{
  "id": 378490,
  "title": "Only one class of breast density is predicted!",
  "url": "/competitions/rsna-breast-cancer-detection/discussion/378490",
  "author_name": "Erdi Kılıç",
  "post_date": "2023-01-16T00:10:55.056000",
  "votes": 2,
  "comment_count": 2,
  "views": 0,
  "content": "<p>This will be a little out of context, but I am just wondering why I'm getting this kind of prediction for breast density via;</p>\n<p><code>model = timm.create_model(\"convnext_tiny\", num_classes=4, pretrained=True, global_pool=\"max\", drop_rate=0.10)</code></p>\n<p>The model predicts only one class. <br>\nI am facing this first time and soo confused about that. What can be wrong here? It predicts better when I use other architectures like efficientnet.</p>\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F5727397%2Fbf18745bb692a1cb52f3b596765c8438%2F__results___30_0.png?generation=1673826670391036&amp;alt=media\" alt=\"\"></p>",
  "messages": [
    {
      "id": 2101475,
      "postDate": "2023-01-16T00:10:55.057Z",
      "content": "<p>This will be a little out of context, but I am just wondering why I'm getting this kind of prediction for breast density via;</p>\n<p><code>model = timm.create_model(\"convnext_tiny\", num_classes=4, pretrained=True, global_pool=\"max\", drop_rate=0.10)</code></p>\n<p>The model predicts only one class. <br>\nI am facing this first time and soo confused about that. What can be wrong here? It predicts better when I use other architectures like efficientnet.</p>\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F5727397%2Fbf18745bb692a1cb52f3b596765c8438%2F__results___30_0.png?generation=1673826670391036&amp;alt=media\" alt=\"\"></p>",
      "rawMarkdown": "This will be a little out of context, but I am just wondering why I'm getting this kind of prediction for breast density via;\n\n`model = timm.create_model(\"convnext_tiny\", num_classes=4, pretrained=True, global_pool=\"max\", drop_rate=0.10) `\n\n\nThe model predicts only one class. \nI am facing this first time and soo confused about that. What can be wrong here? It predicts better when I use other architectures like efficientnet.\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F5727397%2Fbf18745bb692a1cb52f3b596765c8438%2F__results___30_0.png?generation=1673826670391036&alt=media)",
      "votes": 2
    },
    {
      "id": 2102278,
      "postDate": "2023-01-16T13:56:43.187Z",
      "content": "<p>have you fine-tuned it on the your dataset or are you just using the pretrained weights?</p>",
      "rawMarkdown": "have you fine-tuned it on the your dataset or are you just using the pretrained weights?",
      "replies": [
        {
          "id": 2102297,
          "postDate": "2023-01-16T14:13:50.383Z",
          "content": "<p>I loaded the weights and trained for a few epochs.</p>",
          "rawMarkdown": "I loaded the weights and trained for a few epochs."
        }
      ]
    }
  ],
  "comments": [
    {
      "id": 2102278,
      "author_name": "taghados",
      "author_url": "",
      "post_date": "2023-01-16T13:56:43.187000",
      "content": "<p>have you fine-tuned it on the your dataset or are you just using the pretrained weights?</p>",
      "votes": 0,
      "replies": [
        {
          "id": 2102297,
          "author_name": "Erdi Kılıç",
          "author_url": "",
          "post_date": "2023-01-16T14:13:50.383000",
          "content": "<p>I loaded the weights and trained for a few epochs.</p>",
          "votes": 0,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "2101475": "This will be a little out of context, but I am just wondering why I'm getting this kind of prediction for breast density via;\n\n`model = timm.create_model(\"convnext_tiny\", num_classes=4, pretrained=True, global_pool=\"max\", drop_rate=0.10) `\n\n\nThe model predicts only one class. \nI am facing this first time and soo confused about that. What can be wrong here? It predicts better when I use other architectures like efficientnet.\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F5727397%2Fbf18745bb692a1cb52f3b596765c8438%2F__results___30_0.png?generation=1673826670391036&alt=media)",
    "2102278": "have you fine-tuned it on the your dataset or are you just using the pretrained weights?"
  }
}