{
  "id": 583945,
  "title": "Are we expected to use IMU data only? ",
  "url": "/competitions/cmi-detect-behavior-with-sensor-data/discussion/583945",
  "author_name": "",
  "post_date": "2025-06-10T13:58:13.751955Z",
  "votes": null,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Hello everyone,</p>\n<p>I'm seeking some clarification regarding the dataset composition and its implications for model development. I've seen some discussion suggesting that the test data might be split, with one half containing only IMU data and the other half including all available sensor data.</p>\n<p>Could the competition organizers please confirm whether Time-of-Flight (ToF) and Thermal (Thm) data will be consistently present across the full test set, or if there will indeed be a portion of the test data where only IMU measurements are provided?</p>\n<p>Are you guys  developing models that can primarily leverage IMU data, or should we use ToF and Thm data as well?</p>",
  "messages": [
    {
      "id": "3221171",
      "postDate": "06/10/2025 13:58:13",
      "content": "<p>Hello everyone,</p>\n<p>I'm seeking some clarification regarding the dataset composition and its implications for model development. I've seen some discussion suggesting that the test data might be split, with one half containing only IMU data and the other half including all available sensor data.</p>\n<p>Could the competition organizers please confirm whether Time-of-Flight (ToF) and Thermal (Thm) data will be consistently present across the full test set, or if there will indeed be a portion of the test data where only IMU measurements are provided?</p>\n<p>Are you guys  developing models that can primarily leverage IMU data, or should we use ToF and Thm data as well?</p>",
      "rawMarkdown": "Hello everyone,\n\nI'm seeking some clarification regarding the dataset composition and its implications for model development. I've seen some discussion suggesting that the test data might be split, with one half containing only IMU data and the other half including all available sensor data.\n\nCould the competition organizers please confirm whether Time-of-Flight (ToF) and Thermal (Thm) data will be consistently present across the full test set, or if there will indeed be a portion of the test data where only IMU measurements are provided?\n\nAre you guys  developing models that can primarily leverage IMU data, or should we use ToF and Thm data as well?",
      "votes": null
    },
    {
      "id": "3222310",
      "postDate": "06/12/2025 05:47:28",
      "content": "<p>Yes, it is correct that approximately 50% of the test data will only provide IMU measurements; this is shown in the Dataset Description. Currently, the strategy I'm using is to train two separate models: one that uses only IMU data and another that uses all available data. During inference, I check the data to decide which model to use. Hope this helps!</p>",
      "rawMarkdown": "Yes, it is correct that approximately 50% of the test data will only provide IMU measurements; this is shown in the Dataset Description. Currently, the strategy I'm using is to train two separate models: one that uses only IMU data and another that uses all available data. During inference, I check the data to decide which model to use. Hope this helps!",
      "votes": null
    },
    {
      "id": "3226032",
      "postDate": "06/17/2025 07:02:27",
      "content": "<p><a href=\"https://www.kaggle.com/co000l\" target=\"_blank\">@co000l</a> <br>\nHow do you deal with data which provide only IMU measurements when you want to build models using all available data?</p>",
      "rawMarkdown": "co000l \nHow do you deal with data which provide only IMU measurements when you want to build models using all available data?",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 3222310,
      "author_name": "co000l",
      "author_url": "",
      "post_date": "06/12/2025 05:47:28",
      "content": "<p>Yes, it is correct that approximately 50% of the test data will only provide IMU measurements; this is shown in the Dataset Description. Currently, the strategy I'm using is to train two separate models: one that uses only IMU data and another that uses all available data. During inference, I check the data to decide which model to use. Hope this helps!</p>",
      "votes": null,
      "replies": [
        {
          "id": 3226032,
          "author_name": "shanyun",
          "author_url": "",
          "post_date": "06/17/2025 07:02:27",
          "content": "<p><a href=\"https://www.kaggle.com/co000l\" target=\"_blank\">@co000l</a> <br>\nHow do you deal with data which provide only IMU measurements when you want to build models using all available data?</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "3221171": "Hello everyone,\n\nI'm seeking some clarification regarding the dataset composition and its implications for model development. I've seen some discussion suggesting that the test data might be split, with one half containing only IMU data and the other half including all available sensor data.\n\nCould the competition organizers please confirm whether Time-of-Flight (ToF) and Thermal (Thm) data will be consistently present across the full test set, or if there will indeed be a portion of the test data where only IMU measurements are provided?\n\nAre you guys  developing models that can primarily leverage IMU data, or should we use ToF and Thm data as well?",
    "3222310": "Yes, it is correct that approximately 50% of the test data will only provide IMU measurements; this is shown in the Dataset Description. Currently, the strategy I'm using is to train two separate models: one that uses only IMU data and another that uses all available data. During inference, I check the data to decide which model to use. Hope this helps!",
    "3226032": "co000l \nHow do you deal with data which provide only IMU measurements when you want to build models using all available data?"
  },
  "source": "meta"
}