{
  "id": 223220,
  "title": "i think synthetic data generation + transformer will rule this competition",
  "url": "/competitions/bms-molecular-translation/discussion/223220",
  "author_name": "",
  "post_date": "2021-03-03T01:16:25.652600200Z",
  "votes": 31,
  "comment_count": 4,
  "views": 0,
  "content": "<p>this reminds me of the top solutions of the open vaccine challenge …</p>\n<p>more information to come …</p>\n<p>one may want to google for freeware, opensource that does this … something like molecular OCR</p>",
  "messages": [
    {
      "id": "1224677",
      "postDate": "03/03/2021 01:16:25",
      "content": "<p>this reminds me of the top solutions of the open vaccine challenge …</p>\n<p>more information to come …</p>\n<p>one may want to google for freeware, opensource that does this … something like molecular OCR</p>",
      "rawMarkdown": "this reminds me of the top solutions of the open vaccine challenge ...\n\nmore information to come ...\n\none may want to google for freeware, opensource that does this ... something like molecular OCR",
      "votes": null
    },
    {
      "id": "1224719",
      "postDate": "03/03/2021 02:22:48",
      "content": "<p>As the image is similar to the bengaliai-cv19 Competition, it is noisy and grayscale image. The main thing we learn is the image structure. I think it is necessary to use cutmix, mixup, fmix and some denoising enhancement functions in this Competition…</p>",
      "rawMarkdown": "As the image is similar to the bengaliai-cv19 Competition, it is noisy and grayscale image. The main thing we learn is the image structure. I think it is necessary to use cutmix, mixup, fmix and some denoising enhancement functions in this Competition...",
      "votes": null
    },
    {
      "id": "1224739",
      "postDate": "03/03/2021 02:56:26",
      "content": "<p>Hi, could you share the link to the <code>top solution</code> that using data generation technique, I could not find one<br>\nThank you</p>",
      "rawMarkdown": "Hi, could you share the link to the `top solution` that using data generation technique, I could not find one\nThank you",
      "votes": null
    },
    {
      "id": "1225303",
      "postDate": "03/03/2021 14:16:41",
      "content": "<p>Yeah <a href=\"https://www.kaggle.com/hengck23\" target=\"_blank\">@hengck23</a> . It has various variants, <a href=\"https://jcheminf.biomedcentral.com/articles/10.1186/s13321-020-00465-0\" target=\"_blank\">this</a> link has all the related keywords.</p>",
      "rawMarkdown": "Yeah @hengck23 . It has various variants, [this](https://jcheminf.biomedcentral.com/articles/10.1186/s13321-020-00465-0) link has all the related keywords.",
      "votes": null
    },
    {
      "id": "1260196",
      "postDate": "04/01/2021 23:27:56",
      "content": "<p>Don't you think that by using cutmix or mixup here will make it harder for the model to learn the discriminative features well. Unlike classification where these augmentations really improve the performance of the model, I doubt they will help in captioning setting. </p>\n<p>I may be completely wrong. Is there any research paper or your own study supporting this?</p>",
      "rawMarkdown": "Don't you think that by using cutmix or mixup here will make it harder for the model to learn the discriminative features well. Unlike classification where these augmentations really improve the performance of the model, I doubt they will help in captioning setting. \n\nI may be completely wrong. Is there any research paper or your own study supporting this?",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1224719,
      "author_name": "zhangeng",
      "author_url": "",
      "post_date": "03/03/2021 02:22:48",
      "content": "<p>As the image is similar to the bengaliai-cv19 Competition, it is noisy and grayscale image. The main thing we learn is the image structure. I think it is necessary to use cutmix, mixup, fmix and some denoising enhancement functions in this Competition…</p>",
      "votes": null,
      "replies": [
        {
          "id": 1260196,
          "author_name": "ayuraj",
          "author_url": "",
          "post_date": "04/01/2021 23:27:56",
          "content": "<p>Don't you think that by using cutmix or mixup here will make it harder for the model to learn the discriminative features well. Unlike classification where these augmentations really improve the performance of the model, I doubt they will help in captioning setting. </p>\n<p>I may be completely wrong. Is there any research paper or your own study supporting this?</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 1224739,
      "author_name": "ptran1203",
      "author_url": "",
      "post_date": "03/03/2021 02:56:26",
      "content": "<p>Hi, could you share the link to the <code>top solution</code> that using data generation technique, I could not find one<br>\nThank you</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1225303,
      "author_name": "manabendrarout",
      "author_url": "",
      "post_date": "03/03/2021 14:16:41",
      "content": "<p>Yeah <a href=\"https://www.kaggle.com/hengck23\" target=\"_blank\">@hengck23</a> . It has various variants, <a href=\"https://jcheminf.biomedcentral.com/articles/10.1186/s13321-020-00465-0\" target=\"_blank\">this</a> link has all the related keywords.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1224677": "this reminds me of the top solutions of the open vaccine challenge ...\n\nmore information to come ...\n\none may want to google for freeware, opensource that does this ... something like molecular OCR",
    "1224719": "As the image is similar to the bengaliai-cv19 Competition, it is noisy and grayscale image. The main thing we learn is the image structure. I think it is necessary to use cutmix, mixup, fmix and some denoising enhancement functions in this Competition...",
    "1224739": "Hi, could you share the link to the `top solution` that using data generation technique, I could not find one\nThank you",
    "1225303": "Yeah @hengck23 . It has various variants, [this](https://jcheminf.biomedcentral.com/articles/10.1186/s13321-020-00465-0) link has all the related keywords.",
    "1260196": "Don't you think that by using cutmix or mixup here will make it harder for the model to learn the discriminative features well. Unlike classification where these augmentations really improve the performance of the model, I doubt they will help in captioning setting. \n\nI may be completely wrong. Is there any research paper or your own study supporting this?"
  },
  "source": "meta"
}