{
  "id": 346883,
  "title": "do patching image help the model to improve ? ",
  "url": "/competitions/hubmap-organ-segmentation/discussion/346883",
  "author_name": "",
  "post_date": "2022-08-21T21:24:18.220260500Z",
  "votes": 2,
  "comment_count": 1,
  "views": 0,
  "content": "<p>i want to patch the image and the mask into small size like 512x512 . does this help the model like unet improve ? </p>",
  "messages": [
    {
      "id": "1908655",
      "postDate": "08/21/2022 21:24:18",
      "content": "<p>i want to patch the image and the mask into small size like 512x512 . does this help the model like unet improve ? </p>",
      "rawMarkdown": "i want to patch the image and the mask into small size like 512x512 . does this help the model like unet improve ?",
      "votes": null
    },
    {
      "id": "1909445",
      "postDate": "08/22/2022 16:05:07",
      "content": "<p>Not really. Many other demonstrated that, unless you patch the image in very unusual ways (and I'm all hearing!), treating the images as-they-are provides the best results. You should pay attention instead to the pixel dimension between train and test set: in particular, in the training set, we have almost 10 times the dimension for certain classes! D: So you should pay attention to the resizing - and there are plenty of works in this sense. <br>\nBest luck!</p>",
      "rawMarkdown": "Not really. Many other demonstrated that, unless you patch the image in very unusual ways (and I'm all hearing!), treating the images as-they-are provides the best results. You should pay attention instead to the pixel dimension between train and test set: in particular, in the training set, we have almost 10 times the dimension for certain classes! D: So you should pay attention to the resizing - and there are plenty of works in this sense. \nBest luck!",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1909445,
      "author_name": "mawanda",
      "author_url": "",
      "post_date": "08/22/2022 16:05:07",
      "content": "<p>Not really. Many other demonstrated that, unless you patch the image in very unusual ways (and I'm all hearing!), treating the images as-they-are provides the best results. You should pay attention instead to the pixel dimension between train and test set: in particular, in the training set, we have almost 10 times the dimension for certain classes! D: So you should pay attention to the resizing - and there are plenty of works in this sense. <br>\nBest luck!</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1908655": "i want to patch the image and the mask into small size like 512x512 . does this help the model like unet improve ?",
    "1909445": "Not really. Many other demonstrated that, unless you patch the image in very unusual ways (and I'm all hearing!), treating the images as-they-are provides the best results. You should pay attention instead to the pixel dimension between train and test set: in particular, in the training set, we have almost 10 times the dimension for certain classes! D: So you should pay attention to the resizing - and there are plenty of works in this sense. \nBest luck!"
  },
  "source": "meta"
}