{
  "id": 272223,
  "title": "Can we generate captions and then try to match them to the wiki description?",
  "url": "/competitions/wikipedia-image-caption/discussion/272223",
  "author_name": "",
  "post_date": "2021-09-14T16:20:46.377019300Z",
  "votes": 2,
  "comment_count": 3,
  "views": 0,
  "content": "<p>Hi,</p>\n<p>I was just thinking can we take this as a 2 part problem in which let's say for part 1 we try to generate captions for all images and in the part 2 we try to identify the closest caption to what we generated? I know it will be complex since out captions will be generic but we can try to identify those generic things in the wikipedia article? does this seem doable?</p>\n<p>Any thoughts :)</p>",
  "messages": [
    {
      "id": "1512832",
      "postDate": "09/14/2021 16:20:46",
      "content": "<p>Hi,</p>\n<p>I was just thinking can we take this as a 2 part problem in which let's say for part 1 we try to generate captions for all images and in the part 2 we try to identify the closest caption to what we generated? I know it will be complex since out captions will be generic but we can try to identify those generic things in the wikipedia article? does this seem doable?</p>\n<p>Any thoughts :)</p>",
      "rawMarkdown": "Hi,\n\n\nI was just thinking can we take this as a 2 part problem in which let's say for part 1 we try to generate captions for all images and in the part 2 we try to identify the closest caption to what we generated? I know it will be complex since out captions will be generic but we can try to identify those generic things in the wikipedia article? does this seem doable?\n\nAny thoughts :)",
      "votes": null
    },
    {
      "id": "1513621",
      "postDate": "09/15/2021 09:23:39",
      "content": "<p>Hi <a href=\"https://www.kaggle.com/udbhavpangotra\" target=\"_blank\">@udbhavpangotra</a> - thank you for your interest in this competition. This sounds like an amazing option, and we would love to see the raw caption text generated by your models in part 1!</p>\n<blockquote>\n  <p>I was just thinking can we take this as a 2 part problem in which let's say for part 1 we try to generate captions for all images and in the part 2 we try to identify the closest caption to what we generated? I know it will be complex since out captions will be generic but we can try to identify those generic things in the wikipedia article? does this seem doable?</p>\n  <p>Any thoughts :)</p>\n</blockquote>",
      "rawMarkdown": "Hi @udbhavpangotra - thank you for your interest in this competition. This sounds like an amazing option, and we would love to see the raw caption text generated by your models in part 1!\n\n> \n> I was just thinking can we take this as a 2 part problem in which let's say for part 1 we try to generate captions for all images and in the part 2 we try to identify the closest caption to what we generated? I know it will be complex since out captions will be generic but we can try to identify those generic things in the wikipedia article? does this seem doable?\n> \n> Any thoughts :)",
      "votes": null
    },
    {
      "id": "1537726",
      "postDate": "10/07/2021 17:16:31",
      "content": "<p>I have also thought about it) Have you tried it?</p>",
      "rawMarkdown": "I have also thought about it) Have you tried it?",
      "votes": null
    },
    {
      "id": "1538197",
      "postDate": "10/08/2021 06:47:53",
      "content": "<p><a href=\"https://www.kaggle.com/vasyka\" target=\"_blank\">@vasyka</a> , not successfully :( <br>\nkernel for reference - <br>\n<a href=\"https://www.kaggle.com/udbhavpangotra/image-caption-using-tf\" target=\"_blank\">https://www.kaggle.com/udbhavpangotra/image-caption-using-tf</a></p>",
      "rawMarkdown": "vasyka , not successfully :( \nkernel for reference - \nhttps://www.kaggle.com/udbhavpangotra/image-caption-using-tf",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1513621,
      "author_name": "miriamredi",
      "author_url": "",
      "post_date": "09/15/2021 09:23:39",
      "content": "<p>Hi <a href=\"https://www.kaggle.com/udbhavpangotra\" target=\"_blank\">@udbhavpangotra</a> - thank you for your interest in this competition. This sounds like an amazing option, and we would love to see the raw caption text generated by your models in part 1!</p>\n<blockquote>\n  <p>I was just thinking can we take this as a 2 part problem in which let's say for part 1 we try to generate captions for all images and in the part 2 we try to identify the closest caption to what we generated? I know it will be complex since out captions will be generic but we can try to identify those generic things in the wikipedia article? does this seem doable?</p>\n  <p>Any thoughts :)</p>\n</blockquote>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1537726,
      "author_name": "vasyka",
      "author_url": "",
      "post_date": "10/07/2021 17:16:31",
      "content": "<p>I have also thought about it) Have you tried it?</p>",
      "votes": null,
      "replies": [
        {
          "id": 1538197,
          "author_name": "udbhavpangotra",
          "author_url": "",
          "post_date": "10/08/2021 06:47:53",
          "content": "<p><a href=\"https://www.kaggle.com/vasyka\" target=\"_blank\">@vasyka</a> , not successfully :( <br>\nkernel for reference - <br>\n<a href=\"https://www.kaggle.com/udbhavpangotra/image-caption-using-tf\" target=\"_blank\">https://www.kaggle.com/udbhavpangotra/image-caption-using-tf</a></p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1512832": "Hi,\n\n\nI was just thinking can we take this as a 2 part problem in which let's say for part 1 we try to generate captions for all images and in the part 2 we try to identify the closest caption to what we generated? I know it will be complex since out captions will be generic but we can try to identify those generic things in the wikipedia article? does this seem doable?\n\nAny thoughts :)",
    "1513621": "Hi @udbhavpangotra - thank you for your interest in this competition. This sounds like an amazing option, and we would love to see the raw caption text generated by your models in part 1!\n\n> \n> I was just thinking can we take this as a 2 part problem in which let's say for part 1 we try to generate captions for all images and in the part 2 we try to identify the closest caption to what we generated? I know it will be complex since out captions will be generic but we can try to identify those generic things in the wikipedia article? does this seem doable?\n> \n> Any thoughts :)",
    "1537726": "I have also thought about it) Have you tried it?",
    "1538197": "vasyka , not successfully :( \nkernel for reference - \nhttps://www.kaggle.com/udbhavpangotra/image-caption-using-tf"
  },
  "source": "meta"
}