{
  "id": 169634,
  "title": "Morphological features",
  "url": "/competitions/prostate-cancer-grade-assessment/discussion/169634",
  "author_name": "Arkajyoti Bhattacharya",
  "post_date": "2020-07-24T14:43:27.498000",
  "votes": 2,
  "comment_count": 0,
  "views": 0,
  "content": "<p>Congratulations to everyone and thanks to the organizers. I personally learned a lot from the public notebooks of this competition. </p>\n\n<p>I saw that majority of the models which did well in the competition are ensemble of pretrained models. We were trying to extract morphological features in the beginning of the competition and wished to add them in the stacked model. Due to lack of time we ended up not doing it anymore. With only those features in lightgbm, we obtained private LB of 0.38xx.</p>\n\n<p>The list of features used are the following:\n1. Number of white pixels in the image. \n2. Number of objects with certain area (with different cutoffs) using connected component analysis\n3. Compute the Local Binary Pattern representation of the image, and then use the LBP representation to build the histogram of patterns. Thereafter using entropy, skewness, kurtosis, mean, sd of the histogram corresponding to each image.</p>\n\n<p>Feel free to have a look at <a href=\"https://www.kaggle.com/arkajyotib/lbp-cca-features-tune-upsample-better-image-tiles\">this</a> kernel.</p>",
  "messages": [
    {
      "id": 943737,
      "postDate": "2020-07-24T14:43:27.500Z",
      "content": "<p>Congratulations to everyone and thanks to the organizers. I personally learned a lot from the public notebooks of this competition. </p>\n\n<p>I saw that majority of the models which did well in the competition are ensemble of pretrained models. We were trying to extract morphological features in the beginning of the competition and wished to add them in the stacked model. Due to lack of time we ended up not doing it anymore. With only those features in lightgbm, we obtained private LB of 0.38xx.</p>\n\n<p>The list of features used are the following:\n1. Number of white pixels in the image. \n2. Number of objects with certain area (with different cutoffs) using connected component analysis\n3. Compute the Local Binary Pattern representation of the image, and then use the LBP representation to build the histogram of patterns. Thereafter using entropy, skewness, kurtosis, mean, sd of the histogram corresponding to each image.</p>\n\n<p>Feel free to have a look at <a href=\"https://www.kaggle.com/arkajyotib/lbp-cca-features-tune-upsample-better-image-tiles\">this</a> kernel.</p>",
      "rawMarkdown": "Congratulations to everyone and thanks to the organizers. I personally learned a lot from the public notebooks of this competition. \n\nI saw that majority of the models which did well in the competition are ensemble of pretrained models. We were trying to extract morphological features in the beginning of the competition and wished to add them in the stacked model. Due to lack of time we ended up not doing it anymore. With only those features in lightgbm, we obtained private LB of 0.38xx.\n\nThe list of features used are the following:\n1. Number of white pixels in the image. \n2. Number of objects with certain area (with different cutoffs) using connected component analysis\n3. Compute the Local Binary Pattern representation of the image, and then use the LBP representation to build the histogram of patterns. Thereafter using entropy, skewness, kurtosis, mean, sd of the histogram corresponding to each image.\n\nFeel free to have a look at [this](https://www.kaggle.com/arkajyotib/lbp-cca-features-tune-upsample-better-image-tiles) kernel.\n",
      "votes": 2
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "943737": "Congratulations to everyone and thanks to the organizers. I personally learned a lot from the public notebooks of this competition. \n\nI saw that majority of the models which did well in the competition are ensemble of pretrained models. We were trying to extract morphological features in the beginning of the competition and wished to add them in the stacked model. Due to lack of time we ended up not doing it anymore. With only those features in lightgbm, we obtained private LB of 0.38xx.\n\nThe list of features used are the following:\n1. Number of white pixels in the image. \n2. Number of objects with certain area (with different cutoffs) using connected component analysis\n3. Compute the Local Binary Pattern representation of the image, and then use the LBP representation to build the histogram of patterns. Thereafter using entropy, skewness, kurtosis, mean, sd of the histogram corresponding to each image.\n\nFeel free to have a look at [this](https://www.kaggle.com/arkajyotib/lbp-cca-features-tune-upsample-better-image-tiles) kernel.\n"
  }
}