{
  "id": 353949,
  "title": "For those who used tiling, how do you tile the test data?",
  "url": "/competitions/hubmap-organ-segmentation/discussion/353949",
  "author_name": "",
  "post_date": "2022-09-20T13:20:47.166916600Z",
  "votes": 1,
  "comment_count": 1,
  "views": 0,
  "content": "<p>The training images are 3000x3000, and you tile it into e.g. 9 tiles.</p>\n<p>For the test images, some are small sized e.g. 160x160. How do you make predictions on these small images?<br>\nDo you resize it to 3000x3000 then tile it, and then resize back to 160x160? What to do about the paddings when you tile as well?</p>\n<p>Any help will be appreciated!</p>",
  "messages": [
    {
      "id": "1947447",
      "postDate": "09/20/2022 13:20:47",
      "content": "<p>The training images are 3000x3000, and you tile it into e.g. 9 tiles.</p>\n<p>For the test images, some are small sized e.g. 160x160. How do you make predictions on these small images?<br>\nDo you resize it to 3000x3000 then tile it, and then resize back to 160x160? What to do about the paddings when you tile as well?</p>\n<p>Any help will be appreciated!</p>",
      "rawMarkdown": "The training images are 3000x3000, and you tile it into e.g. 9 tiles.\n\nFor the test images, some are small sized e.g. 160x160. How do you make predictions on these small images?\nDo you resize it to 3000x3000 then tile it, and then resize back to 160x160? What to do about the paddings when you tile as well?\n\nAny help will be appreciated!",
      "votes": null
    },
    {
      "id": "1948188",
      "postDate": "09/20/2022 22:45:54",
      "content": "<p>For example, I added some padding to 160size image to be 256(tile size) and predicted its mask. And then I removed the padding as well as adjusted the mask position. Also, you can upsize 160size image to 256 and predict, after that, downsize it back to 160.  I used <a href=\"https://www.kaggle.com/code/thedevastator/converting-to-256x256\" target=\"_blank\">this code</a> to pad: you can check the detail in <code>class HuBMAPDataset(Dataset)</code></p>",
      "rawMarkdown": "For example, I added some padding to 160size image to be 256(tile size) and predicted its mask. And then I removed the padding as well as adjusted the mask position. Also, you can upsize 160size image to 256 and predict, after that, downsize it back to 160.  I used [this code](https://www.kaggle.com/code/thedevastator/converting-to-256x256) to pad: you can check the detail in ```class HuBMAPDataset(Dataset)```",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1948188,
      "author_name": "cheulkay",
      "author_url": "",
      "post_date": "09/20/2022 22:45:54",
      "content": "<p>For example, I added some padding to 160size image to be 256(tile size) and predicted its mask. And then I removed the padding as well as adjusted the mask position. Also, you can upsize 160size image to 256 and predict, after that, downsize it back to 160.  I used <a href=\"https://www.kaggle.com/code/thedevastator/converting-to-256x256\" target=\"_blank\">this code</a> to pad: you can check the detail in <code>class HuBMAPDataset(Dataset)</code></p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1947447": "The training images are 3000x3000, and you tile it into e.g. 9 tiles.\n\nFor the test images, some are small sized e.g. 160x160. How do you make predictions on these small images?\nDo you resize it to 3000x3000 then tile it, and then resize back to 160x160? What to do about the paddings when you tile as well?\n\nAny help will be appreciated!",
    "1948188": "For example, I added some padding to 160size image to be 256(tile size) and predicted its mask. And then I removed the padding as well as adjusted the mask position. Also, you can upsize 160size image to 256 and predict, after that, downsize it back to 160.  I used [this code](https://www.kaggle.com/code/thedevastator/converting-to-256x256) to pad: you can check the detail in ```class HuBMAPDataset(Dataset)```"
  },
  "source": "meta"
}