{
  "id": 269437,
  "title": "All [MRI TYPES] once is what u need",
  "url": "/competitions/rsna-miccai-brain-tumor-radiogenomic-classification/discussion/269437",
  "author_name": "",
  "post_date": "2021-08-31T17:13:32.453217900Z",
  "votes": null,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Hello,</p>\n<p>I tried to feed all <strong>MRI TYPES</strong> at once to PyTorch effnet01 model here:</p>\n<p><a href=\"https://www.kaggle.com/elcaiseri/pytorch-all-mri-types-once-is-what-u-need/\">NoteBook: Efficientnet3D with [ALL MRI TYPES ONCE]</a><br>\n<a href=\"https://www.kaggle.com/elcaiseri/pytorch-all-mri-types-once-is-what-u-need\" target=\"_blank\">https://www.kaggle.com/elcaiseri/pytorch-all-mri-types-once-is-what-u-need</a></p>\n<p><strong>Advantages:</strong></p>\n<ul>\n<li>Run Faster</li>\n<li>Use a larger batch size</li>\n</ul>\n<p>The inferences results were quite good, the best weight file score <strong>0.568 on Public LB.\nwhich is good for only single-weight file.</strong></p>\n<p>Take a look, I am looking for ur feedback and how we can improve this idea?</p>",
  "messages": [
    {
      "id": "1498118",
      "postDate": "08/31/2021 17:13:32",
      "content": "<p>Hello,</p>\n<p>I tried to feed all <strong>MRI TYPES</strong> at once to PyTorch effnet01 model here:</p>\n<p><a href=\"https://www.kaggle.com/elcaiseri/pytorch-all-mri-types-once-is-what-u-need/\">NoteBook: Efficientnet3D with [ALL MRI TYPES ONCE]</a><br>\n<a href=\"https://www.kaggle.com/elcaiseri/pytorch-all-mri-types-once-is-what-u-need\" target=\"_blank\">https://www.kaggle.com/elcaiseri/pytorch-all-mri-types-once-is-what-u-need</a></p>\n<p><strong>Advantages:</strong></p>\n<ul>\n<li>Run Faster</li>\n<li>Use a larger batch size</li>\n</ul>\n<p>The inferences results were quite good, the best weight file score <strong>0.568 on Public LB.\nwhich is good for only single-weight file.</strong></p>\n<p>Take a look, I am looking for ur feedback and how we can improve this idea?</p>",
      "rawMarkdown": "Hello,\n\nI tried to feed all **MRI TYPES** at once to PyTorch effnet01 model here:\n\n<a href='https://www.kaggle.com/elcaiseri/pytorch-all-mri-types-once-is-what-u-need/'>NoteBook: Efficientnet3D with [ALL MRI TYPES ONCE]</a>\nhttps://www.kaggle.com/elcaiseri/pytorch-all-mri-types-once-is-what-u-need\n\n**Advantages:**\n- Run Faster\n- Use a larger batch size\n\nThe inferences results were quite good, the best weight file score **0.568 on Public LB.\nwhich is good for only single-weight file.**\n\n\nTake a look, I am looking for ur feedback and how we can improve this idea?",
      "votes": null
    },
    {
      "id": "1500337",
      "postDate": "09/02/2021 09:45:38",
      "content": "<p>Nice &amp; compact solution! To improve, I would suggest, referring to:</p>\n<p><a href=\"https://www.kaggle.com/experienceinai/aivo-vii-nopea\" target=\"_blank\">https://www.kaggle.com/experienceinai/aivo-vii-nopea</a></p>\n<p>that:</p>\n<p>-remove the \"zero pixels\" from images by applying function like def yhden_kuvan_kasittely(hakupolku,kuvakoko)<br>\n-opitmize \"time domain place\" where to take the images by applying function like def analysoi_yhden_kansion_kuvapinon_piirteet(inputdir,nayta)</p>\n<p>Hope those would improve the results!</p>\n<p>Also, referring to the same notebook I think after you have polished \"all performance out\" from the current architecture, if you try approach named \"Level 2\" you perhaps get even better results. And, if you find applicable \"Level 3\", and get the idea out from the picture, go on! I think the time-domain may include important infomation (like you have already done by Efficientnet3D). </p>",
      "rawMarkdown": "Nice & compact solution! To improve, I would suggest, referring to:\n\nhttps://www.kaggle.com/experienceinai/aivo-vii-nopea\n\nthat:\n\n-remove the \"zero pixels\" from images by applying function like def yhden_kuvan_kasittely(hakupolku,kuvakoko)\n-opitmize \"time domain place\" where to take the images by applying function like def analysoi_yhden_kansion_kuvapinon_piirteet(inputdir,nayta)\n\nHope those would improve the results!\n\nAlso, referring to the same notebook I think after you have polished \"all performance out\" from the current architecture, if you try approach named \"Level 2\" you perhaps get even better results. And, if you find applicable \"Level 3\", and get the idea out from the picture, go on! I think the time-domain may include important infomation (like you have already done by Efficientnet3D).",
      "votes": null
    },
    {
      "id": "1501382",
      "postDate": "09/03/2021 06:46:26",
      "content": "<p>Thank u <a href=\"https://www.kaggle.com/experienceinai\" target=\"_blank\">@experienceinai</a> very much.</p>\n<p>I would try to do it and I will mention u in the next version ☺️</p>",
      "rawMarkdown": "Thank u @experienceinai very much.\n\nI would try to do it and I will mention u in the next version ☺️",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1500337,
      "author_name": "experienceinai",
      "author_url": "",
      "post_date": "09/02/2021 09:45:38",
      "content": "<p>Nice &amp; compact solution! To improve, I would suggest, referring to:</p>\n<p><a href=\"https://www.kaggle.com/experienceinai/aivo-vii-nopea\" target=\"_blank\">https://www.kaggle.com/experienceinai/aivo-vii-nopea</a></p>\n<p>that:</p>\n<p>-remove the \"zero pixels\" from images by applying function like def yhden_kuvan_kasittely(hakupolku,kuvakoko)<br>\n-opitmize \"time domain place\" where to take the images by applying function like def analysoi_yhden_kansion_kuvapinon_piirteet(inputdir,nayta)</p>\n<p>Hope those would improve the results!</p>\n<p>Also, referring to the same notebook I think after you have polished \"all performance out\" from the current architecture, if you try approach named \"Level 2\" you perhaps get even better results. And, if you find applicable \"Level 3\", and get the idea out from the picture, go on! I think the time-domain may include important infomation (like you have already done by Efficientnet3D). </p>",
      "votes": null,
      "replies": [
        {
          "id": 1501382,
          "author_name": "elcaiseri",
          "author_url": "",
          "post_date": "09/03/2021 06:46:26",
          "content": "<p>Thank u <a href=\"https://www.kaggle.com/experienceinai\" target=\"_blank\">@experienceinai</a> very much.</p>\n<p>I would try to do it and I will mention u in the next version ☺️</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1498118": "Hello,\n\nI tried to feed all **MRI TYPES** at once to PyTorch effnet01 model here:\n\n<a href='https://www.kaggle.com/elcaiseri/pytorch-all-mri-types-once-is-what-u-need/'>NoteBook: Efficientnet3D with [ALL MRI TYPES ONCE]</a>\nhttps://www.kaggle.com/elcaiseri/pytorch-all-mri-types-once-is-what-u-need\n\n**Advantages:**\n- Run Faster\n- Use a larger batch size\n\nThe inferences results were quite good, the best weight file score **0.568 on Public LB.\nwhich is good for only single-weight file.**\n\n\nTake a look, I am looking for ur feedback and how we can improve this idea?",
    "1500337": "Nice & compact solution! To improve, I would suggest, referring to:\n\nhttps://www.kaggle.com/experienceinai/aivo-vii-nopea\n\nthat:\n\n-remove the \"zero pixels\" from images by applying function like def yhden_kuvan_kasittely(hakupolku,kuvakoko)\n-opitmize \"time domain place\" where to take the images by applying function like def analysoi_yhden_kansion_kuvapinon_piirteet(inputdir,nayta)\n\nHope those would improve the results!\n\nAlso, referring to the same notebook I think after you have polished \"all performance out\" from the current architecture, if you try approach named \"Level 2\" you perhaps get even better results. And, if you find applicable \"Level 3\", and get the idea out from the picture, go on! I think the time-domain may include important infomation (like you have already done by Efficientnet3D).",
    "1501382": "Thank u @experienceinai very much.\n\nI would try to do it and I will mention u in the next version ☺️"
  },
  "source": "meta"
}