{
  "id": 679129,
  "title": "About using data",
  "url": "/competitions/cvpr-mslr-2026-track-2/discussion/679129",
  "author_name": "",
  "post_date": "2026-02-27T09:56:42.245199600Z",
  "votes": null,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Hi organizers,\nSince the 2026 challenge is built around the MultiMeDaLIS dataset, is it permissible to use the publicly available RGB and RDM data from the 2025 version of the same dataset for training purposes only?\nWe understand that: \n    1. The leaderboard is Radar-Only.\n    2. Our final model will strictly take only RTM files as input during inference.\nHowever, we intend to use the 2025 multimodal data as auxiliary information to improve the feature representation of our radar-only model. Does this approach comply with the \"External Data\" policy, provided the data is publicly accessible to all? \nThank you for your guidance.</p>",
  "messages": [
    {
      "id": "3414626",
      "postDate": "02/27/2026 09:56:42",
      "content": "<p>Hi organizers,\nSince the 2026 challenge is built around the MultiMeDaLIS dataset, is it permissible to use the publicly available RGB and RDM data from the 2025 version of the same dataset for training purposes only?\nWe understand that: \n    1. The leaderboard is Radar-Only.\n    2. Our final model will strictly take only RTM files as input during inference.\nHowever, we intend to use the 2025 multimodal data as auxiliary information to improve the feature representation of our radar-only model. Does this approach comply with the \"External Data\" policy, provided the data is publicly accessible to all? \nThank you for your guidance.</p>",
      "rawMarkdown": "Hi organizers,\nSince the 2026 challenge is built around the MultiMeDaLIS dataset, is it permissible to use the publicly available RGB and RDM data from the 2025 version of the same dataset for training purposes only?\nWe understand that: \n    1. The leaderboard is Radar-Only.\n    2. Our final model will strictly take only RTM files as input during inference.\nHowever, we intend to use the 2025 multimodal data as auxiliary information to improve the feature representation of our radar-only model. Does this approach comply with the \"External Data\" policy, provided the data is publicly accessible to all? \nThank you for your guidance.",
      "votes": null
    },
    {
      "id": "3414633",
      "postDate": "02/27/2026 10:40:12",
      "content": "<p>Thank you for raising this point: <strong>using the 2025 MultiMeDaLIS RGB or RDM data is not allowed</strong> in this challenge.</p>\n<p>Although external public datasets for generic pretraining may be acceptable, the 2025 RGB/RDM data belongs to the same dataset family as the current competition. Using those modalities during training would introduce multimodal supervision directly linked to the challenge dataset, which is not consistent with the intended Radar-Only setting.</p>\n<p>Moreover, sessions and class distributions were intentionally reorganized across releases to prevent cross-release matching or indirect exploitation of related data.</p>",
      "rawMarkdown": "Thank you for raising this point: **using the 2025 MultiMeDaLIS RGB or RDM data is not allowed** in this challenge.\n\nAlthough external public datasets for generic pretraining may be acceptable, the 2025 RGB/RDM data belongs to the same dataset family as the current competition. Using those modalities during training would introduce multimodal supervision directly linked to the challenge dataset, which is not consistent with the intended Radar-Only setting.\n\nMoreover, sessions and class distributions were intentionally reorganized across releases to prevent cross-release matching or indirect exploitation of related data.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 3414633,
      "author_name": "raffaelemineo",
      "author_url": "",
      "post_date": "02/27/2026 10:40:12",
      "content": "<p>Thank you for raising this point: <strong>using the 2025 MultiMeDaLIS RGB or RDM data is not allowed</strong> in this challenge.</p>\n<p>Although external public datasets for generic pretraining may be acceptable, the 2025 RGB/RDM data belongs to the same dataset family as the current competition. Using those modalities during training would introduce multimodal supervision directly linked to the challenge dataset, which is not consistent with the intended Radar-Only setting.</p>\n<p>Moreover, sessions and class distributions were intentionally reorganized across releases to prevent cross-release matching or indirect exploitation of related data.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "3414626": "Hi organizers,\nSince the 2026 challenge is built around the MultiMeDaLIS dataset, is it permissible to use the publicly available RGB and RDM data from the 2025 version of the same dataset for training purposes only?\nWe understand that: \n    1. The leaderboard is Radar-Only.\n    2. Our final model will strictly take only RTM files as input during inference.\nHowever, we intend to use the 2025 multimodal data as auxiliary information to improve the feature representation of our radar-only model. Does this approach comply with the \"External Data\" policy, provided the data is publicly accessible to all? \nThank you for your guidance.",
    "3414633": "Thank you for raising this point: **using the 2025 MultiMeDaLIS RGB or RDM data is not allowed** in this challenge.\n\nAlthough external public datasets for generic pretraining may be acceptable, the 2025 RGB/RDM data belongs to the same dataset family as the current competition. Using those modalities during training would introduce multimodal supervision directly linked to the challenge dataset, which is not consistent with the intended Radar-Only setting.\n\nMoreover, sessions and class distributions were intentionally reorganized across releases to prevent cross-release matching or indirect exploitation of related data."
  },
  "source": "meta"
}