{
  "id": 500432,
  "title": "Use Transformer architecture to fuse multi-modal information",
  "url": "/competitions/geolifeclef-2024/discussion/500432",
  "author_name": "",
  "post_date": "2024-05-05T14:51:19.978183Z",
  "votes": 1,
  "comment_count": 1,
  "views": 0,
  "content": "<p>I tried to improve ViT based on the transformer architecture to convert image features at different positions that should have been input into features obtained from different dimensions. Judging from the results, it seems to be an effective attempt.</p>\n<p><a href=\"url\" target=\"_blank\">https://www.kaggle.com/code/liangyz2019/multimodal-transformer-inference</a></p>\n<p>I hope to get more opinions from everyone and improve this model together.😃</p>",
  "messages": [
    {
      "id": "2794868",
      "postDate": "05/05/2024 14:51:19",
      "content": "<p>I tried to improve ViT based on the transformer architecture to convert image features at different positions that should have been input into features obtained from different dimensions. Judging from the results, it seems to be an effective attempt.</p>\n<p><a href=\"url\" target=\"_blank\">https://www.kaggle.com/code/liangyz2019/multimodal-transformer-inference</a></p>\n<p>I hope to get more opinions from everyone and improve this model together.😃</p>",
      "rawMarkdown": "I tried to improve ViT based on the transformer architecture to convert image features at different positions that should have been input into features obtained from different dimensions. Judging from the results, it seems to be an effective attempt.\n\n[https://www.kaggle.com/code/liangyz2019/multimodal-transformer-inference](url)\n\nI hope to get more opinions from everyone and improve this model together.😃",
      "votes": null
    },
    {
      "id": "2795746",
      "postDate": "05/05/2024 23:30:49",
      "content": "<p>Well done. Good Job! This is what we were looking for. <br>\nHope you will inspire other teams and someone will surpass the 0.35 score mark :)</p>\n<p>Best,<br>\nLukas</p>",
      "rawMarkdown": "Well done. Good Job! This is what we were looking for. \nHope you will inspire other teams and someone will surpass the 0.35 score mark :)\n\nBest,\nLukas",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2795746,
      "author_name": "picekl",
      "author_url": "",
      "post_date": "05/05/2024 23:30:49",
      "content": "<p>Well done. Good Job! This is what we were looking for. <br>\nHope you will inspire other teams and someone will surpass the 0.35 score mark :)</p>\n<p>Best,<br>\nLukas</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2794868": "I tried to improve ViT based on the transformer architecture to convert image features at different positions that should have been input into features obtained from different dimensions. Judging from the results, it seems to be an effective attempt.\n\n[https://www.kaggle.com/code/liangyz2019/multimodal-transformer-inference](url)\n\nI hope to get more opinions from everyone and improve this model together.😃",
    "2795746": "Well done. Good Job! This is what we were looking for. \nHope you will inspire other teams and someone will surpass the 0.35 score mark :)\n\nBest,\nLukas"
  },
  "source": "meta"
}