{
  "id": 156890,
  "title": "Melanoma Classification with Attention. ",
  "url": "/competitions/siim-isic-melanoma-classification/discussion/156890",
  "author_name": "",
  "post_date": "2020-06-08T11:23:37.341053500Z",
  "votes": 8,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Attention mechanism is very popular with NLP models however the same can't be seen when it comes to computer vision tasks. Therefore, I wrote a kernel to see how it does in this competition. \n<a href=\"https://www.kaggle.com/ibtesama/melanoma-classification-with-attention?scriptVersionId=35705148\">https://www.kaggle.com/ibtesama/melanoma-classification-with-attention?scriptVersionId=35705148</a></p>\n\n<p>I have only trained on a sample of data(2500 images) for 2 epochs without much tuning or k-fold or tabular data with VGG16 as the backbone. It gave an LB score of <code>0.886</code>.</p>\n\n<p>I'm quite sure when trained on more images or tabular data or a popular network for these competitions like <code>Resnet</code> or <code>EfficientNet</code> or even with some good hyperparameter tuning, attention mechanism will yield really good results.</p>\n\n<p>So, try it out guys and share your findings in the comments.</p>",
  "messages": [
    {
      "id": "878226",
      "postDate": "06/08/2020 11:23:37",
      "content": "<p>Attention mechanism is very popular with NLP models however the same can't be seen when it comes to computer vision tasks. Therefore, I wrote a kernel to see how it does in this competition. \n<a href=\"https://www.kaggle.com/ibtesama/melanoma-classification-with-attention?scriptVersionId=35705148\">https://www.kaggle.com/ibtesama/melanoma-classification-with-attention?scriptVersionId=35705148</a></p>\n\n<p>I have only trained on a sample of data(2500 images) for 2 epochs without much tuning or k-fold or tabular data with VGG16 as the backbone. It gave an LB score of <code>0.886</code>.</p>\n\n<p>I'm quite sure when trained on more images or tabular data or a popular network for these competitions like <code>Resnet</code> or <code>EfficientNet</code> or even with some good hyperparameter tuning, attention mechanism will yield really good results.</p>\n\n<p>So, try it out guys and share your findings in the comments.</p>",
      "rawMarkdown": "Attention mechanism is very popular with NLP models however the same can't be seen when it comes to computer vision tasks. Therefore, I wrote a kernel to see how it does in this competition. \n[https://www.kaggle.com/ibtesama/melanoma-classification-with-attention?scriptVersionId=35705148](https://www.kaggle.com/ibtesama/melanoma-classification-with-attention?scriptVersionId=35705148)\n \nI have only trained on a sample of data(2500 images) for 2 epochs without much tuning or k-fold or tabular data with VGG16 as the backbone. It gave an LB score of `0.886`.\n\nI'm quite sure when trained on more images or tabular data or a popular network for these competitions like `Resnet` or `EfficientNet` or even with some good hyperparameter tuning, attention mechanism will yield really good results.\n\nSo, try it out guys and share your findings in the comments.",
      "votes": null
    },
    {
      "id": "878538",
      "postDate": "06/08/2020 16:12:10",
      "content": "<p>Valuable paper and tip about attention mechanism.</p>",
      "rawMarkdown": "Valuable paper and tip about attention mechanism.",
      "votes": null
    },
    {
      "id": "879781",
      "postDate": "06/09/2020 18:02:06",
      "content": "<p>Effectively applied Attention mechanism in Computer Vision task. Very well articulated. </p>",
      "rawMarkdown": "Effectively applied Attention mechanism in Computer Vision task. Very well articulated.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 878538,
      "author_name": "mpwolke",
      "author_url": "",
      "post_date": "06/08/2020 16:12:10",
      "content": "<p>Valuable paper and tip about attention mechanism.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 879781,
      "author_name": "suvralipi",
      "author_url": "",
      "post_date": "06/09/2020 18:02:06",
      "content": "<p>Effectively applied Attention mechanism in Computer Vision task. Very well articulated. </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "878226": "Attention mechanism is very popular with NLP models however the same can't be seen when it comes to computer vision tasks. Therefore, I wrote a kernel to see how it does in this competition. \n[https://www.kaggle.com/ibtesama/melanoma-classification-with-attention?scriptVersionId=35705148](https://www.kaggle.com/ibtesama/melanoma-classification-with-attention?scriptVersionId=35705148)\n \nI have only trained on a sample of data(2500 images) for 2 epochs without much tuning or k-fold or tabular data with VGG16 as the backbone. It gave an LB score of `0.886`.\n\nI'm quite sure when trained on more images or tabular data or a popular network for these competitions like `Resnet` or `EfficientNet` or even with some good hyperparameter tuning, attention mechanism will yield really good results.\n\nSo, try it out guys and share your findings in the comments.",
    "878538": "Valuable paper and tip about attention mechanism.",
    "879781": "Effectively applied Attention mechanism in Computer Vision task. Very well articulated."
  },
  "source": "meta"
}