{
  "id": 351550,
  "title": "RLE Encode Question",
  "url": "/competitions/hubmap-organ-segmentation/discussion/351550",
  "author_name": "",
  "post_date": "2022-09-10T19:10:59.145472Z",
  "votes": 3,
  "comment_count": 3,
  "views": 0,
  "content": "<p>I know I am too late to ask this question. I am a beginner, and I made an Unet model, so I am resizing the input image to 1024,1024, and the output mask is 1016,1016. Can I use the same RLE decode function everyone is using or do I need to make some changes to it? How should I resize this 1016,1016 mask and make a submission?</p>",
  "messages": [
    {
      "id": "1933780",
      "postDate": "09/10/2022 19:10:59",
      "content": "<p>I know I am too late to ask this question. I am a beginner, and I made an Unet model, so I am resizing the input image to 1024,1024, and the output mask is 1016,1016. Can I use the same RLE decode function everyone is using or do I need to make some changes to it? How should I resize this 1016,1016 mask and make a submission?</p>",
      "rawMarkdown": "I know I am too late to ask this question. I am a beginner, and I made an Unet model, so I am resizing the input image to 1024,1024, and the output mask is 1016,1016. Can I use the same RLE decode function everyone is using or do I need to make some changes to it? How should I resize this 1016,1016 mask and make a submission?",
      "votes": null
    },
    {
      "id": "1933995",
      "postDate": "09/11/2022 03:46:47",
      "content": "<p><a href=\"https://www.kaggle.com/akshatdevve\" target=\"_blank\">@akshatdevve</a> You could resize predicted mask to the original image size by cv2 resizing or PIL Image resizing, and then make rle encoding on that resized mask. <a href=\"https://www.kaggle.com/paulorzp/rle-functions-run-lenght-encode-decode\" target=\"_blank\">Check rle encoding/deconding functions</a></p>",
      "rawMarkdown": "akshatdevve You could resize predicted mask to the original image size by cv2 resizing or PIL Image resizing, and then make rle encoding on that resized mask. [Check rle encoding/deconding functions](https://www.kaggle.com/paulorzp/rle-functions-run-lenght-encode-decode)",
      "votes": null
    },
    {
      "id": "1934822",
      "postDate": "09/11/2022 16:07:16",
      "content": "<p><a href=\"https://www.kaggle.com/electro\" target=\"_blank\">@electro</a> Thank you.</p>",
      "rawMarkdown": "electro Thank you.",
      "votes": null
    },
    {
      "id": "1935223",
      "postDate": "09/12/2022 00:29:34",
      "content": "<p>You mean mask prediction size is 1016, 1016 right?<br>\nFirst, you can check the conv2d, if use 3x3 conv2d padding = 0 will be reducing your mask prediction size.<br>\nAnd If you need resize your mask prediction, use interpolate. in pytorch, check torch.nn.functional.interpolate</p>",
      "rawMarkdown": "You mean mask prediction size is 1016, 1016 right?\nFirst, you can check the conv2d, if use 3x3 conv2d padding = 0 will be reducing your mask prediction size.\nAnd If you need resize your mask prediction, use interpolate. in pytorch, check torch.nn.functional.interpolate",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1933995,
      "author_name": "electro",
      "author_url": "",
      "post_date": "09/11/2022 03:46:47",
      "content": "<p><a href=\"https://www.kaggle.com/akshatdevve\" target=\"_blank\">@akshatdevve</a> You could resize predicted mask to the original image size by cv2 resizing or PIL Image resizing, and then make rle encoding on that resized mask. <a href=\"https://www.kaggle.com/paulorzp/rle-functions-run-lenght-encode-decode\" target=\"_blank\">Check rle encoding/deconding functions</a></p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1934822,
      "author_name": "akshatdevve",
      "author_url": "",
      "post_date": "09/11/2022 16:07:16",
      "content": "<p><a href=\"https://www.kaggle.com/electro\" target=\"_blank\">@electro</a> Thank you.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1935223,
      "author_name": "methyl",
      "author_url": "",
      "post_date": "09/12/2022 00:29:34",
      "content": "<p>You mean mask prediction size is 1016, 1016 right?<br>\nFirst, you can check the conv2d, if use 3x3 conv2d padding = 0 will be reducing your mask prediction size.<br>\nAnd If you need resize your mask prediction, use interpolate. in pytorch, check torch.nn.functional.interpolate</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1933780": "I know I am too late to ask this question. I am a beginner, and I made an Unet model, so I am resizing the input image to 1024,1024, and the output mask is 1016,1016. Can I use the same RLE decode function everyone is using or do I need to make some changes to it? How should I resize this 1016,1016 mask and make a submission?",
    "1933995": "akshatdevve You could resize predicted mask to the original image size by cv2 resizing or PIL Image resizing, and then make rle encoding on that resized mask. [Check rle encoding/deconding functions](https://www.kaggle.com/paulorzp/rle-functions-run-lenght-encode-decode)",
    "1934822": "electro Thank you.",
    "1935223": "You mean mask prediction size is 1016, 1016 right?\nFirst, you can check the conv2d, if use 3x3 conv2d padding = 0 will be reducing your mask prediction size.\nAnd If you need resize your mask prediction, use interpolate. in pytorch, check torch.nn.functional.interpolate"
  },
  "source": "meta"
}