{
  "id": 573738,
  "title": "Number of epochs in yolo model training",
  "url": "/competitions/byu-locating-bacterial-flagellar-motors-2025/discussion/573738",
  "author_name": "",
  "post_date": "2025-04-17T15:48:33.841174300Z",
  "votes": null,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Number of epochs in model training for approximately 20,000 training data points. What number is needed to avoid both overfitting and making the model too simple?</p>",
  "messages": [
    {
      "id": "3181230",
      "postDate": "04/17/2025 15:48:33",
      "content": "<p>Number of epochs in model training for approximately 20,000 training data points. What number is needed to avoid both overfitting and making the model too simple?</p>",
      "rawMarkdown": "Number of epochs in model training for approximately 20,000 training data points. What number is needed to avoid both overfitting and making the model too simple?",
      "votes": null
    },
    {
      "id": "3181263",
      "postDate": "04/17/2025 16:37:27",
      "content": "<p>It's hard! The <code>COCO dataset has a total of 330,000 images</code>, and <code>over 200,000 of them are labeled</code> for object detection. </p>\n<p>Models are usually trained for 100 to 300 epochs. So the key point is that you have to constantly monitor whether the model is <code>overfitting</code>. </p>\n<p>That means you need to understand the loss values. An easy way to help with that is to set patience=5 or 8, which is generally optimal. Also, there are many parameters to tune, like <code>lr0</code> (initial learning rate) to <code>lr_final</code> (final learning rate)!</p>",
      "rawMarkdown": "It's hard! The `COCO dataset has a total of 330,000 images`, and `over 200,000 of them are labeled` for object detection. \n\nModels are usually trained for 100 to 300 epochs. So the key point is that you have to constantly monitor whether the model is `overfitting`. \n\nThat means you need to understand the loss values. An easy way to help with that is to set patience=5 or 8, which is generally optimal. Also, there are many parameters to tune, like `lr0` (initial learning rate) to `lr_final` (final learning rate)!",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 3181263,
      "author_name": "sangrampatil5150",
      "author_url": "",
      "post_date": "04/17/2025 16:37:27",
      "content": "<p>It's hard! The <code>COCO dataset has a total of 330,000 images</code>, and <code>over 200,000 of them are labeled</code> for object detection. </p>\n<p>Models are usually trained for 100 to 300 epochs. So the key point is that you have to constantly monitor whether the model is <code>overfitting</code>. </p>\n<p>That means you need to understand the loss values. An easy way to help with that is to set patience=5 or 8, which is generally optimal. Also, there are many parameters to tune, like <code>lr0</code> (initial learning rate) to <code>lr_final</code> (final learning rate)!</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "3181230": "Number of epochs in model training for approximately 20,000 training data points. What number is needed to avoid both overfitting and making the model too simple?",
    "3181263": "It's hard! The `COCO dataset has a total of 330,000 images`, and `over 200,000 of them are labeled` for object detection. \n\nModels are usually trained for 100 to 300 epochs. So the key point is that you have to constantly monitor whether the model is `overfitting`. \n\nThat means you need to understand the loss values. An easy way to help with that is to set patience=5 or 8, which is generally optimal. Also, there are many parameters to tune, like `lr0` (initial learning rate) to `lr_final` (final learning rate)!"
  },
  "source": "meta"
}