{
  "id": 685562,
  "title": "Question about Partial Annotation for Ablation Study and Accident Type Definition",
  "url": "/competitions/accident/discussion/685562",
  "author_name": "",
  "post_date": "2026-03-28T15:16:03.223775400Z",
  "votes": null,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Hello,</p>\n<p>I have a couple of questions regarding the competition:</p>\n<ol>\n<li><p>Due to the long inference time of our current pipeline, we are considering manually annotating a subset of around 200 samples and using them for evaluation in our paper (e.g., for ablation studies and comparative analysis). Would this be acceptable, or are there any restrictions on using a partially labeled subset for research purposes?</p></li>\n<li><p>For accident type classification, how should we categorize a case where a moving vehicle collides with a parked vehicle? Should this be considered a t-bone collision or a single accident?</p></li>\n</ol>\n<p>Thank you for your clarification!</p>",
  "messages": [
    {
      "id": "3430493",
      "postDate": "03/28/2026 15:16:03",
      "content": "<p>Hello,</p>\n<p>I have a couple of questions regarding the competition:</p>\n<ol>\n<li><p>Due to the long inference time of our current pipeline, we are considering manually annotating a subset of around 200 samples and using them for evaluation in our paper (e.g., for ablation studies and comparative analysis). Would this be acceptable, or are there any restrictions on using a partially labeled subset for research purposes?</p></li>\n<li><p>For accident type classification, how should we categorize a case where a moving vehicle collides with a parked vehicle? Should this be considered a t-bone collision or a single accident?</p></li>\n</ol>\n<p>Thank you for your clarification!</p>",
      "rawMarkdown": "Hello,\n\nI have a couple of questions regarding the competition:\n\n1. Due to the long inference time of our current pipeline, we are considering manually annotating a subset of around 200 samples and using them for evaluation in our paper (e.g., for ablation studies and comparative analysis). Would this be acceptable, or are there any restrictions on using a partially labeled subset for research purposes?\n\n2. For accident type classification, how should we categorize a case where a moving vehicle collides with a parked vehicle? Should this be considered a t-bone collision or a single accident?\n\nThank you for your clarification!",
      "votes": null
    },
    {
      "id": "3430633",
      "postDate": "03/28/2026 18:24:00",
      "content": "<p>Hi <a href=\"https://www.kaggle.com/tjrhyun\" target=\"_blank\">@tjrhyun</a> ,</p>\n<p>Please note that <strong>this is a zero-shot focused competition</strong>, so manually annotating any portion of the competition test set would be contrary to both the rules and the intended evaluation setting. This applies even if the annotations are used only on a subset and only for analysis in a paper.</p>\n<p>I would like to stress that participants are expected to comply with the competition requirements, including the code submission requirements where applicable. Entries that do not satisfy these requirements, or otherwise violate the competition rules, may be removed from the leaderboard.</p>\n<p>Regarding accident type classification, if two vehicles are involved, it is never a single vehicle.</p>\n<p>Best,\nLukas</p>",
      "rawMarkdown": "Hi @tjrhyun ,\n\nPlease note that **this is a zero-shot focused competition**, so manually annotating any portion of the competition test set would be contrary to both the rules and the intended evaluation setting. This applies even if the annotations are used only on a subset and only for analysis in a paper.\n\nI would like to stress that participants are expected to comply with the competition requirements, including the code submission requirements where applicable. Entries that do not satisfy these requirements, or otherwise violate the competition rules, may be removed from the leaderboard.\n\nRegarding accident type classification, if two vehicles are involved, it is never a single vehicle.\n\nBest,\nLukas",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 3430633,
      "author_name": "picekl",
      "author_url": "",
      "post_date": "03/28/2026 18:24:00",
      "content": "<p>Hi <a href=\"https://www.kaggle.com/tjrhyun\" target=\"_blank\">@tjrhyun</a> ,</p>\n<p>Please note that <strong>this is a zero-shot focused competition</strong>, so manually annotating any portion of the competition test set would be contrary to both the rules and the intended evaluation setting. This applies even if the annotations are used only on a subset and only for analysis in a paper.</p>\n<p>I would like to stress that participants are expected to comply with the competition requirements, including the code submission requirements where applicable. Entries that do not satisfy these requirements, or otherwise violate the competition rules, may be removed from the leaderboard.</p>\n<p>Regarding accident type classification, if two vehicles are involved, it is never a single vehicle.</p>\n<p>Best,\nLukas</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "3430493": "Hello,\n\nI have a couple of questions regarding the competition:\n\n1. Due to the long inference time of our current pipeline, we are considering manually annotating a subset of around 200 samples and using them for evaluation in our paper (e.g., for ablation studies and comparative analysis). Would this be acceptable, or are there any restrictions on using a partially labeled subset for research purposes?\n\n2. For accident type classification, how should we categorize a case where a moving vehicle collides with a parked vehicle? Should this be considered a t-bone collision or a single accident?\n\nThank you for your clarification!",
    "3430633": "Hi @tjrhyun ,\n\nPlease note that **this is a zero-shot focused competition**, so manually annotating any portion of the competition test set would be contrary to both the rules and the intended evaluation setting. This applies even if the annotations are used only on a subset and only for analysis in a paper.\n\nI would like to stress that participants are expected to comply with the competition requirements, including the code submission requirements where applicable. Entries that do not satisfy these requirements, or otherwise violate the competition rules, may be removed from the leaderboard.\n\nRegarding accident type classification, if two vehicles are involved, it is never a single vehicle.\n\nBest,\nLukas"
  },
  "source": "meta"
}