{
  "id": 137552,
  "title": "Key of the 10th solution? : Where sSE Block looks",
  "url": "/competitions/bengaliai-cv19/discussion/137552",
  "author_name": "Tawara",
  "post_date": "2020-03-21T08:34:43.917000",
  "votes": 8,
  "comment_count": 0,
  "views": 0,
  "content": "<p>I just published a new notebook, which visualizes where sSE Block looks.</p>\n\n<p>I suppose that sSE Block plays an important role as well as RandomErasing in my solution. </p>\n\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F473234%2Fe38553c42caffc6e6f3a8f79c6f36554%2F__results___40_0.png?generation=1584780074014706&amp;alt=media\" alt=\"example: consonant == 1\"></p>\n\n<p>If you want to see more, check this:  <a href=\"https://www.kaggle.com/ttahara/visualize-10th-place-model-where-sse-block-looks\">https://www.kaggle.com/ttahara/visualize-10th-place-model-where-sse-block-looks</a></p>\n\n<p>As the result, for me, difference between squeezed features of each component looks <strong>not so large</strong>. <br>\nBut this result gives me findings that <strong>sSE Block's behavior is different from simple GAP</strong>.</p>\n\n<p>This difference, may be, contributes to my simple solution getting the 10th place .\n<br>\nThanks.</p>",
  "messages": [
    {
      "id": 781404,
      "postDate": "2020-03-21T08:34:43.917Z",
      "content": "<p>I just published a new notebook, which visualizes where sSE Block looks.</p>\n\n<p>I suppose that sSE Block plays an important role as well as RandomErasing in my solution. </p>\n\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F473234%2Fe38553c42caffc6e6f3a8f79c6f36554%2F__results___40_0.png?generation=1584780074014706&amp;alt=media\" alt=\"example: consonant == 1\"></p>\n\n<p>If you want to see more, check this:  <a href=\"https://www.kaggle.com/ttahara/visualize-10th-place-model-where-sse-block-looks\">https://www.kaggle.com/ttahara/visualize-10th-place-model-where-sse-block-looks</a></p>\n\n<p>As the result, for me, difference between squeezed features of each component looks <strong>not so large</strong>. <br>\nBut this result gives me findings that <strong>sSE Block's behavior is different from simple GAP</strong>.</p>\n\n<p>This difference, may be, contributes to my simple solution getting the 10th place .\n<br>\nThanks.</p>",
      "rawMarkdown": "I just published a new notebook, which visualizes where sSE Block looks.\n\nI suppose that sSE Block plays an important role as well as RandomErasing in my solution. \n\n![example: consonant == 1](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F473234%2Fe38553c42caffc6e6f3a8f79c6f36554%2F__results___40_0.png?generation=1584780074014706&amp;alt=media)\n\n\nIf you want to see more, check this:  https://www.kaggle.com/ttahara/visualize-10th-place-model-where-sse-block-looks\n\nAs the result, for me, difference between squeezed features of each component looks **not so large**.  \nBut this result gives me findings that **sSE Block's behavior is different from simple GAP**.\n\nThis difference, may be, contributes to my simple solution getting the 10th place .\n<br>\nThanks.",
      "votes": 8
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "781404": "I just published a new notebook, which visualizes where sSE Block looks.\n\nI suppose that sSE Block plays an important role as well as RandomErasing in my solution. \n\n![example: consonant == 1](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F473234%2Fe38553c42caffc6e6f3a8f79c6f36554%2F__results___40_0.png?generation=1584780074014706&amp;alt=media)\n\n\nIf you want to see more, check this:  https://www.kaggle.com/ttahara/visualize-10th-place-model-where-sse-block-looks\n\nAs the result, for me, difference between squeezed features of each component looks **not so large**.  \nBut this result gives me findings that **sSE Block's behavior is different from simple GAP**.\n\nThis difference, may be, contributes to my simple solution getting the 10th place .\n<br>\nThanks."
  }
}