{
  "id": 567953,
  "title": " Impact of Objects Number in Training and Testing Data on YOLO Training",
  "url": "/competitions/byu-locating-bacterial-flagellar-motors-2025/discussion/567953",
  "author_name": "Leo Yang",
  "post_date": "2025-03-13T02:17:37.054000",
  "votes": 6,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Hi Kagglers, </p>\n<p>The training data includes images with 0, 1, or more objects. However, the testing data will only contain images with only 1 or 0 objects. I'm concerned about how this discrepancy might influence the performance of my YOLO model.</p>\n<p>For example, if I remove all training data with more objects, will it improve the performance on testing data ? I assume it will lower the chance to give 2 or more detect object, is it right?</p>\n<p>thanks for any opinions.</p>",
  "messages": [
    {
      "id": 3149375,
      "postDate": "2025-03-14T05:47:53.907Z",
      "content": "<p>It would be better to use all training data. You can select the highest-confidence output, given that the test data only contains images with 0 or 1 object during inference. In this case, NMS would be unnecessary as well.</p>",
      "rawMarkdown": "It would be better to use all training data. You can select the highest-confidence output, given that the test data only contains images with 0 or 1 object during inference. In this case, NMS would be unnecessary as well.",
      "votes": 6
    },
    {
      "id": 3148313,
      "postDate": "2025-03-13T02:17:37.053Z",
      "content": "<p>Hi Kagglers, </p>\n<p>The training data includes images with 0, 1, or more objects. However, the testing data will only contain images with only 1 or 0 objects. I'm concerned about how this discrepancy might influence the performance of my YOLO model.</p>\n<p>For example, if I remove all training data with more objects, will it improve the performance on testing data ? I assume it will lower the chance to give 2 or more detect object, is it right?</p>\n<p>thanks for any opinions.</p>",
      "rawMarkdown": "Hi Kagglers, \n\nThe training data includes images with 0, 1, or more objects. However, the testing data will only contain images with only 1 or 0 objects. I'm concerned about how this discrepancy might influence the performance of my YOLO model.\n\nFor example, if I remove all training data with more objects, will it improve the performance on testing data ? I assume it will lower the chance to give 2 or more detect object, is it right?\n\nthanks for any opinions.\n\n\n\n",
      "votes": 6
    },
    {
      "id": 3148737,
      "postDate": "2025-03-13T13:26:19.090Z",
      "rawMarkdown": "",
      "isDeleted": true
    }
  ],
  "comments": [
    {
      "id": 3149375,
      "author_name": "I really don't like crustle",
      "author_url": "",
      "post_date": "2025-03-14T05:47:53.907000",
      "content": "<p>It would be better to use all training data. You can select the highest-confidence output, given that the test data only contains images with 0 or 1 object during inference. In this case, NMS would be unnecessary as well.</p>",
      "votes": 6,
      "replies": []
    },
    {
      "id": 3148737,
      "author_name": "",
      "author_url": "",
      "post_date": "2025-03-13T13:26:19.090000",
      "content": "",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "3149375": "It would be better to use all training data. You can select the highest-confidence output, given that the test data only contains images with 0 or 1 object during inference. In this case, NMS would be unnecessary as well.",
    "3148313": "Hi Kagglers, \n\nThe training data includes images with 0, 1, or more objects. However, the testing data will only contain images with only 1 or 0 objects. I'm concerned about how this discrepancy might influence the performance of my YOLO model.\n\nFor example, if I remove all training data with more objects, will it improve the performance on testing data ? I assume it will lower the chance to give 2 or more detect object, is it right?\n\nthanks for any opinions.\n\n\n\n",
    "3148737": ""
  }
}