{
  "id": 104933,
  "title": "Segmentation Crop 데이터셋을 공유합니다.",
  "url": "/competitions/2019-3rd-ml-month-with-kakr/discussion/104933",
  "author_name": "",
  "post_date": "2019-08-20T05:09:52.679769700Z",
  "votes": 3,
  "comment_count": 2,
  "views": 0,
  "content": "<p>대회가 종료되서 생성했던 전처리 데이터셋을 지울까 생각하다가, 누군가 공부삼아 써보고 싶으시거나 연구하고 싶으신 분이 있을지 몰라서 공유합니다.</p>\n\n<p>코드는 아래 커널에 바탕색만 칠한 정도라서 참조하시면 될것 같습니다.\n<a href=\"https://www.kaggle.com/cruiserx/3rd-ml-month-car-image-segmentation-crop\">https://www.kaggle.com/cruiserx/3rd-ml-month-car-image-segmentation-crop</a></p>\n\n<p>모델 튜닝하기 전의 결과라 다시 돌려보면 다를지 모르겠습니다만,\n실험기록을 찾아보니 아래와 같습니다. 그나마 회색으로 채운게 높았습니다.\n더 정밀하게 자른게 왜 더 점수가 낮은지는 아직도 의문입니다..</p>\n\n<ul>\n<li>태명님set   : Acc 0.8862</li>\n<li>3번 회색set : Acc 0.8546</li>\n<li>2번 흰색set : Acc 0.8475</li>\n<li>1번 검정색set:Acc 0.8438</li>\n<li>4번 랜덤색set:Acc 0.8418</li>\n<li><p>5번 겹치기set:Acc 0.8322</p>\n\n<ol><li><p>Segmentation Crop + 검정배경\n<a href=\"https://www.kaggle.com/cruiserx/3rdrcnnsegcrop\">https://www.kaggle.com/cruiserx/3rdrcnnsegcrop</a>\n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2340279%2F982d9aab75b2634880e91c7a9b38ff56%2Fblack_test_00002.jpg?generation=1566277631684120&amp;alt=media\" alt=\"\"></p></li>\n<li><p>Segmentation Crop + 흰색배경\n<a href=\"https://www.kaggle.com/cruiserx/3rdrcnnsegcropwhite\">https://www.kaggle.com/cruiserx/3rdrcnnsegcropwhite</a>\n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2340279%2F11f6c88854b298f2b4a88ec842a84ffa%2Fwhite_test_00002.jpg?generation=1566277659550932&amp;alt=media\" alt=\"\"></p></li>\n<li><p>Segmentation Crop + 회색배경\n<a href=\"https://www.kaggle.com/cruiserx/3rdrcnnsegcropgray\">https://www.kaggle.com/cruiserx/3rdrcnnsegcropgray</a>\n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2340279%2Ff74835d0b382b1c008c33f68614f0c3e%2Fgray_test_00002.jpg?generation=1566277685310107&amp;alt=media\" alt=\"\"></p></li>\n<li><p>Segmentation Crop + 랜덤배경\n<a href=\"https://www.kaggle.com/cruiserx/3rdrcnnsegcroprand\">https://www.kaggle.com/cruiserx/3rdrcnnsegcroprand</a>\n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2340279%2F688949ad59431e40ff27c97fe2e4e602%2Frandom_test_00002.jpg?generation=1566277707325977&amp;alt=media\" alt=\"\"></p></li>\n<li><p>Segmentation Crop + 겹치기\n<a href=\"https://www.kaggle.com/cruiserx/3rdrcnnsegcroppatch\">https://www.kaggle.com/cruiserx/3rdrcnnsegcroppatch</a>\n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2340279%2F08a468e3e97277404c3c552c3aec405a%2Foverlap_test_00002.jpg?generation=1566277731091006&amp;alt=media\" alt=\"\"></p></li></ol></li>\n</ul>",
  "messages": [
    {
      "id": "603282",
      "postDate": "08/20/2019 05:09:52",
      "content": "<p>대회가 종료되서 생성했던 전처리 데이터셋을 지울까 생각하다가, 누군가 공부삼아 써보고 싶으시거나 연구하고 싶으신 분이 있을지 몰라서 공유합니다.</p>\n\n<p>코드는 아래 커널에 바탕색만 칠한 정도라서 참조하시면 될것 같습니다.\n<a href=\"https://www.kaggle.com/cruiserx/3rd-ml-month-car-image-segmentation-crop\">https://www.kaggle.com/cruiserx/3rd-ml-month-car-image-segmentation-crop</a></p>\n\n<p>모델 튜닝하기 전의 결과라 다시 돌려보면 다를지 모르겠습니다만,\n실험기록을 찾아보니 아래와 같습니다. 그나마 회색으로 채운게 높았습니다.\n더 정밀하게 자른게 왜 더 점수가 낮은지는 아직도 의문입니다..</p>\n\n<ul>\n<li>태명님set   : Acc 0.8862</li>\n<li>3번 회색set : Acc 0.8546</li>\n<li>2번 흰색set : Acc 0.8475</li>\n<li>1번 검정색set:Acc 0.8438</li>\n<li>4번 랜덤색set:Acc 0.8418</li>\n<li><p>5번 겹치기set:Acc 0.8322</p>\n\n<ol><li><p>Segmentation Crop + 검정배경\n<a href=\"https://www.kaggle.com/cruiserx/3rdrcnnsegcrop\">https://www.kaggle.com/cruiserx/3rdrcnnsegcrop</a>\n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2340279%2F982d9aab75b2634880e91c7a9b38ff56%2Fblack_test_00002.jpg?generation=1566277631684120&amp;alt=media\" alt=\"\"></p></li>\n<li><p>Segmentation Crop + 흰색배경\n<a href=\"https://www.kaggle.com/cruiserx/3rdrcnnsegcropwhite\">https://www.kaggle.com/cruiserx/3rdrcnnsegcropwhite</a>\n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2340279%2F11f6c88854b298f2b4a88ec842a84ffa%2Fwhite_test_00002.jpg?generation=1566277659550932&amp;alt=media\" alt=\"\"></p></li>\n<li><p>Segmentation Crop + 회색배경\n<a href=\"https://www.kaggle.com/cruiserx/3rdrcnnsegcropgray\">https://www.kaggle.com/cruiserx/3rdrcnnsegcropgray</a>\n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2340279%2Ff74835d0b382b1c008c33f68614f0c3e%2Fgray_test_00002.jpg?generation=1566277685310107&amp;alt=media\" alt=\"\"></p></li>\n<li><p>Segmentation Crop + 랜덤배경\n<a href=\"https://www.kaggle.com/cruiserx/3rdrcnnsegcroprand\">https://www.kaggle.com/cruiserx/3rdrcnnsegcroprand</a>\n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2340279%2F688949ad59431e40ff27c97fe2e4e602%2Frandom_test_00002.jpg?generation=1566277707325977&amp;alt=media\" alt=\"\"></p></li>\n<li><p>Segmentation Crop + 겹치기\n<a href=\"https://www.kaggle.com/cruiserx/3rdrcnnsegcroppatch\">https://www.kaggle.com/cruiserx/3rdrcnnsegcroppatch</a>\n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2340279%2F08a468e3e97277404c3c552c3aec405a%2Foverlap_test_00002.jpg?generation=1566277731091006&amp;alt=media\" alt=\"\"></p></li></ol></li>\n</ul>",
      "rawMarkdown": "대회가 종료되서 생성했던 전처리 데이터셋을 지울까 생각하다가, 누군가 공부삼아 써보고 싶으시거나 연구하고 싶으신 분이 있을지 몰라서 공유합니다.\n\n코드는 아래 커널에 바탕색만 칠한 정도라서 참조하시면 될것 같습니다.\nhttps://www.kaggle.com/cruiserx/3rd-ml-month-car-image-segmentation-crop\n\n모델 튜닝하기 전의 결과라 다시 돌려보면 다를지 모르겠습니다만,\n실험기록을 찾아보니 아래와 같습니다. 그나마 회색으로 채운게 높았습니다.\n더 정밀하게 자른게 왜 더 점수가 낮은지는 아직도 의문입니다..\n\n - 태명님set   : Acc 0.8862\n - 3번 회색set : Acc 0.8546\n - 2번 흰색set : Acc 0.8475\n - 1번 검정색set:Acc 0.8438\n - 4번 랜덤색set:Acc 0.8418\n - 5번 겹치기set:Acc 0.8322\n\n\n1. Segmentation Crop + 검정배경\nhttps://www.kaggle.com/cruiserx/3rdrcnnsegcrop\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2340279%2F982d9aab75b2634880e91c7a9b38ff56%2Fblack_test_00002.jpg?generation=1566277631684120&amp;alt=media)\n\n2. Segmentation Crop + 흰색배경\nhttps://www.kaggle.com/cruiserx/3rdrcnnsegcropwhite\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2340279%2F11f6c88854b298f2b4a88ec842a84ffa%2Fwhite_test_00002.jpg?generation=1566277659550932&amp;alt=media)\n\n3. Segmentation Crop + 회색배경\nhttps://www.kaggle.com/cruiserx/3rdrcnnsegcropgray\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2340279%2Ff74835d0b382b1c008c33f68614f0c3e%2Fgray_test_00002.jpg?generation=1566277685310107&amp;alt=media)\n\n\n4. Segmentation Crop + 랜덤배경\nhttps://www.kaggle.com/cruiserx/3rdrcnnsegcroprand\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2340279%2F688949ad59431e40ff27c97fe2e4e602%2Frandom_test_00002.jpg?generation=1566277707325977&amp;alt=media)\n\n\n5. Segmentation Crop + 겹치기\nhttps://www.kaggle.com/cruiserx/3rdrcnnsegcroppatch\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2340279%2F08a468e3e97277404c3c552c3aec405a%2Foverlap_test_00002.jpg?generation=1566277731091006&amp;alt=media)",
      "votes": null
    },
    {
      "id": "603650",
      "postDate": "08/20/2019 14:20:49",
      "content": "<p>혹시 흰 차나 검정 차가 많아서 그런건 아닐까요?</p>",
      "rawMarkdown": "혹시 흰 차나 검정 차가 많아서 그런건 아닐까요?",
      "votes": null
    },
    {
      "id": "605122",
      "postDate": "08/22/2019 04:49:21",
      "content": "<p>그럴수도 있겠네요. 전체 사진들 색상을 확인해보진 않았으니.. 시간날때 연구좀 해봐야겠습니다 ^^</p>",
      "rawMarkdown": "그럴수도 있겠네요. 전체 사진들 색상을 확인해보진 않았으니.. 시간날때 연구좀 해봐야겠습니다 ^^",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 603650,
      "author_name": "seriousran",
      "author_url": "",
      "post_date": "08/20/2019 14:20:49",
      "content": "<p>혹시 흰 차나 검정 차가 많아서 그런건 아닐까요?</p>",
      "votes": null,
      "replies": [
        {
          "id": 605122,
          "author_name": "cruiserx",
          "author_url": "",
          "post_date": "08/22/2019 04:49:21",
          "content": "<p>그럴수도 있겠네요. 전체 사진들 색상을 확인해보진 않았으니.. 시간날때 연구좀 해봐야겠습니다 ^^</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "603282": "대회가 종료되서 생성했던 전처리 데이터셋을 지울까 생각하다가, 누군가 공부삼아 써보고 싶으시거나 연구하고 싶으신 분이 있을지 몰라서 공유합니다.\n\n코드는 아래 커널에 바탕색만 칠한 정도라서 참조하시면 될것 같습니다.\nhttps://www.kaggle.com/cruiserx/3rd-ml-month-car-image-segmentation-crop\n\n모델 튜닝하기 전의 결과라 다시 돌려보면 다를지 모르겠습니다만,\n실험기록을 찾아보니 아래와 같습니다. 그나마 회색으로 채운게 높았습니다.\n더 정밀하게 자른게 왜 더 점수가 낮은지는 아직도 의문입니다..\n\n - 태명님set   : Acc 0.8862\n - 3번 회색set : Acc 0.8546\n - 2번 흰색set : Acc 0.8475\n - 1번 검정색set:Acc 0.8438\n - 4번 랜덤색set:Acc 0.8418\n - 5번 겹치기set:Acc 0.8322\n\n\n1. Segmentation Crop + 검정배경\nhttps://www.kaggle.com/cruiserx/3rdrcnnsegcrop\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2340279%2F982d9aab75b2634880e91c7a9b38ff56%2Fblack_test_00002.jpg?generation=1566277631684120&amp;alt=media)\n\n2. Segmentation Crop + 흰색배경\nhttps://www.kaggle.com/cruiserx/3rdrcnnsegcropwhite\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2340279%2F11f6c88854b298f2b4a88ec842a84ffa%2Fwhite_test_00002.jpg?generation=1566277659550932&amp;alt=media)\n\n3. Segmentation Crop + 회색배경\nhttps://www.kaggle.com/cruiserx/3rdrcnnsegcropgray\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2340279%2Ff74835d0b382b1c008c33f68614f0c3e%2Fgray_test_00002.jpg?generation=1566277685310107&amp;alt=media)\n\n\n4. Segmentation Crop + 랜덤배경\nhttps://www.kaggle.com/cruiserx/3rdrcnnsegcroprand\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2340279%2F688949ad59431e40ff27c97fe2e4e602%2Frandom_test_00002.jpg?generation=1566277707325977&amp;alt=media)\n\n\n5. Segmentation Crop + 겹치기\nhttps://www.kaggle.com/cruiserx/3rdrcnnsegcroppatch\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2340279%2F08a468e3e97277404c3c552c3aec405a%2Foverlap_test_00002.jpg?generation=1566277731091006&amp;alt=media)",
    "603650": "혹시 흰 차나 검정 차가 많아서 그런건 아닐까요?",
    "605122": "그럴수도 있겠네요. 전체 사진들 색상을 확인해보진 않았으니.. 시간날때 연구좀 해봐야겠습니다 ^^"
  },
  "source": "meta"
}