{
  "id": 81885,
  "title": "List of duplicate whale IDs",
  "url": "/competitions/humpback-whale-identification/discussion/81885",
  "author_name": "Oleksandr Mokin",
  "post_date": "2019-02-25T20:25:08.273000",
  "votes": 26,
  "comment_count": 22,
  "views": 0,
  "content": "<p>Here are the ones that I managed to find. Yes, quite a lot. And unfortunately, they are not recognized by the assessment system as the same whale. So be careful.</p>\n\n<pre><code>w_84b1b28,w_163b2b8\nw_297e53a,w_9f2cfff\nw_b43d0fa,w_19f507e\nw_e4a9205,w_5c4c166\nw_5fcf350,w_e21b25f\nw_bd18f3c,w_185044d\nw_3d7bc85,w_cb0e8c2\nw_fae4be5,w_f84e6e1,w_1d35a02\nw_2ac2bac,w_3bf887a\nw_2f8de4f,w_f7234f1\nw_7efe670,w_597758e\nw_28218e1,w_62bb54b\nw_c1b90d1,w_950c471\nw_94e8c02,w_169253a\nw_eb31833,w_662c89b\nw_a44a9b2,w_6fe3566\nw_5ce4848,w_81ed5f8\nw_b2ef717,w_07d4b0a\nw_808d40b,w_b035775\nw_c97ad94,w_a3181a0\nw_b82d0eb,w_aa0dacc\nw_298f605,w_43614bc\nw_bbf0c38,w_32e705d\nw_7d46ac3,w_3729146\nw_25d662d,w_d2facb4\nw_5010896,w_a1a0090\nw_8431ae8,w_115e880\nw_f5d4627,w_2623921\nw_b6e4761,w_bfecb74\nw_52162eb,w_a112639\nw_60b027a,w_c87817e\nw_022b708,w_d215a68\nw_da651f6,w_d92b4c4\nw_5072c08,w_8694cb5\nw_f94fc92,w_228c7ee\nw_0532483,w_444b771\nw_0cc0430,w_97074d8\nw_d7aad03,w_e56d7b2\nw_6525e6b,w_f1b016e\nw_ddda4a4,w_9353e7c\nw_ea2d2f9,w_3789389\nw_26ba5fd,w_8daccba\nw_792ffd7,w_a39b2e2\nw_e12b78d,w_9ffc9aa\nw_353d249,w_af1d57b\nw_4130233,w_cc93694\nw_c1715f5,w_706e4fb\nw_ac6aee2,w_a8276b4\n</code></pre>",
  "messages": [
    {
      "id": 478168,
      "postDate": "2019-02-25T20:25:08.273Z",
      "content": "<p>Here are the ones that I managed to find. Yes, quite a lot. And unfortunately, they are not recognized by the assessment system as the same whale. So be careful.</p>\n\n<pre><code>w_84b1b28,w_163b2b8\nw_297e53a,w_9f2cfff\nw_b43d0fa,w_19f507e\nw_e4a9205,w_5c4c166\nw_5fcf350,w_e21b25f\nw_bd18f3c,w_185044d\nw_3d7bc85,w_cb0e8c2\nw_fae4be5,w_f84e6e1,w_1d35a02\nw_2ac2bac,w_3bf887a\nw_2f8de4f,w_f7234f1\nw_7efe670,w_597758e\nw_28218e1,w_62bb54b\nw_c1b90d1,w_950c471\nw_94e8c02,w_169253a\nw_eb31833,w_662c89b\nw_a44a9b2,w_6fe3566\nw_5ce4848,w_81ed5f8\nw_b2ef717,w_07d4b0a\nw_808d40b,w_b035775\nw_c97ad94,w_a3181a0\nw_b82d0eb,w_aa0dacc\nw_298f605,w_43614bc\nw_bbf0c38,w_32e705d\nw_7d46ac3,w_3729146\nw_25d662d,w_d2facb4\nw_5010896,w_a1a0090\nw_8431ae8,w_115e880\nw_f5d4627,w_2623921\nw_b6e4761,w_bfecb74\nw_52162eb,w_a112639\nw_60b027a,w_c87817e\nw_022b708,w_d215a68\nw_da651f6,w_d92b4c4\nw_5072c08,w_8694cb5\nw_f94fc92,w_228c7ee\nw_0532483,w_444b771\nw_0cc0430,w_97074d8\nw_d7aad03,w_e56d7b2\nw_6525e6b,w_f1b016e\nw_ddda4a4,w_9353e7c\nw_ea2d2f9,w_3789389\nw_26ba5fd,w_8daccba\nw_792ffd7,w_a39b2e2\nw_e12b78d,w_9ffc9aa\nw_353d249,w_af1d57b\nw_4130233,w_cc93694\nw_c1715f5,w_706e4fb\nw_ac6aee2,w_a8276b4\n</code></pre>",
      "rawMarkdown": "Here are the ones that I managed to find. Yes, quite a lot. And unfortunately, they are not recognized by the assessment system as the same whale. So be careful.\n\n    w_84b1b28,w_163b2b8\n    w_297e53a,w_9f2cfff\n    w_b43d0fa,w_19f507e\n    w_e4a9205,w_5c4c166\n    w_5fcf350,w_e21b25f\n    w_bd18f3c,w_185044d\n    w_3d7bc85,w_cb0e8c2\n    w_fae4be5,w_f84e6e1,w_1d35a02\n    w_2ac2bac,w_3bf887a\n    w_2f8de4f,w_f7234f1\n    w_7efe670,w_597758e\n    w_28218e1,w_62bb54b\n    w_c1b90d1,w_950c471\n    w_94e8c02,w_169253a\n    w_eb31833,w_662c89b\n    w_a44a9b2,w_6fe3566\n    w_5ce4848,w_81ed5f8\n    w_b2ef717,w_07d4b0a\n    w_808d40b,w_b035775\n    w_c97ad94,w_a3181a0\n    w_b82d0eb,w_aa0dacc\n    w_298f605,w_43614bc\n    w_bbf0c38,w_32e705d\n    w_7d46ac3,w_3729146\n    w_25d662d,w_d2facb4\n    w_5010896,w_a1a0090\n    w_8431ae8,w_115e880\n    w_f5d4627,w_2623921\n    w_b6e4761,w_bfecb74\n    w_52162eb,w_a112639\n    w_60b027a,w_c87817e\n    w_022b708,w_d215a68\n    w_da651f6,w_d92b4c4\n    w_5072c08,w_8694cb5\n    w_f94fc92,w_228c7ee\n    w_0532483,w_444b771\n    w_0cc0430,w_97074d8\n    w_d7aad03,w_e56d7b2\n    w_6525e6b,w_f1b016e\n    w_ddda4a4,w_9353e7c\n    w_ea2d2f9,w_3789389\n    w_26ba5fd,w_8daccba\n    w_792ffd7,w_a39b2e2\n    w_e12b78d,w_9ffc9aa\n    w_353d249,w_af1d57b\n    w_4130233,w_cc93694\n    w_c1715f5,w_706e4fb\n    w_ac6aee2,w_a8276b4\n\n",
      "votes": 26
    },
    {
      "id": 478231,
      "postDate": "2019-02-25T22:56:59.747Z",
      "content": "<p>thanks. this will affect the LB score so there is an element of luck here.</p>",
      "rawMarkdown": "thanks. this will affect the LB score so there is an element of luck here.",
      "votes": 3,
      "replies": [
        {
          "id": 478315,
          "postDate": "2019-02-26T02:25:51.033Z",
          "content": "<p>I bet all samples with ambiguous labels are removed from private test set, thus don't worry : )</p>",
          "rawMarkdown": "I bet all samples with ambiguous labels are removed from private test set, thus don't worry : )",
          "votes": -1
        },
        {
          "id": 478334,
          "postDate": "2019-02-26T03:08:29.743Z",
          "content": "<p>i don't think so. they can only remove those that are published before.</p>\n\n<p>i don't think they know there are so many new duplicates.</p>",
          "rawMarkdown": "i don't think so. they can only remove those that are published before.\n\ni don't think they know there are so many new duplicates.",
          "votes": 3
        },
        {
          "id": 478341,
          "postDate": "2019-02-26T03:23:39.847Z",
          "rawMarkdown": "",
          "isDeleted": true
        },
        {
          "id": 478513,
          "postDate": "2019-02-26T08:52:59.057Z",
          "content": "<p>I bet no :)</p>",
          "rawMarkdown": "I bet no :)",
          "votes": 3
        },
        {
          "id": 478534,
          "postDate": "2019-02-26T09:49:35.327Z",
          "content": "<p>Well, I decide not to make any changes for the original labels. </p>",
          "rawMarkdown": "Well, I decide not to make any changes for the original labels. ",
          "votes": 1
        },
        {
          "id": 478709,
          "postDate": "2019-02-26T14:15:47.787Z",
          "content": "<p>Ignoring duplicated Ids might be the best way. I think original labels incorporate appearance (eg: color tone, angle of camera) in some way.</p>",
          "rawMarkdown": "Ignoring duplicated Ids might be the best way. I think original labels incorporate appearance (eg: color tone, angle of camera) in some way."
        }
      ]
    },
    {
      "id": 478918,
      "postDate": "2019-02-26T19:57:10.743Z",
      "content": "<p>Hi all,</p>\n\n<p>We also found some cases considering intersection between classes 'new_whale' and other ids. </p>\n\n<p>I'm really wondering how all these cases will be fairly evaluated. </p>\n\n<p><img src=\"https://storage.googleapis.com/kaggle-forum-message-attachments/478918/11433/same_nw_2.png\" alt=\"enter image description here\">\n<img src=\"https://storage.googleapis.com/kaggle-forum-message-attachments/478918/11432/same_nw_1.png\" alt=\"enter image description here\">\n<img src=\"https://storage.googleapis.com/kaggle-forum-message-attachments/478918/11431/nw_same.png\" alt=\"enter image description here\"></p>",
      "rawMarkdown": "Hi all,\n\nWe also found some cases considering intersection between classes 'new_whale' and other ids. \n\nI'm really wondering how all these cases will be fairly evaluated. \n\n![enter image description here][1]\n![enter image description here][2]\n![enter image description here][3]\n\n\n  [1]: https://storage.googleapis.com/kaggle-forum-message-attachments/478918/11433/same_nw_2.png\n  [2]: https://storage.googleapis.com/kaggle-forum-message-attachments/478918/11432/same_nw_1.png\n  [3]: https://storage.googleapis.com/kaggle-forum-message-attachments/478918/11431/nw_same.png\n\n",
      "votes": 4,
      "replies": [
        {
          "id": 479020,
          "postDate": "2019-02-26T23:26:29.820Z",
          "content": "<p>I’ve found at least 300 new whales that are not so new after all. </p>",
          "rawMarkdown": "I’ve found at least 300 new whales that are not so new after all. ",
          "votes": 4
        },
        {
          "id": 479102,
          "postDate": "2019-02-27T01:06:40.107Z",
          "content": "<p>i realise this too. in my training, all new-whale images are flipped to make sure they are really new-whales. Note that there are three kind of errors:</p>\n\n<ol>\n<li><p>same identity whale split into different id</p></li>\n<li><p>identity whale labeled as new whale</p></li>\n<li><p>different identity whale merge into same id (about  3 cases?)</p></li>\n</ol>\n\n<p>It is interesting to note that such noise are common problem in face recognition. Several open face recognition dataset are not cleaned as well. (e.g. LFW has some error  in the test set and hence if an algorithm  claims 100%, it is definitely in-accurate!)</p>\n\n<p>There are papers that deals with cleaning of the dataset and how to noise affect results, e.g</p>\n\n<p><a href=\"https://arxiv.org/abs/1807.11649\">https://arxiv.org/abs/1807.11649</a></p>\n\n<p>\"The Devil of Face Recognition is in the Noise\"</p>\n\n<p>in summary, handling label noise is part of face recogniton.</p>",
          "rawMarkdown": "i realise this too. in my training, all new-whale images are flipped to make sure they are really new-whales. Note that there are three kind of errors:\n\n1. same identity whale split into different id\n\n2. identity whale labeled as new whale\n\n3. different identity whale merge into same id (about  3 cases?)\n\nIt is interesting to note that such noise are common problem in face recognition. Several open face recognition dataset are not cleaned as well. (e.g. LFW has some error  in the test set and hence if an algorithm  claims 100%, it is definitely in-accurate!)\n\nThere are papers that deals with cleaning of the dataset and how to noise affect results, e.g\n\nhttps://arxiv.org/abs/1807.11649\n\n\"The Devil of Face Recognition is in the Noise\"\n\n\nin summary, handling label noise is part of face recogniton.",
          "votes": 3
        },
        {
          "id": 479596,
          "postDate": "2019-02-27T09:14:57.490Z",
          "content": "<blockquote>\n  <p>in summary, handling label noise is part of face recogniton.</p>\n</blockquote>\n\n<p><a href=\"/hengck23\">@hengck23</a> You are right - in real world. But we are at kaggle :)</p>",
          "rawMarkdown": "&gt;in summary, handling label noise is part of face recogniton.\n\n@hengck23 You are right - in real world. But we are at kaggle :)\n"
        }
      ]
    },
    {
      "id": 480242,
      "postDate": "2019-02-28T01:06:15.663Z",
      "content": "<p>A dictionary to merge duplicated whale IDs into a single ID for training (two lists were combined).\nIn addition to the evaluation problem, these duplicate whale IDs would have negative effect on hard negative mining-based training...</p>\n\n<p><code>\n    dup_id_dict = {\n        \"w_c7e1b12\": \"w_62b631e\",\n        \"w_c543f7c\": \"w_2d1d67a\",\n        \"w_a335fc2\": \"w_c99807e\",\n        \"w_f119baa\": \"w_3f213d5\",\n        \"w_2f350be\": \"w_6f22173\",\n        \"w_954fec8\": \"w_0ee7878\",\n        \"w_27272a5\": \"w_7e5b9da\",\n        \"w_b5e6c9c\": \"w_0a0c768\",\n        \"w_e20dcb5\": \"w_6822dbc\",\n        \"w_92a47eb\": \"w_ea6a9f4\",\n        \"w_5a3e0de\": \"w_0b398b2\",\n        \"w_53ce2e1\": \"w_59109e5\",\n        \"w_f91389a\": \"w_7d8c37c\",\n        \"w_1d532b4\": \"w_b54f70f\",\n        \"w_169a302\": \"w_feb4051\",\n        \"w_596f6bc\": \"w_d588fb4\",\n        \"w_2ed0d2f\": \"w_7537a43\",\n        \"w_4e00a18\": \"w_6c2aa5f\",\n        \"w_012678c\": \"w_f497e3f\",\n        \"w_6eb10d2\": \"w_2e5ad54\",\n        \"w_20950a9\": \"w_1b2bf0f\",\n        \"w_993320b\": \"w_6dbfd7a\",\n        \"w_21fd105\": \"w_04de239\",\n        \"w_8be0240\": \"w_dcbdd01\",\n        \"w_4135cb8\": \"w_8d4c9f7\",\n        \"w_fae4be5\": \"w_1d35a02\",\n        \"w_f84e6e1\": \"w_1d35a02\",\n        \"w_cded134\": \"w_a6df3a5\",\n        \"w_84b1b28\": \"w_163b2b8\",\n        \"w_297e53a\": \"w_9f2cfff\",\n        \"w_b43d0fa\": \"w_19f507e\",\n        \"w_e4a9205\": \"w_5c4c166\",\n        \"w_5fcf350\": \"w_e21b25f\",\n        \"w_bd18f3c\": \"w_185044d\",\n        \"w_3d7bc85\": \"w_cb0e8c2\",\n        \"w_2ac2bac\": \"w_3bf887a\",\n        \"w_2f8de4f\": \"w_f7234f1\",\n        \"w_7efe670\": \"w_597758e\",\n        \"w_28218e1\": \"w_62bb54b\",\n        \"w_c1b90d1\": \"w_950c471\",\n        \"w_94e8c02\": \"w_169253a\",\n        \"w_eb31833\": \"w_662c89b\",\n        \"w_a44a9b2\": \"w_6fe3566\",\n        \"w_5ce4848\": \"w_81ed5f8\",\n        \"w_b2ef717\": \"w_07d4b0a\",\n        \"w_808d40b\": \"w_b035775\",\n        \"w_c97ad94\": \"w_a3181a0\",\n        \"w_b82d0eb\": \"w_aa0dacc\",\n        \"w_298f605\": \"w_43614bc\",\n        \"w_bbf0c38\": \"w_32e705d\",\n        \"w_7d46ac3\": \"w_3729146\",\n        \"w_25d662d\": \"w_d2facb4\",\n        \"w_5010896\": \"w_a1a0090\",\n        \"w_8431ae8\": \"w_115e880\",\n        \"w_f5d4627\": \"w_2623921\",\n        \"w_b6e4761\": \"w_bfecb74\",\n        \"w_52162eb\": \"w_a112639\",\n        \"w_60b027a\": \"w_c87817e\",\n        \"w_022b708\": \"w_d215a68\",\n        \"w_da651f6\": \"w_d92b4c4\",\n        \"w_5072c08\": \"w_8694cb5\",\n        \"w_f94fc92\": \"w_228c7ee\",\n        \"w_0532483\": \"w_444b771\",\n        \"w_0cc0430\": \"w_97074d8\",\n        \"w_d7aad03\": \"w_e56d7b2\",\n        \"w_6525e6b\": \"w_f1b016e\",\n        \"w_ddda4a4\": \"w_9353e7c\",\n        \"w_ea2d2f9\": \"w_3789389\",\n        \"w_26ba5fd\": \"w_8daccba\",\n        \"w_792ffd7\": \"w_a39b2e2\",\n        \"w_e12b78d\": \"w_9ffc9aa\",\n        \"w_353d249\": \"w_af1d57b\",\n        \"w_4130233\": \"w_cc93694\",\n        \"w_c1715f5\": \"w_706e4fb\",\n        \"w_ac6aee2\": \"w_a8276b4\",\n    }\n</code></p>",
      "rawMarkdown": "A dictionary to merge duplicated whale IDs into a single ID for training (two lists were combined).\nIn addition to the evaluation problem, these duplicate whale IDs would have negative effect on hard negative mining-based training...\n\n```\n    dup_id_dict = {\n        \"w_c7e1b12\": \"w_62b631e\",\n        \"w_c543f7c\": \"w_2d1d67a\",\n        \"w_a335fc2\": \"w_c99807e\",\n        \"w_f119baa\": \"w_3f213d5\",\n        \"w_2f350be\": \"w_6f22173\",\n        \"w_954fec8\": \"w_0ee7878\",\n        \"w_27272a5\": \"w_7e5b9da\",\n        \"w_b5e6c9c\": \"w_0a0c768\",\n        \"w_e20dcb5\": \"w_6822dbc\",\n        \"w_92a47eb\": \"w_ea6a9f4\",\n        \"w_5a3e0de\": \"w_0b398b2\",\n        \"w_53ce2e1\": \"w_59109e5\",\n        \"w_f91389a\": \"w_7d8c37c\",\n        \"w_1d532b4\": \"w_b54f70f\",\n        \"w_169a302\": \"w_feb4051\",\n        \"w_596f6bc\": \"w_d588fb4\",\n        \"w_2ed0d2f\": \"w_7537a43\",\n        \"w_4e00a18\": \"w_6c2aa5f\",\n        \"w_012678c\": \"w_f497e3f\",\n        \"w_6eb10d2\": \"w_2e5ad54\",\n        \"w_20950a9\": \"w_1b2bf0f\",\n        \"w_993320b\": \"w_6dbfd7a\",\n        \"w_21fd105\": \"w_04de239\",\n        \"w_8be0240\": \"w_dcbdd01\",\n        \"w_4135cb8\": \"w_8d4c9f7\",\n        \"w_fae4be5\": \"w_1d35a02\",\n        \"w_f84e6e1\": \"w_1d35a02\",\n        \"w_cded134\": \"w_a6df3a5\",\n        \"w_84b1b28\": \"w_163b2b8\",\n        \"w_297e53a\": \"w_9f2cfff\",\n        \"w_b43d0fa\": \"w_19f507e\",\n        \"w_e4a9205\": \"w_5c4c166\",\n        \"w_5fcf350\": \"w_e21b25f\",\n        \"w_bd18f3c\": \"w_185044d\",\n        \"w_3d7bc85\": \"w_cb0e8c2\",\n        \"w_2ac2bac\": \"w_3bf887a\",\n        \"w_2f8de4f\": \"w_f7234f1\",\n        \"w_7efe670\": \"w_597758e\",\n        \"w_28218e1\": \"w_62bb54b\",\n        \"w_c1b90d1\": \"w_950c471\",\n        \"w_94e8c02\": \"w_169253a\",\n        \"w_eb31833\": \"w_662c89b\",\n        \"w_a44a9b2\": \"w_6fe3566\",\n        \"w_5ce4848\": \"w_81ed5f8\",\n        \"w_b2ef717\": \"w_07d4b0a\",\n        \"w_808d40b\": \"w_b035775\",\n        \"w_c97ad94\": \"w_a3181a0\",\n        \"w_b82d0eb\": \"w_aa0dacc\",\n        \"w_298f605\": \"w_43614bc\",\n        \"w_bbf0c38\": \"w_32e705d\",\n        \"w_7d46ac3\": \"w_3729146\",\n        \"w_25d662d\": \"w_d2facb4\",\n        \"w_5010896\": \"w_a1a0090\",\n        \"w_8431ae8\": \"w_115e880\",\n        \"w_f5d4627\": \"w_2623921\",\n        \"w_b6e4761\": \"w_bfecb74\",\n        \"w_52162eb\": \"w_a112639\",\n        \"w_60b027a\": \"w_c87817e\",\n        \"w_022b708\": \"w_d215a68\",\n        \"w_da651f6\": \"w_d92b4c4\",\n        \"w_5072c08\": \"w_8694cb5\",\n        \"w_f94fc92\": \"w_228c7ee\",\n        \"w_0532483\": \"w_444b771\",\n        \"w_0cc0430\": \"w_97074d8\",\n        \"w_d7aad03\": \"w_e56d7b2\",\n        \"w_6525e6b\": \"w_f1b016e\",\n        \"w_ddda4a4\": \"w_9353e7c\",\n        \"w_ea2d2f9\": \"w_3789389\",\n        \"w_26ba5fd\": \"w_8daccba\",\n        \"w_792ffd7\": \"w_a39b2e2\",\n        \"w_e12b78d\": \"w_9ffc9aa\",\n        \"w_353d249\": \"w_af1d57b\",\n        \"w_4130233\": \"w_cc93694\",\n        \"w_c1715f5\": \"w_706e4fb\",\n        \"w_ac6aee2\": \"w_a8276b4\",\n    }\n```",
      "votes": 1
    },
    {
      "id": 478478,
      "postDate": "2019-02-26T07:51:43.803Z",
      "content": "<p>Thanks for the post. What do you mean by </p>\n\n<blockquote>\n  <p>they are not recognized by the assessment system as the same whale</p>\n</blockquote>",
      "rawMarkdown": "Thanks for the post. What do you mean by \n\n&gt; they are not recognized by the assessment system as the same whale\n\n",
      "votes": 1,
      "replies": [
        {
          "id": 478518,
          "postDate": "2019-02-26T09:02:10.047Z",
          "content": "<p>If you use only one of the labels of the whales in your submission you will get significantly worse results.</p>",
          "rawMarkdown": "If you use only one of the labels of the whales in your submission you will get significantly worse results.",
          "votes": 2
        }
      ]
    },
    {
      "id": 478237,
      "postDate": "2019-02-25T23:09:17.723Z",
      "content": "<p>The number of duplicated IDs is larger than I expected. What we can do is put together them in TOP5 prediction 😟</p>",
      "rawMarkdown": "The number of duplicated IDs is larger than I expected. What we can do is put together them in TOP5 prediction 😟",
      "votes": 1
    },
    {
      "id": 478939,
      "postDate": "2019-02-26T20:23:15.547Z",
      "content": "<p>And also our extension to the list of duplicates: </p>\n\n<pre><code>w_c7e1b12 w_62b631e\nw_c543f7c w_2d1d67a\nw_a335fc2 w_c99807e\nw_f119baa w_3f213d5\nw_2f350be w_6f22173\nw_954fec8 w_0ee7878\nw_27272a5 w_7e5b9da\nw_b5e6c9c w_0a0c768\nw_e20dcb5 w_6822dbc\nw_92a47eb w_ea6a9f4\nw_5a3e0de w_0b398b2\nw_53ce2e1 w_59109e5\nw_f91389a w_7d8c37c\nw_1d532b4 w_b54f70f\nw_169a302 w_feb4051\nw_596f6bc w_d588fb4\nw_2ed0d2f w_7537a43\nw_4e00a18 w_6c2aa5f\nw_012678c w_f497e3f\nw_6eb10d2 w_2e5ad54\nw_20950a9 w_1b2bf0f\nw_993320b w_6dbfd7a\nw_21fd105 w_04de239\nw_8be0240 w_dcbdd01\nw_4135cb8 w_8d4c9f7\nw_1d35a02 w_f84e6e1\nw_cded134 w_a6df3a5\n</code></pre>",
      "rawMarkdown": "And also our extension to the list of duplicates: \n\n    w_c7e1b12 w_62b631e\n    w_c543f7c w_2d1d67a\n    w_a335fc2 w_c99807e\n    w_f119baa w_3f213d5\n    w_2f350be w_6f22173\n    w_954fec8 w_0ee7878\n    w_27272a5 w_7e5b9da\n    w_b5e6c9c w_0a0c768\n    w_e20dcb5 w_6822dbc\n    w_92a47eb w_ea6a9f4\n    w_5a3e0de w_0b398b2\n    w_53ce2e1 w_59109e5\n    w_f91389a w_7d8c37c\n    w_1d532b4 w_b54f70f\n    w_169a302 w_feb4051\n    w_596f6bc w_d588fb4\n    w_2ed0d2f w_7537a43\n    w_4e00a18 w_6c2aa5f\n    w_012678c w_f497e3f\n    w_6eb10d2 w_2e5ad54\n    w_20950a9 w_1b2bf0f\n    w_993320b w_6dbfd7a\n    w_21fd105 w_04de239\n    w_8be0240 w_dcbdd01\n    w_4135cb8 w_8d4c9f7\n    w_1d35a02 w_f84e6e1\n    w_cded134 w_a6df3a5\n\n",
      "votes": 2
    },
    {
      "id": 478343,
      "postDate": "2019-02-26T03:26:09.250Z",
      "content": "<p>How did you know about them? Did you check them manually?</p>\n\n<p>I used  <a href=\"https://www.kaggle.com/c/humpback-whale-identification/discussion/80379\">my code</a>  to sort the most wrong labels and could come up with ~80% of your list. </p>\n\n<p>I got significantly more images that are not on your list.. \nRunning the code should be easy and you can get your own analysis with sorting from the most wrong to the most confused to the most right labels.. And the function is fairly generalized to be used for other competitions.. </p>\n\n<p>One such triple wrong labeled example that is <strong>not</strong> available in your list:</p>\n\n<p><img src=\"https://www.dropbox.com/s/bkhvbip2peyv2lt/000000.51074_new_whale_w_07125ed_17bb74c11.jpg?dl=1\" alt=\"enter image description here\"></p>",
      "rawMarkdown": "How did you know about them? Did you check them manually?\n\nI used  [my code][1]  to sort the most wrong labels and could come up with ~80% of your list. \n\nI got significantly more images that are not on your list.. \nRunning the code should be easy and you can get your own analysis with sorting from the most wrong to the most confused to the most right labels.. And the function is fairly generalized to be used for other competitions.. \n\nOne such triple wrong labeled example that is **not** available in your list:\n\n![enter image description here][2]\n\n\n  [1]: https://www.kaggle.com/c/humpback-whale-identification/discussion/80379\n  [2]: https://www.dropbox.com/s/bkhvbip2peyv2lt/000000.51074_new_whale_w_07125ed_17bb74c11.jpg?dl=1",
      "votes": 2,
      "replies": [
        {
          "id": 478519,
          "postDate": "2019-02-26T09:04:38.900Z",
          "content": "<p>These are the IDs that my model found. There may be much more.</p>",
          "rawMarkdown": "These are the IDs that my model found. There may be much more.",
          "votes": 1
        }
      ]
    },
    {
      "id": 478638,
      "postDate": "2019-02-26T12:38:37.037Z",
      "rawMarkdown": "",
      "isDeleted": true
    },
    {
      "id": 480006,
      "postDate": "2019-02-27T17:06:43.527Z",
      "content": "<p>Thanks~</p>",
      "rawMarkdown": "Thanks~"
    },
    {
      "id": 478987,
      "postDate": "2019-02-26T22:09:13.137Z",
      "content": "<p>Thanks for sharing</p>",
      "rawMarkdown": "Thanks for sharing"
    },
    {
      "id": 478644,
      "postDate": "2019-02-26T12:42:00.007Z",
      "content": "<p>Thanks</p>",
      "rawMarkdown": "Thanks"
    }
  ],
  "comments": [
    {
      "id": 478231,
      "author_name": "hengck23",
      "author_url": "",
      "post_date": "2019-02-25T22:56:59.747000",
      "content": "<p>thanks. this will affect the LB score so there is an element of luck here.</p>",
      "votes": 3,
      "replies": [
        {
          "id": 478315,
          "author_name": "Yiheng Wang",
          "author_url": "",
          "post_date": "2019-02-26T02:25:51.033000",
          "content": "<p>I bet all samples with ambiguous labels are removed from private test set, thus don't worry : )</p>",
          "votes": -1,
          "replies": []
        },
        {
          "id": 478334,
          "author_name": "hengck23",
          "author_url": "",
          "post_date": "2019-02-26T03:08:29.743000",
          "content": "<p>i don't think so. they can only remove those that are published before.</p>\n\n<p>i don't think they know there are so many new duplicates.</p>",
          "votes": 3,
          "replies": []
        },
        {
          "id": 478341,
          "author_name": "",
          "author_url": "",
          "post_date": "2019-02-26T03:23:39.847000",
          "content": "",
          "votes": 0,
          "replies": []
        },
        {
          "id": 478513,
          "author_name": "old-ufo",
          "author_url": "",
          "post_date": "2019-02-26T08:52:59.057000",
          "content": "<p>I bet no :)</p>",
          "votes": 3,
          "replies": []
        },
        {
          "id": 478534,
          "author_name": "Yiheng Wang",
          "author_url": "",
          "post_date": "2019-02-26T09:49:35.327000",
          "content": "<p>Well, I decide not to make any changes for the original labels. </p>",
          "votes": 1,
          "replies": []
        },
        {
          "id": 478709,
          "author_name": "toshi_k",
          "author_url": "",
          "post_date": "2019-02-26T14:15:47.787000",
          "content": "<p>Ignoring duplicated Ids might be the best way. I think original labels incorporate appearance (eg: color tone, angle of camera) in some way.</p>",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 478918,
      "author_name": "Sorceress",
      "author_url": "",
      "post_date": "2019-02-26T19:57:10.743000",
      "content": "<p>Hi all,</p>\n\n<p>We also found some cases considering intersection between classes 'new_whale' and other ids. </p>\n\n<p>I'm really wondering how all these cases will be fairly evaluated. </p>\n\n<p><img src=\"https://storage.googleapis.com/kaggle-forum-message-attachments/478918/11433/same_nw_2.png\" alt=\"enter image description here\">\n<img src=\"https://storage.googleapis.com/kaggle-forum-message-attachments/478918/11432/same_nw_1.png\" alt=\"enter image description here\">\n<img src=\"https://storage.googleapis.com/kaggle-forum-message-attachments/478918/11431/nw_same.png\" alt=\"enter image description here\"></p>",
      "votes": 4,
      "replies": [
        {
          "id": 479020,
          "author_name": "interneuron",
          "author_url": "",
          "post_date": "2019-02-26T23:26:29.820000",
          "content": "<p>I’ve found at least 300 new whales that are not so new after all. </p>",
          "votes": 4,
          "replies": []
        },
        {
          "id": 479102,
          "author_name": "hengck23",
          "author_url": "",
          "post_date": "2019-02-27T01:06:40.107000",
          "content": "<p>i realise this too. in my training, all new-whale images are flipped to make sure they are really new-whales. Note that there are three kind of errors:</p>\n\n<ol>\n<li><p>same identity whale split into different id</p></li>\n<li><p>identity whale labeled as new whale</p></li>\n<li><p>different identity whale merge into same id (about  3 cases?)</p></li>\n</ol>\n\n<p>It is interesting to note that such noise are common problem in face recognition. Several open face recognition dataset are not cleaned as well. (e.g. LFW has some error  in the test set and hence if an algorithm  claims 100%, it is definitely in-accurate!)</p>\n\n<p>There are papers that deals with cleaning of the dataset and how to noise affect results, e.g</p>\n\n<p><a href=\"https://arxiv.org/abs/1807.11649\">https://arxiv.org/abs/1807.11649</a></p>\n\n<p>\"The Devil of Face Recognition is in the Noise\"</p>\n\n<p>in summary, handling label noise is part of face recogniton.</p>",
          "votes": 3,
          "replies": []
        },
        {
          "id": 479596,
          "author_name": "old-ufo",
          "author_url": "",
          "post_date": "2019-02-27T09:14:57.490000",
          "content": "<blockquote>\n  <p>in summary, handling label noise is part of face recogniton.</p>\n</blockquote>\n\n<p><a href=\"/hengck23\">@hengck23</a> You are right - in real world. But we are at kaggle :)</p>",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 480242,
      "author_name": "yu4u",
      "author_url": "",
      "post_date": "2019-02-28T01:06:15.663000",
      "content": "<p>A dictionary to merge duplicated whale IDs into a single ID for training (two lists were combined).\nIn addition to the evaluation problem, these duplicate whale IDs would have negative effect on hard negative mining-based training...</p>\n\n<p><code>\n    dup_id_dict = {\n        \"w_c7e1b12\": \"w_62b631e\",\n        \"w_c543f7c\": \"w_2d1d67a\",\n        \"w_a335fc2\": \"w_c99807e\",\n        \"w_f119baa\": \"w_3f213d5\",\n        \"w_2f350be\": \"w_6f22173\",\n        \"w_954fec8\": \"w_0ee7878\",\n        \"w_27272a5\": \"w_7e5b9da\",\n        \"w_b5e6c9c\": \"w_0a0c768\",\n        \"w_e20dcb5\": \"w_6822dbc\",\n        \"w_92a47eb\": \"w_ea6a9f4\",\n        \"w_5a3e0de\": \"w_0b398b2\",\n        \"w_53ce2e1\": \"w_59109e5\",\n        \"w_f91389a\": \"w_7d8c37c\",\n        \"w_1d532b4\": \"w_b54f70f\",\n        \"w_169a302\": \"w_feb4051\",\n        \"w_596f6bc\": \"w_d588fb4\",\n        \"w_2ed0d2f\": \"w_7537a43\",\n        \"w_4e00a18\": \"w_6c2aa5f\",\n        \"w_012678c\": \"w_f497e3f\",\n        \"w_6eb10d2\": \"w_2e5ad54\",\n        \"w_20950a9\": \"w_1b2bf0f\",\n        \"w_993320b\": \"w_6dbfd7a\",\n        \"w_21fd105\": \"w_04de239\",\n        \"w_8be0240\": \"w_dcbdd01\",\n        \"w_4135cb8\": \"w_8d4c9f7\",\n        \"w_fae4be5\": \"w_1d35a02\",\n        \"w_f84e6e1\": \"w_1d35a02\",\n        \"w_cded134\": \"w_a6df3a5\",\n        \"w_84b1b28\": \"w_163b2b8\",\n        \"w_297e53a\": \"w_9f2cfff\",\n        \"w_b43d0fa\": \"w_19f507e\",\n        \"w_e4a9205\": \"w_5c4c166\",\n        \"w_5fcf350\": \"w_e21b25f\",\n        \"w_bd18f3c\": \"w_185044d\",\n        \"w_3d7bc85\": \"w_cb0e8c2\",\n        \"w_2ac2bac\": \"w_3bf887a\",\n        \"w_2f8de4f\": \"w_f7234f1\",\n        \"w_7efe670\": \"w_597758e\",\n        \"w_28218e1\": \"w_62bb54b\",\n        \"w_c1b90d1\": \"w_950c471\",\n        \"w_94e8c02\": \"w_169253a\",\n        \"w_eb31833\": \"w_662c89b\",\n        \"w_a44a9b2\": \"w_6fe3566\",\n        \"w_5ce4848\": \"w_81ed5f8\",\n        \"w_b2ef717\": \"w_07d4b0a\",\n        \"w_808d40b\": \"w_b035775\",\n        \"w_c97ad94\": \"w_a3181a0\",\n        \"w_b82d0eb\": \"w_aa0dacc\",\n        \"w_298f605\": \"w_43614bc\",\n        \"w_bbf0c38\": \"w_32e705d\",\n        \"w_7d46ac3\": \"w_3729146\",\n        \"w_25d662d\": \"w_d2facb4\",\n        \"w_5010896\": \"w_a1a0090\",\n        \"w_8431ae8\": \"w_115e880\",\n        \"w_f5d4627\": \"w_2623921\",\n        \"w_b6e4761\": \"w_bfecb74\",\n        \"w_52162eb\": \"w_a112639\",\n        \"w_60b027a\": \"w_c87817e\",\n        \"w_022b708\": \"w_d215a68\",\n        \"w_da651f6\": \"w_d92b4c4\",\n        \"w_5072c08\": \"w_8694cb5\",\n        \"w_f94fc92\": \"w_228c7ee\",\n        \"w_0532483\": \"w_444b771\",\n        \"w_0cc0430\": \"w_97074d8\",\n        \"w_d7aad03\": \"w_e56d7b2\",\n        \"w_6525e6b\": \"w_f1b016e\",\n        \"w_ddda4a4\": \"w_9353e7c\",\n        \"w_ea2d2f9\": \"w_3789389\",\n        \"w_26ba5fd\": \"w_8daccba\",\n        \"w_792ffd7\": \"w_a39b2e2\",\n        \"w_e12b78d\": \"w_9ffc9aa\",\n        \"w_353d249\": \"w_af1d57b\",\n        \"w_4130233\": \"w_cc93694\",\n        \"w_c1715f5\": \"w_706e4fb\",\n        \"w_ac6aee2\": \"w_a8276b4\",\n    }\n</code></p>",
      "votes": 1,
      "replies": []
    },
    {
      "id": 478478,
      "author_name": "good good study",
      "author_url": "",
      "post_date": "2019-02-26T07:51:43.803000",
      "content": "<p>Thanks for the post. What do you mean by </p>\n\n<blockquote>\n  <p>they are not recognized by the assessment system as the same whale</p>\n</blockquote>",
      "votes": 1,
      "replies": [
        {
          "id": 478518,
          "author_name": "Oleksandr Mokin",
          "author_url": "",
          "post_date": "2019-02-26T09:02:10.047000",
          "content": "<p>If you use only one of the labels of the whales in your submission you will get significantly worse results.</p>",
          "votes": 2,
          "replies": []
        }
      ]
    },
    {
      "id": 478237,
      "author_name": "toshi_k",
      "author_url": "",
      "post_date": "2019-02-25T23:09:17.723000",
      "content": "<p>The number of duplicated IDs is larger than I expected. What we can do is put together them in TOP5 prediction 😟</p>",
      "votes": 1,
      "replies": []
    },
    {
      "id": 478939,
      "author_name": "Sorceress",
      "author_url": "",
      "post_date": "2019-02-26T20:23:15.547000",
      "content": "<p>And also our extension to the list of duplicates: </p>\n\n<pre><code>w_c7e1b12 w_62b631e\nw_c543f7c w_2d1d67a\nw_a335fc2 w_c99807e\nw_f119baa w_3f213d5\nw_2f350be w_6f22173\nw_954fec8 w_0ee7878\nw_27272a5 w_7e5b9da\nw_b5e6c9c w_0a0c768\nw_e20dcb5 w_6822dbc\nw_92a47eb w_ea6a9f4\nw_5a3e0de w_0b398b2\nw_53ce2e1 w_59109e5\nw_f91389a w_7d8c37c\nw_1d532b4 w_b54f70f\nw_169a302 w_feb4051\nw_596f6bc w_d588fb4\nw_2ed0d2f w_7537a43\nw_4e00a18 w_6c2aa5f\nw_012678c w_f497e3f\nw_6eb10d2 w_2e5ad54\nw_20950a9 w_1b2bf0f\nw_993320b w_6dbfd7a\nw_21fd105 w_04de239\nw_8be0240 w_dcbdd01\nw_4135cb8 w_8d4c9f7\nw_1d35a02 w_f84e6e1\nw_cded134 w_a6df3a5\n</code></pre>",
      "votes": 2,
      "replies": []
    },
    {
      "id": 478343,
      "author_name": "Haider Alwasiti",
      "author_url": "",
      "post_date": "2019-02-26T03:26:09.250000",
      "content": "<p>How did you know about them? Did you check them manually?</p>\n\n<p>I used  <a href=\"https://www.kaggle.com/c/humpback-whale-identification/discussion/80379\">my code</a>  to sort the most wrong labels and could come up with ~80% of your list. </p>\n\n<p>I got significantly more images that are not on your list.. \nRunning the code should be easy and you can get your own analysis with sorting from the most wrong to the most confused to the most right labels.. And the function is fairly generalized to be used for other competitions.. </p>\n\n<p>One such triple wrong labeled example that is <strong>not</strong> available in your list:</p>\n\n<p><img src=\"https://www.dropbox.com/s/bkhvbip2peyv2lt/000000.51074_new_whale_w_07125ed_17bb74c11.jpg?dl=1\" alt=\"enter image description here\"></p>",
      "votes": 2,
      "replies": [
        {
          "id": 478519,
          "author_name": "Oleksandr Mokin",
          "author_url": "",
          "post_date": "2019-02-26T09:04:38.900000",
          "content": "<p>These are the IDs that my model found. There may be much more.</p>",
          "votes": 1,
          "replies": []
        }
      ]
    },
    {
      "id": 478638,
      "author_name": "",
      "author_url": "",
      "post_date": "2019-02-26T12:38:37.037000",
      "content": "",
      "votes": 0,
      "replies": []
    },
    {
      "id": 480006,
      "author_name": "Mars2020",
      "author_url": "",
      "post_date": "2019-02-27T17:06:43.527000",
      "content": "<p>Thanks~</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 478987,
      "author_name": "klickmal",
      "author_url": "",
      "post_date": "2019-02-26T22:09:13.137000",
      "content": "<p>Thanks for sharing</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 478644,
      "author_name": "< Afzal Sayed >",
      "author_url": "",
      "post_date": "2019-02-26T12:42:00.007000",
      "content": "<p>Thanks</p>",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "478168": "Here are the ones that I managed to find. Yes, quite a lot. And unfortunately, they are not recognized by the assessment system as the same whale. So be careful.\n\n    w_84b1b28,w_163b2b8\n    w_297e53a,w_9f2cfff\n    w_b43d0fa,w_19f507e\n    w_e4a9205,w_5c4c166\n    w_5fcf350,w_e21b25f\n    w_bd18f3c,w_185044d\n    w_3d7bc85,w_cb0e8c2\n    w_fae4be5,w_f84e6e1,w_1d35a02\n    w_2ac2bac,w_3bf887a\n    w_2f8de4f,w_f7234f1\n    w_7efe670,w_597758e\n    w_28218e1,w_62bb54b\n    w_c1b90d1,w_950c471\n    w_94e8c02,w_169253a\n    w_eb31833,w_662c89b\n    w_a44a9b2,w_6fe3566\n    w_5ce4848,w_81ed5f8\n    w_b2ef717,w_07d4b0a\n    w_808d40b,w_b035775\n    w_c97ad94,w_a3181a0\n    w_b82d0eb,w_aa0dacc\n    w_298f605,w_43614bc\n    w_bbf0c38,w_32e705d\n    w_7d46ac3,w_3729146\n    w_25d662d,w_d2facb4\n    w_5010896,w_a1a0090\n    w_8431ae8,w_115e880\n    w_f5d4627,w_2623921\n    w_b6e4761,w_bfecb74\n    w_52162eb,w_a112639\n    w_60b027a,w_c87817e\n    w_022b708,w_d215a68\n    w_da651f6,w_d92b4c4\n    w_5072c08,w_8694cb5\n    w_f94fc92,w_228c7ee\n    w_0532483,w_444b771\n    w_0cc0430,w_97074d8\n    w_d7aad03,w_e56d7b2\n    w_6525e6b,w_f1b016e\n    w_ddda4a4,w_9353e7c\n    w_ea2d2f9,w_3789389\n    w_26ba5fd,w_8daccba\n    w_792ffd7,w_a39b2e2\n    w_e12b78d,w_9ffc9aa\n    w_353d249,w_af1d57b\n    w_4130233,w_cc93694\n    w_c1715f5,w_706e4fb\n    w_ac6aee2,w_a8276b4\n\n",
    "478231": "thanks. this will affect the LB score so there is an element of luck here.",
    "478918": "Hi all,\n\nWe also found some cases considering intersection between classes 'new_whale' and other ids. \n\nI'm really wondering how all these cases will be fairly evaluated. \n\n![enter image description here][1]\n![enter image description here][2]\n![enter image description here][3]\n\n\n  [1]: https://storage.googleapis.com/kaggle-forum-message-attachments/478918/11433/same_nw_2.png\n  [2]: https://storage.googleapis.com/kaggle-forum-message-attachments/478918/11432/same_nw_1.png\n  [3]: https://storage.googleapis.com/kaggle-forum-message-attachments/478918/11431/nw_same.png\n\n",
    "480242": "A dictionary to merge duplicated whale IDs into a single ID for training (two lists were combined).\nIn addition to the evaluation problem, these duplicate whale IDs would have negative effect on hard negative mining-based training...\n\n```\n    dup_id_dict = {\n        \"w_c7e1b12\": \"w_62b631e\",\n        \"w_c543f7c\": \"w_2d1d67a\",\n        \"w_a335fc2\": \"w_c99807e\",\n        \"w_f119baa\": \"w_3f213d5\",\n        \"w_2f350be\": \"w_6f22173\",\n        \"w_954fec8\": \"w_0ee7878\",\n        \"w_27272a5\": \"w_7e5b9da\",\n        \"w_b5e6c9c\": \"w_0a0c768\",\n        \"w_e20dcb5\": \"w_6822dbc\",\n        \"w_92a47eb\": \"w_ea6a9f4\",\n        \"w_5a3e0de\": \"w_0b398b2\",\n        \"w_53ce2e1\": \"w_59109e5\",\n        \"w_f91389a\": \"w_7d8c37c\",\n        \"w_1d532b4\": \"w_b54f70f\",\n        \"w_169a302\": \"w_feb4051\",\n        \"w_596f6bc\": \"w_d588fb4\",\n        \"w_2ed0d2f\": \"w_7537a43\",\n        \"w_4e00a18\": \"w_6c2aa5f\",\n        \"w_012678c\": \"w_f497e3f\",\n        \"w_6eb10d2\": \"w_2e5ad54\",\n        \"w_20950a9\": \"w_1b2bf0f\",\n        \"w_993320b\": \"w_6dbfd7a\",\n        \"w_21fd105\": \"w_04de239\",\n        \"w_8be0240\": \"w_dcbdd01\",\n        \"w_4135cb8\": \"w_8d4c9f7\",\n        \"w_fae4be5\": \"w_1d35a02\",\n        \"w_f84e6e1\": \"w_1d35a02\",\n        \"w_cded134\": \"w_a6df3a5\",\n        \"w_84b1b28\": \"w_163b2b8\",\n        \"w_297e53a\": \"w_9f2cfff\",\n        \"w_b43d0fa\": \"w_19f507e\",\n        \"w_e4a9205\": \"w_5c4c166\",\n        \"w_5fcf350\": \"w_e21b25f\",\n        \"w_bd18f3c\": \"w_185044d\",\n        \"w_3d7bc85\": \"w_cb0e8c2\",\n        \"w_2ac2bac\": \"w_3bf887a\",\n        \"w_2f8de4f\": \"w_f7234f1\",\n        \"w_7efe670\": \"w_597758e\",\n        \"w_28218e1\": \"w_62bb54b\",\n        \"w_c1b90d1\": \"w_950c471\",\n        \"w_94e8c02\": \"w_169253a\",\n        \"w_eb31833\": \"w_662c89b\",\n        \"w_a44a9b2\": \"w_6fe3566\",\n        \"w_5ce4848\": \"w_81ed5f8\",\n        \"w_b2ef717\": \"w_07d4b0a\",\n        \"w_808d40b\": \"w_b035775\",\n        \"w_c97ad94\": \"w_a3181a0\",\n        \"w_b82d0eb\": \"w_aa0dacc\",\n        \"w_298f605\": \"w_43614bc\",\n        \"w_bbf0c38\": \"w_32e705d\",\n        \"w_7d46ac3\": \"w_3729146\",\n        \"w_25d662d\": \"w_d2facb4\",\n        \"w_5010896\": \"w_a1a0090\",\n        \"w_8431ae8\": \"w_115e880\",\n        \"w_f5d4627\": \"w_2623921\",\n        \"w_b6e4761\": \"w_bfecb74\",\n        \"w_52162eb\": \"w_a112639\",\n        \"w_60b027a\": \"w_c87817e\",\n        \"w_022b708\": \"w_d215a68\",\n        \"w_da651f6\": \"w_d92b4c4\",\n        \"w_5072c08\": \"w_8694cb5\",\n        \"w_f94fc92\": \"w_228c7ee\",\n        \"w_0532483\": \"w_444b771\",\n        \"w_0cc0430\": \"w_97074d8\",\n        \"w_d7aad03\": \"w_e56d7b2\",\n        \"w_6525e6b\": \"w_f1b016e\",\n        \"w_ddda4a4\": \"w_9353e7c\",\n        \"w_ea2d2f9\": \"w_3789389\",\n        \"w_26ba5fd\": \"w_8daccba\",\n        \"w_792ffd7\": \"w_a39b2e2\",\n        \"w_e12b78d\": \"w_9ffc9aa\",\n        \"w_353d249\": \"w_af1d57b\",\n        \"w_4130233\": \"w_cc93694\",\n        \"w_c1715f5\": \"w_706e4fb\",\n        \"w_ac6aee2\": \"w_a8276b4\",\n    }\n```",
    "478478": "Thanks for the post. What do you mean by \n\n&gt; they are not recognized by the assessment system as the same whale\n\n",
    "478237": "The number of duplicated IDs is larger than I expected. What we can do is put together them in TOP5 prediction 😟",
    "478939": "And also our extension to the list of duplicates: \n\n    w_c7e1b12 w_62b631e\n    w_c543f7c w_2d1d67a\n    w_a335fc2 w_c99807e\n    w_f119baa w_3f213d5\n    w_2f350be w_6f22173\n    w_954fec8 w_0ee7878\n    w_27272a5 w_7e5b9da\n    w_b5e6c9c w_0a0c768\n    w_e20dcb5 w_6822dbc\n    w_92a47eb w_ea6a9f4\n    w_5a3e0de w_0b398b2\n    w_53ce2e1 w_59109e5\n    w_f91389a w_7d8c37c\n    w_1d532b4 w_b54f70f\n    w_169a302 w_feb4051\n    w_596f6bc w_d588fb4\n    w_2ed0d2f w_7537a43\n    w_4e00a18 w_6c2aa5f\n    w_012678c w_f497e3f\n    w_6eb10d2 w_2e5ad54\n    w_20950a9 w_1b2bf0f\n    w_993320b w_6dbfd7a\n    w_21fd105 w_04de239\n    w_8be0240 w_dcbdd01\n    w_4135cb8 w_8d4c9f7\n    w_1d35a02 w_f84e6e1\n    w_cded134 w_a6df3a5\n\n",
    "478343": "How did you know about them? Did you check them manually?\n\nI used  [my code][1]  to sort the most wrong labels and could come up with ~80% of your list. \n\nI got significantly more images that are not on your list.. \nRunning the code should be easy and you can get your own analysis with sorting from the most wrong to the most confused to the most right labels.. And the function is fairly generalized to be used for other competitions.. \n\nOne such triple wrong labeled example that is **not** available in your list:\n\n![enter image description here][2]\n\n\n  [1]: https://www.kaggle.com/c/humpback-whale-identification/discussion/80379\n  [2]: https://www.dropbox.com/s/bkhvbip2peyv2lt/000000.51074_new_whale_w_07125ed_17bb74c11.jpg?dl=1",
    "478638": "",
    "480006": "Thanks~",
    "478987": "Thanks for sharing",
    "478644": "Thanks"
  }
}