{
  "id": 694474,
  "title": "Is normal get 0.00536 use the generate_preds_csv.py",
  "url": "/competitions/dishcovery-mission-ii-cvpr-2026/discussion/694474",
  "author_name": "",
  "post_date": "2026-04-25T02:21:25.852637400Z",
  "votes": null,
  "comment_count": 1,
  "views": 0,
  "content": "<p>I'm new to the competition, and I just use generate_preds_csv.py to get result, but the result is really low</p>",
  "messages": [
    {
      "id": "3448253",
      "postDate": "04/25/2026 02:21:25",
      "content": "<p>I'm new to the competition, and I just use generate_preds_csv.py to get result, but the result is really low</p>",
      "rawMarkdown": "I'm new to the competition, and I just use generate_preds_csv.py to get result, but the result is really low",
      "votes": null
    },
    {
      "id": "3448667",
      "postDate": "04/26/2026 10:32:51",
      "content": "<p>Welcome to the competition! <a href=\"https://www.kaggle.com/zxyccc\" target=\"_blank\">@zxyccc</a>  . The baseline script is only meant to demonstrate the correct output format, so its low score is expected and not something to worry about. To see a solid improvement, start by switching to a stronger CLIP-style model like SigLIP or OpenCLIP, as they’re far more capable than the demo model used. Even without complex tweaks, this one change can significantly boost your results and give you a much better starting point for further experimentation.</p>",
      "rawMarkdown": "Welcome to the competition! @zxyccc  . The baseline script is only meant to demonstrate the correct output format, so its low score is expected and not something to worry about. To see a solid improvement, start by switching to a stronger CLIP-style model like SigLIP or OpenCLIP, as they’re far more capable than the demo model used. Even without complex tweaks, this one change can significantly boost your results and give you a much better starting point for further experimentation.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 3448667,
      "author_name": "ignaneswara",
      "author_url": "",
      "post_date": "04/26/2026 10:32:51",
      "content": "<p>Welcome to the competition! <a href=\"https://www.kaggle.com/zxyccc\" target=\"_blank\">@zxyccc</a>  . The baseline script is only meant to demonstrate the correct output format, so its low score is expected and not something to worry about. To see a solid improvement, start by switching to a stronger CLIP-style model like SigLIP or OpenCLIP, as they’re far more capable than the demo model used. Even without complex tweaks, this one change can significantly boost your results and give you a much better starting point for further experimentation.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "3448253": "I'm new to the competition, and I just use generate_preds_csv.py to get result, but the result is really low",
    "3448667": "Welcome to the competition! @zxyccc  . The baseline script is only meant to demonstrate the correct output format, so its low score is expected and not something to worry about. To see a solid improvement, start by switching to a stronger CLIP-style model like SigLIP or OpenCLIP, as they’re far more capable than the demo model used. Even without complex tweaks, this one change can significantly boost your results and give you a much better starting point for further experimentation."
  },
  "source": "meta"
}