{
  "id": 604010,
  "title": "ResNet-50 Baseline for Division B — Beginner-Friendly Chest X-Ray Model",
  "url": "/competitions/grand-xray-slam-division-b/discussion/604010",
  "author_name": "Guntas Dhanjal",
  "post_date": "2025-09-05T16:34:50.878000",
  "votes": 0,
  "comment_count": 0,
  "views": 0,
  "content": "<p>Hi everyone,</p>\n<p>I’m excited to share a <strong><a href=\"https://www.kaggle.com/code/guntasdhanjal/catch-this-0-90-with-resnet-50-on-division-b\" target=\"_blank\">beginner-friendly ResNet-50 baseline</a></strong> for the <strong>Grand X-Ray Slam: Division B</strong> competition! 🚀</p>\n<h2>Overview</h2>\n<p>This notebook provides a simple educational approach to classify 14 chest conditions from X-ray images using <strong>transfer learning with ResNet-50</strong>. It’s designed to help newcomers quickly get started with model training, evaluation, and submission.</p>\n<h2>Dataset</h2>\n<ul>\n<li><code>train2.csv</code> for labels  </li>\n<li>Images in <code>train2/</code> and <code>test2/</code> directories  </li>\n</ul>\n<h2>Model &amp; Training</h2>\n<ul>\n<li>Pretrained <strong>ResNet-50</strong> as feature extractor  </li>\n<li>Custom classifier head  </li>\n<li>Trained for <strong>3 epochs</strong> to balance speed and performance  </li>\n</ul>\n<h2>Results</h2>\n<ul>\n<li>Validation AUC-ROC: <strong>~0.8969</strong>  </li>\n<li>Predictions saved in <code>submission.csv</code> for 14 chest conditions  </li>\n</ul>\n<h2>Next Steps / Suggestions</h2>\n<ul>\n<li>Experiment with <strong>data augmentation</strong> (flips, rotations)  </li>\n<li>Fine-tune more ResNet-50 layers for higher accuracy  </li>\n<li>Apply <strong>class weights</strong> to handle label imbalance  </li>\n<li>Incorporate <strong>metadata</strong> (Age, Sex, View) for improved predictions  </li>\n</ul>\n<h2>Additional Context</h2>\n<ul>\n<li>For further learning, check out <strong>Division A notebooks</strong>, which explore baseline CNNs and <strong><a href=\"https://www.kaggle.com/code/guntasdhanjal/this-efficientnetb0-will-surprise-you\" target=\"_blank\">EfficientNet with metadata</a></strong>, providing alternative approaches and comparisons.</li>\n</ul>\n<h2>Get Involved</h2>\n<ul>\n<li>Try out this notebook, experiment with different strategies, and share your results!  </li>\n<li>Happy coding and good luck on the leaderboard! 🚀</li>\n</ul>",
  "messages": [
    {
      "id": 3282097,
      "postDate": "2025-09-05T16:34:50.880Z",
      "content": "<p>Hi everyone,</p>\n<p>I’m excited to share a <strong><a href=\"https://www.kaggle.com/code/guntasdhanjal/catch-this-0-90-with-resnet-50-on-division-b\" target=\"_blank\">beginner-friendly ResNet-50 baseline</a></strong> for the <strong>Grand X-Ray Slam: Division B</strong> competition! 🚀</p>\n<h2>Overview</h2>\n<p>This notebook provides a simple educational approach to classify 14 chest conditions from X-ray images using <strong>transfer learning with ResNet-50</strong>. It’s designed to help newcomers quickly get started with model training, evaluation, and submission.</p>\n<h2>Dataset</h2>\n<ul>\n<li><code>train2.csv</code> for labels  </li>\n<li>Images in <code>train2/</code> and <code>test2/</code> directories  </li>\n</ul>\n<h2>Model &amp; Training</h2>\n<ul>\n<li>Pretrained <strong>ResNet-50</strong> as feature extractor  </li>\n<li>Custom classifier head  </li>\n<li>Trained for <strong>3 epochs</strong> to balance speed and performance  </li>\n</ul>\n<h2>Results</h2>\n<ul>\n<li>Validation AUC-ROC: <strong>~0.8969</strong>  </li>\n<li>Predictions saved in <code>submission.csv</code> for 14 chest conditions  </li>\n</ul>\n<h2>Next Steps / Suggestions</h2>\n<ul>\n<li>Experiment with <strong>data augmentation</strong> (flips, rotations)  </li>\n<li>Fine-tune more ResNet-50 layers for higher accuracy  </li>\n<li>Apply <strong>class weights</strong> to handle label imbalance  </li>\n<li>Incorporate <strong>metadata</strong> (Age, Sex, View) for improved predictions  </li>\n</ul>\n<h2>Additional Context</h2>\n<ul>\n<li>For further learning, check out <strong>Division A notebooks</strong>, which explore baseline CNNs and <strong><a href=\"https://www.kaggle.com/code/guntasdhanjal/this-efficientnetb0-will-surprise-you\" target=\"_blank\">EfficientNet with metadata</a></strong>, providing alternative approaches and comparisons.</li>\n</ul>\n<h2>Get Involved</h2>\n<ul>\n<li>Try out this notebook, experiment with different strategies, and share your results!  </li>\n<li>Happy coding and good luck on the leaderboard! 🚀</li>\n</ul>",
      "rawMarkdown": "Hi everyone,\n\nI’m excited to share a **[beginner-friendly ResNet-50 baseline](https://www.kaggle.com/code/guntasdhanjal/catch-this-0-90-with-resnet-50-on-division-b)** for the **Grand X-Ray Slam: Division B** competition! 🚀\n\n## Overview\nThis notebook provides a simple educational approach to classify 14 chest conditions from X-ray images using **transfer learning with ResNet-50**. It’s designed to help newcomers quickly get started with model training, evaluation, and submission.\n\n## Dataset\n- `train2.csv` for labels  \n- Images in `train2/` and `test2/` directories  \n\n## Model & Training\n- Pretrained **ResNet-50** as feature extractor  \n- Custom classifier head  \n- Trained for **3 epochs** to balance speed and performance  \n\n## Results\n- Validation AUC-ROC: **~0.8969**  \n- Predictions saved in `submission.csv` for 14 chest conditions  \n\n## Next Steps / Suggestions\n- Experiment with **data augmentation** (flips, rotations)  \n- Fine-tune more ResNet-50 layers for higher accuracy  \n- Apply **class weights** to handle label imbalance  \n- Incorporate **metadata** (Age, Sex, View) for improved predictions  \n\n## Additional Context\n- For further learning, check out **Division A notebooks**, which explore baseline CNNs and **[EfficientNet with metadata](https://www.kaggle.com/code/guntasdhanjal/this-efficientnetb0-will-surprise-you)**, providing alternative approaches and comparisons.\n\n## Get Involved\n- Try out this notebook, experiment with different strategies, and share your results!  \n- Happy coding and good luck on the leaderboard! 🚀\n"
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "3282097": "Hi everyone,\n\nI’m excited to share a **[beginner-friendly ResNet-50 baseline](https://www.kaggle.com/code/guntasdhanjal/catch-this-0-90-with-resnet-50-on-division-b)** for the **Grand X-Ray Slam: Division B** competition! 🚀\n\n## Overview\nThis notebook provides a simple educational approach to classify 14 chest conditions from X-ray images using **transfer learning with ResNet-50**. It’s designed to help newcomers quickly get started with model training, evaluation, and submission.\n\n## Dataset\n- `train2.csv` for labels  \n- Images in `train2/` and `test2/` directories  \n\n## Model & Training\n- Pretrained **ResNet-50** as feature extractor  \n- Custom classifier head  \n- Trained for **3 epochs** to balance speed and performance  \n\n## Results\n- Validation AUC-ROC: **~0.8969**  \n- Predictions saved in `submission.csv` for 14 chest conditions  \n\n## Next Steps / Suggestions\n- Experiment with **data augmentation** (flips, rotations)  \n- Fine-tune more ResNet-50 layers for higher accuracy  \n- Apply **class weights** to handle label imbalance  \n- Incorporate **metadata** (Age, Sex, View) for improved predictions  \n\n## Additional Context\n- For further learning, check out **Division A notebooks**, which explore baseline CNNs and **[EfficientNet with metadata](https://www.kaggle.com/code/guntasdhanjal/this-efficientnetb0-will-surprise-you)**, providing alternative approaches and comparisons.\n\n## Get Involved\n- Try out this notebook, experiment with different strategies, and share your results!  \n- Happy coding and good luck on the leaderboard! 🚀\n"
  }
}