing a few models, I'm training ViT models such as DeiT.</p>\n<p>The CV score were just around 0.9.<br>\n(I did not submit results for ViT because CV was lower than other models.)<br>\nThe resnet200d seems to be doing well compared to other models.</p>\n<p>Could you share something about your experience with ViT?</p>\n<p>Thank you!</p>",
  "messages": [
    {
      "id": "1225386",
      "postDate": "03/03/2021 15:26:25",
      "content": "<p>I joined late after the cassava competition.<br>\n( Maybe it's too late. :) )</p>\n<p>After testing a few models, I'm training ViT models such as DeiT.</p>\n<p>The CV score were just around 0.9.<br>\n(I did not submit results for ViT because CV was lower than other models.)<br>\nThe resnet200d seems to be doing well compared to other models.</p>\n<p>Could you share something about your experience with ViT?</p>\n<p>Thank you!</p>",
      "rawMarkdown": "I joined late after the cassava competition.\n( Maybe it's too late. :) )\n\nAfter testing a few models, I'm training ViT models such as DeiT.\n\nThe CV score were just around 0.9.\n(I did not submit results for ViT because CV was lower than other models.)\nThe resnet200d seems to be doing well compared to other models.\n\nCould you share something about your experience with ViT?\n\nThank you!",
      "votes": null
    },
    {
      "id": "1225390",
      "postDate": "03/03/2021 15:29:53",
      "content": "<p>I tried VIT and got single fold CV of close to 91. I used Image size of 800 but did not submit because of low CV. </p>",
      "rawMarkdown": "I tried VIT and got single fold CV of close to 91. I used Image size of 800 but did not submit because of low CV.",
      "votes": null
    },
    {
      "id": "1225396",
      "postDate": "03/03/2021 15:35:26",
      "content": "<p>There is not much information in the discussion about ViT. Thanks for sharing your result!</p>",
      "rawMarkdown": "There is not much information in the discussion about ViT. Thanks for sharing your result!",
      "votes": null
    },
    {
      "id": "1225419",
      "postDate": "03/03/2021 15:53:01",
      "content": "<p>I tried vit with 384 image size, but cv score 0.84</p>",
      "rawMarkdown": "I tried vit with 384 image size, but cv score 0.84",
      "votes": null
    },
    {
      "id": "1225437",
      "postDate": "03/03/2021 16:02:00",
      "content": "<p>It seems difficult to survive when we only see CV scores. :)</p>",
      "rawMarkdown": "It seems difficult to survive when we only see CV scores. :)",
      "votes": null
    },
    {
      "id": "1225494",
      "postDate": "03/03/2021 16:34:26",
      "content": "<p>I got single fold CV 0.945 LB 0.953 with vit_base_resnet50_384, image size: 384</p>",
      "rawMarkdown": "I got single fold CV 0.945 LB 0.953 with vit_base_resnet50_384, image size: 384",
      "votes": null
    },
    {
      "id": "1225502",
      "postDate": "03/03/2021 16:39:49",
      "content": "<p>Nice work. Looks like the CV is pretty high. Thanks for sharing.</p>",
      "rawMarkdown": "Nice work. Looks like the CV is pretty high. Thanks for sharing.",
      "votes": null
    },
    {
      "id": "1225509",
      "postDate": "03/03/2021 16:46:51",
      "content": "<p>thank you for sharing </p>",
      "rawMarkdown": "thank you for sharing",
      "votes": null
    },
    {
      "id": "1225749",
      "postDate": "03/03/2021 22:10:15",
      "content": "<p>In this competition, image size is really important (the bigger the better). You can see improvement CV from moving to larger image size. ViT and Deit require input size that is relative small, hence why they are not popular here and performance is not that great too.</p>",
      "rawMarkdown": "In this competition, image size is really important (the bigger the better). You can see improvement CV from moving to larger image size. ViT and Deit require input size that is relative small, hence why they are not popular here and performance is not that great too.",
      "votes": null
    },
    {
      "id": "1274343",
      "postDate": "04/15/2021 08:07:55",
      "content": "<p>I am trying to use deit for this task, but i also got a poor result. Could you share something about your experience with ViT or DeiT?🙏</p>",
      "rawMarkdown": "I am trying to use deit for this task, but i also got a poor result. Could you share something about your experience with ViT or DeiT?🙏",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1225390,
      "author_name": "vickygoyal",
      "author_url": "",
      "post_date": "03/03/2021 15:29:53",
      "content": "<p>I tried VIT and got single fold CV of close to 91. I used Image size of 800 but did not submit because of low CV. </p>",
      "votes": null,
      "replies": [
        {
          "id": 1225396,
          "author_name": "piantic",
          "author_url": "",
          "post_date": "03/03/2021 15:35:26",
          "content": "<p>There is not much information in the discussion about ViT. Thanks for sharing your result!</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 1225419,
      "author_name": "likezerocoke",
      "author_url": "",
      "post_date": "03/03/2021 15:53:01",
      "content": "<p>I tried vit with 384 image size, but cv score 0.84</p>",
      "votes": null,
      "replies": [
        {
          "id": 1225437,
          "author_name": "piantic",
          "author_url": "",
          "post_date": "03/03/2021 16:02:00",
          "content": "<p>It seems difficult to survive when we only see CV scores. :)</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 1225494,
      "author_name": "mknzfr",
      "author_url": "",
      "post_date": "03/03/2021 16:34:26",
      "content": "<p>I got single fold CV 0.945 LB 0.953 with vit_base_resnet50_384, image size: 384</p>",
      "votes": null,
      "replies": [
        {
          "id": 1225502,
          "author_name": "piantic",
          "author_url": "",
          "post_date": "03/03/2021 16:39:49",
          "content": "<p>Nice work. Looks like the CV is pretty high. Thanks for sharing.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1225509,
          "author_name": "likezerocoke",
          "author_url": "",
          "post_date": "03/03/2021 16:46:51",
          "content": "<p>thank you for sharing </p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1225749,
          "author_name": "tom88jerry",
          "author_url": "",
          "post_date": "03/03/2021 22:10:15",
          "content": "<p>In this competition, image size is really important (the bigger the better). You can see improvement CV from moving to larger image size. ViT and Deit require input size that is relative small, hence why they are not popular here and performance is not that great too.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 1274343,
      "author_name": "mrjiangxiaoben",
      "author_url": "",
      "post_date": "04/15/2021 08:07:55",
      "content": "<p>I am trying to use deit for this task, but i also got a poor result. Could you share something about your experience with ViT or DeiT?🙏</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1225386": "I joined late after the cassava competition.\n( Maybe it's too late. :) )\n\nAfter testing a few models, I'm training ViT models such as DeiT.\n\nThe CV score were just around 0.9.\n(I did not submit results for ViT because CV was lower than other models.)\nThe resnet200d seems to be doing well compared to other models.\n\nCould you share something about your experience with ViT?\n\nThank you!",
    "1225390": "I tried VIT and got single fold CV of close to 91. I used Image size of 800 but did not submit because of low CV.",
    "1225396": "There is not much information in the discussion about ViT. Thanks for sharing your result!",
    "1225419": "I tried vit with 384 image size, but cv score 0.84",
    "1225437": "It seems difficult to survive when we only see CV scores. :)",
    "1225494": "I got single fold CV 0.945 LB 0.953 with vit_base_resnet50_384, image size: 384",
    "1225502": "Nice work. Looks like the CV is pretty high. Thanks for sharing.",
    "1225509": "thank you for sharing",
    "1225749": "In this competition, image size is really important (the bigger the better). You can see improvement CV from moving to larger image size. ViT and Deit require input size that is relative small, hence why they are not popular here and performance is not that great too.",
    "1274343": "I am trying to use deit for this task, but i also got a poor result. Could you share something about your experience with ViT or DeiT?🙏"
  },
  "source": "meta"
}