{
  "id": 54134,
  "title": "MaskRCNN for Nuclei Detection on Google Colab (LB=0.382)",
  "url": "/competitions/data-science-bowl-2018/discussion/54134",
  "author_name": "aftab khan",
  "post_date": "2018-04-10T06:01:54.554000",
  "votes": 8,
  "comment_count": 0,
  "views": 0,
  "content": "<p><a href=\"https://github.com/MLKng/KaggleDSBowl2018\">https://github.com/MLKng/KaggleDSBowl2018</a></p>\n\n<p>The above link is to a Jupyter Notebook which implements the following:</p>\n\n<ol>\n<li>MaskRCNN implementation at <a href=\"https://github.com/killthekitten/kaggle-ds-bowl-2018-baseline#kaggle-ds-bowl-2018\">https://github.com/killthekitten/kaggle-ds-bowl-2018-baseline#kaggle-ds-bowl-2018</a> in Jupyter Notebook form.\n<ul><li>No change in code, only changed modules train.py and inference.py into procedures so they can be run within this notebook and take parameters</li>\n<li>Using this code I made 6 submissions after every 5 epochs going upto 35 epochs (5, 10, 15, 20, 25, 30, 35). Scores increased on LB until 20 epochs, then went down (probably overfitting). Max LB score at 20 epochs = 0.382</li></ul></li>\n<li>Be able to run this notebook in Google Collaboratory (<a href=\"https://colab.research.google.com/\">https://colab.research.google.com/</a>) and utilize the free GPU support. (Takes an average 10 minutes to run 1 epoch with GPU)\n<ul><li>Loads all the competition data from Kaggle into Google Colab (the virtual environment allocated to the user)</li>\n<li>Loads the model weights as required by item 1 above into Google Colab</li>\n<li>Loads the code from github for item 1 implementation</li>\n<li>Can make competition submissions to Kaggle site directly from Google colab</li>\n<li>After every 5 epochs during RCNN training uploads h5 weights file to your Google drive</li></ul></li>\n</ol>",
  "messages": [
    {
      "id": 311475,
      "postDate": "2018-04-10T06:01:54.553Z",
      "content": "<p><a href=\"https://github.com/MLKng/KaggleDSBowl2018\">https://github.com/MLKng/KaggleDSBowl2018</a></p>\n\n<p>The above link is to a Jupyter Notebook which implements the following:</p>\n\n<ol>\n<li>MaskRCNN implementation at <a href=\"https://github.com/killthekitten/kaggle-ds-bowl-2018-baseline#kaggle-ds-bowl-2018\">https://github.com/killthekitten/kaggle-ds-bowl-2018-baseline#kaggle-ds-bowl-2018</a> in Jupyter Notebook form.\n<ul><li>No change in code, only changed modules train.py and inference.py into procedures so they can be run within this notebook and take parameters</li>\n<li>Using this code I made 6 submissions after every 5 epochs going upto 35 epochs (5, 10, 15, 20, 25, 30, 35). Scores increased on LB until 20 epochs, then went down (probably overfitting). Max LB score at 20 epochs = 0.382</li></ul></li>\n<li>Be able to run this notebook in Google Collaboratory (<a href=\"https://colab.research.google.com/\">https://colab.research.google.com/</a>) and utilize the free GPU support. (Takes an average 10 minutes to run 1 epoch with GPU)\n<ul><li>Loads all the competition data from Kaggle into Google Colab (the virtual environment allocated to the user)</li>\n<li>Loads the model weights as required by item 1 above into Google Colab</li>\n<li>Loads the code from github for item 1 implementation</li>\n<li>Can make competition submissions to Kaggle site directly from Google colab</li>\n<li>After every 5 epochs during RCNN training uploads h5 weights file to your Google drive</li></ul></li>\n</ol>",
      "rawMarkdown": "https://github.com/MLKng/KaggleDSBowl2018\n\nThe above link is to a Jupyter Notebook which implements the following:\n\n 1. MaskRCNN implementation at https://github.com/killthekitten/kaggle-ds-bowl-2018-baseline#kaggle-ds-bowl-2018 in Jupyter Notebook form.\n  - No change in code, only changed modules train.py and inference.py into procedures so they can be run within this notebook and take parameters\n  - Using this code I made 6 submissions after every 5 epochs going upto 35 epochs (5, 10, 15, 20, 25, 30, 35). Scores increased on LB until 20 epochs, then went down (probably overfitting). Max LB score at 20 epochs = 0.382\n 2. Be able to run this notebook in Google Collaboratory (https://colab.research.google.com/) and utilize the free GPU support. (Takes an average 10 minutes to run 1 epoch with GPU)\n  - Loads all the competition data from Kaggle into Google Colab (the virtual environment allocated to the user)\n  - Loads the model weights as required by item 1 above into Google Colab\n - Loads the code from github for item 1 implementation\n - Can make competition submissions to Kaggle site directly from Google colab\n - After every 5 epochs during RCNN training uploads h5 weights file to your Google drive\n\n\n  [1]: https://github.com/MLKng/KaggleDSBowl2018",
      "votes": 8
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "311475": "https://github.com/MLKng/KaggleDSBowl2018\n\nThe above link is to a Jupyter Notebook which implements the following:\n\n 1. MaskRCNN implementation at https://github.com/killthekitten/kaggle-ds-bowl-2018-baseline#kaggle-ds-bowl-2018 in Jupyter Notebook form.\n  - No change in code, only changed modules train.py and inference.py into procedures so they can be run within this notebook and take parameters\n  - Using this code I made 6 submissions after every 5 epochs going upto 35 epochs (5, 10, 15, 20, 25, 30, 35). Scores increased on LB until 20 epochs, then went down (probably overfitting). Max LB score at 20 epochs = 0.382\n 2. Be able to run this notebook in Google Collaboratory (https://colab.research.google.com/) and utilize the free GPU support. (Takes an average 10 minutes to run 1 epoch with GPU)\n  - Loads all the competition data from Kaggle into Google Colab (the virtual environment allocated to the user)\n  - Loads the model weights as required by item 1 above into Google Colab\n - Loads the code from github for item 1 implementation\n - Can make competition submissions to Kaggle site directly from Google colab\n - After every 5 epochs during RCNN training uploads h5 weights file to your Google drive\n\n\n  [1]: https://github.com/MLKng/KaggleDSBowl2018"
  }
}