{
  "id": 245216,
  "title": "Best Single Model",
  "url": "/competitions/siim-covid19-detection/discussion/245216",
  "author_name": "",
  "post_date": "2021-06-10T07:19:32.088866600Z",
  "votes": 2,
  "comment_count": 2,
  "views": 0,
  "content": "<p>I've been exploring this competition for a while. <br>\nIs there anyone who achieved good results with Image level data or by using both Image + Study level data?</p>\n<p>For me, My best CV right now is resnet18d on Study level data.</p>",
  "messages": [
    {
      "id": "1343383",
      "postDate": "06/10/2021 07:19:32",
      "content": "<p>I've been exploring this competition for a while. <br>\nIs there anyone who achieved good results with Image level data or by using both Image + Study level data?</p>\n<p>For me, My best CV right now is resnet18d on Study level data.</p>",
      "rawMarkdown": "I've been exploring this competition for a while. \nIs there anyone who achieved good results with Image level data or by using both Image + Study level data?\n\nFor me, My best CV right now is resnet18d on Study level data.",
      "votes": null
    },
    {
      "id": "1343642",
      "postDate": "06/10/2021 10:46:48",
      "content": "<p>On my stratified 4 fold local data, image-level has huge train/test inconsistency.  Validation loss is like 5x of training loss. I started to investigate my data pipeline now, but it's hard to tell whether my processed GT is correct or not without domain knowledge.</p>",
      "rawMarkdown": "On my stratified 4 fold local data, image-level has huge train/test inconsistency.  Validation loss is like 5x of training loss. I started to investigate my data pipeline now, but it's hard to tell whether my processed GT is correct or not without domain knowledge.",
      "votes": null
    },
    {
      "id": "1343735",
      "postDate": "06/10/2021 11:58:01",
      "content": "<p>EfficientnetB7 is best CV at this moment. As there are issues in image level data submission I am not sure about that part. </p>",
      "rawMarkdown": "EfficientnetB7 is best CV at this moment. As there are issues in image level data submission I am not sure about that part.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1343642,
      "author_name": "artnotintelligence",
      "author_url": "",
      "post_date": "06/10/2021 10:46:48",
      "content": "<p>On my stratified 4 fold local data, image-level has huge train/test inconsistency.  Validation loss is like 5x of training loss. I started to investigate my data pipeline now, but it's hard to tell whether my processed GT is correct or not without domain knowledge.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1343735,
      "author_name": "eakdag",
      "author_url": "",
      "post_date": "06/10/2021 11:58:01",
      "content": "<p>EfficientnetB7 is best CV at this moment. As there are issues in image level data submission I am not sure about that part. </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1343383": "I've been exploring this competition for a while. \nIs there anyone who achieved good results with Image level data or by using both Image + Study level data?\n\nFor me, My best CV right now is resnet18d on Study level data.",
    "1343642": "On my stratified 4 fold local data, image-level has huge train/test inconsistency.  Validation loss is like 5x of training loss. I started to investigate my data pipeline now, but it's hard to tell whether my processed GT is correct or not without domain knowledge.",
    "1343735": "EfficientnetB7 is best CV at this moment. As there are issues in image level data submission I am not sure about that part."
  },
  "source": "meta"
}