{
  "id": 569604,
  "title": "External pre-trained models",
  "url": "/competitions/plantclef-2025/discussion/569604",
  "author_name": "",
  "post_date": "2025-03-22T22:08:12.788964900Z",
  "votes": null,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Are pre-trained models that not the ones provided permitted?</p>",
  "messages": [
    {
      "id": "3157053",
      "postDate": "03/22/2025 22:08:12",
      "content": "<p>Are pre-trained models that not the ones provided permitted?</p>",
      "rawMarkdown": "Are pre-trained models that not the ones provided permitted?",
      "votes": null
    },
    {
      "id": "3158063",
      "postDate": "03/24/2025 07:46:35",
      "content": "<p>Hello, yes of course we allow the use of other models. We've provided pre-trained models to save a little time if you want to start with the proposed weights and/or focus on the inference part, but I have no doubt that other approaches may be more relevant (large vit architecture instead of vit base to better manage the number of simultaneous species in an image? loss explicitly multi-label? Integration of the second dataset without labels, but visually closer to the test data? etc).</p>",
      "rawMarkdown": "Hello, yes of course we allow the use of other models. We've provided pre-trained models to save a little time if you want to start with the proposed weights and/or focus on the inference part, but I have no doubt that other approaches may be more relevant (large vit architecture instead of vit base to better manage the number of simultaneous species in an image? loss explicitly multi-label? Integration of the second dataset without labels, but visually closer to the test data? etc).",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 3158063,
      "author_name": "hgoeau",
      "author_url": "",
      "post_date": "03/24/2025 07:46:35",
      "content": "<p>Hello, yes of course we allow the use of other models. We've provided pre-trained models to save a little time if you want to start with the proposed weights and/or focus on the inference part, but I have no doubt that other approaches may be more relevant (large vit architecture instead of vit base to better manage the number of simultaneous species in an image? loss explicitly multi-label? Integration of the second dataset without labels, but visually closer to the test data? etc).</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "3157053": "Are pre-trained models that not the ones provided permitted?",
    "3158063": "Hello, yes of course we allow the use of other models. We've provided pre-trained models to save a little time if you want to start with the proposed weights and/or focus on the inference part, but I have no doubt that other approaches may be more relevant (large vit architecture instead of vit base to better manage the number of simultaneous species in an image? loss explicitly multi-label? Integration of the second dataset without labels, but visually closer to the test data? etc)."
  },
  "source": "meta"
}