{
  "id": 285177,
  "title": "[Trainable] Pytorch Starter",
  "url": "/competitions/wikipedia-image-caption/discussion/285177",
  "author_name": "",
  "post_date": "2021-11-03T16:39:49.903055600Z",
  "votes": 7,
  "comment_count": 2,
  "views": 0,
  "content": "<p>I have published a completely trainable notebook for the Image/Caption matching challenge<br>\nNotebook: <a href=\"https://www.kaggle.com/debarshichanda/pytorch-wikipedia-image-caption-starter\" target=\"_blank\">Pytorch Image/Caption Starter</a><br>\n</p><hr><br>\nModel Architecture<br>\n<img src=\"https://i.imgur.com/PzooBWU.png\" alt=\"\"><br>\nAs of now the efficientnet backbone and xlm-backbone are not set to trainable and only the final layers are trainable but the entire network can be easily made trainable by not calling <code>self.freeze_backbone()</code><p></p>\n<p>I hope this helps!</p>",
  "messages": [
    {
      "id": "1569676",
      "postDate": "11/03/2021 16:39:49",
      "content": "<p>I have published a completely trainable notebook for the Image/Caption matching challenge<br>\nNotebook: <a href=\"https://www.kaggle.com/debarshichanda/pytorch-wikipedia-image-caption-starter\" target=\"_blank\">Pytorch Image/Caption Starter</a><br>\n</p><hr><br>\nModel Architecture<br>\n<img src=\"https://i.imgur.com/PzooBWU.png\" alt=\"\"><br>\nAs of now the efficientnet backbone and xlm-backbone are not set to trainable and only the final layers are trainable but the entire network can be easily made trainable by not calling <code>self.freeze_backbone()</code><p></p>\n<p>I hope this helps!</p>",
      "rawMarkdown": "I have published a completely trainable notebook for the Image/Caption matching challenge\nNotebook: [Pytorch Image/Caption Starter](https://www.kaggle.com/debarshichanda/pytorch-wikipedia-image-caption-starter)\n<hr>\nModel Architecture\n![](https://i.imgur.com/PzooBWU.png)\nAs of now the efficientnet backbone and xlm-backbone are not set to trainable and only the final layers are trainable but the entire network can be easily made trainable by not calling `self.freeze_backbone()`\n\nI hope this helps!",
      "votes": null
    },
    {
      "id": "1575751",
      "postDate": "11/08/2021 16:14:00",
      "content": "<p>Hi! this is a really good baseline actually! did you try maybe some different backbones to see how the results change?</p>",
      "rawMarkdown": "Hi! this is a really good baseline actually! did you try maybe some different backbones to see how the results change?",
      "votes": null
    },
    {
      "id": "1576649",
      "postDate": "11/09/2021 11:15:17",
      "content": "<p>I haven't really found time to experiment with different backbones as of now</p>",
      "rawMarkdown": "I haven't really found time to experiment with different backbones as of now",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1575751,
      "author_name": "",
      "author_url": "",
      "post_date": "11/08/2021 16:14:00",
      "content": "<p>Hi! this is a really good baseline actually! did you try maybe some different backbones to see how the results change?</p>",
      "votes": null,
      "replies": [
        {
          "id": 1576649,
          "author_name": "debarshichanda",
          "author_url": "",
          "post_date": "11/09/2021 11:15:17",
          "content": "<p>I haven't really found time to experiment with different backbones as of now</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1569676": "I have published a completely trainable notebook for the Image/Caption matching challenge\nNotebook: [Pytorch Image/Caption Starter](https://www.kaggle.com/debarshichanda/pytorch-wikipedia-image-caption-starter)\n<hr>\nModel Architecture\n![](https://i.imgur.com/PzooBWU.png)\nAs of now the efficientnet backbone and xlm-backbone are not set to trainable and only the final layers are trainable but the entire network can be easily made trainable by not calling `self.freeze_backbone()`\n\nI hope this helps!",
    "1575751": "Hi! this is a really good baseline actually! did you try maybe some different backbones to see how the results change?",
    "1576649": "I haven't really found time to experiment with different backbones as of now"
  },
  "source": "meta"
}