![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2594941%2Fff42cda2aa3da8c9679a118572758332%2F41c0123cc.jpg?generation=1592430201106442&alt=media\")
\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2594941%2F8457982067947413f1fc0b597839673a%2F409a8490c.jpg?generation=1592430201086194&alt=media\")
\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2594941%2Fc6068002843c8af6e930b5e121baaa18%2Fd067ac2b1.jpg?generation=1592430201241471&alt=media\")
\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2594941%2F2daf1ed173874755161a8be0f23e4a28%2F2cc75e9f5.jpg?generation=1592430201423627&alt=media\")
\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2594941%2F7f0eff0b0a9f37878cba2d362b48f4c0%2F893938464.jpg?generation=1592430201433365&alt=media\")
\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2594941%2Febc45233e34ab988476eb64fc76668a2%2F42e6efaaa.jpg?generation=1592430201488989&alt=media\")
\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2594941%2F79250b5c797cff009ed602b5d32245bb%2Fa1321ca95.jpg?generation=1592430201785339&alt=media\")
\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2594941%2Fd1340daf6a8ca35375afa8544fc0f1e4%2Fd60e832a5.jpg?generation=1592430202436979&alt=media\")
So, I've looked at images with bboxes. And created a list of huge bboxes without wheat in them. There it is:\n\n41c0123cc.jpg\na1321ca95.jpg\n2cc75e9f5.jpg\n42e6efaaa.jpg\n409a8490c.jpg\nd067ac2b1.jpg\nd60e832a5.jpg\n893938464.jpg\n
If somebody will find more of them, please, add in comments. Thanks to @artgor for mentioning this issue.\nImages:
\n\n\n\n\n\n\n\n\n
So, I've looked at images with bboxes. And created a list of huge bboxes without wheat in them. There it is:\n\n41c0123cc.jpg\na1321ca95.jpg\n2cc75e9f5.jpg\n42e6efaaa.jpg\n409a8490c.jpg\nd067ac2b1.jpg\nd60e832a5.jpg\n893938464.jpg\n
If somebody will find more of them, please, add in comments. Thanks to @artgor for mentioning this issue.\nImages:
\n\n\n\n\n\n\n\n\n
Quick comb through and you missed a few images
\n\n9a30dd802.jpg
\n\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F3543139%2F411151e9aba835e682c77101b8e4afe3%2Ff6ee205e130aa6667783099226d6a433.png?generation=1592444356331280&alt=media\")
d7a02151d.jpg
\n\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F3543139%2F685b91e6b0b08d4b561f1c48c625b353%2F2b095055faca0bf464e39e42206734df.png?generation=1592444575783772&alt=media\")
dcafcae79.jpg
\n\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F3543139%2F866edb5ce33df7e40c3c3a0fb2ca8e18%2F000911c6b3ee0b5a3bdfa94303263ce5.png?generation=1592444588680453&alt=media\")
Also here is a list of the indices of the boxes in train.csv so you don't have to remove the entire image from training
\n\n\n3687\n117344\n173\n113947\n52868\n2159\n2169\n121633\n121634\n147504\n118211\n52727\n147552\n\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F3543139%2F9afac44dddb7c21a6397bd2042b788b8%2F1fd440b56b5a2602c50ad52ca5162103.png?generation=1592444459264846&alt=media\")
@greatgamedota Cool!
", "rawMarkdown": "@greatgamedota Cool!" }, { "id": 901315, "postDate": "2020-06-25T11:39:50.867Z", "content": "bad_boxes = [3687, 117344,173,113947,52868,2159,2169,121633,121634,147504,118211,52727,147552]
", "rawMarkdown": "bad_boxes = [3687, 117344,173,113947,52868,2159,2169,121633,121634,147504,118211,52727,147552]", "votes": 3 }, { "id": 942803, "postDate": "2020-07-24T02:26:58.713Z", "content": "Might I inquire as to how you're styling that notebook—I'm assuming some extension?
", "rawMarkdown": "Might I inquire as to how you're styling that notebook—I'm assuming some extension?" }, { "id": 942836, "postDate": "2020-07-24T03:12:57.260Z", "content": "I would assume they are using an extension like this.
", "rawMarkdown": "I would assume they are using an extension like [this](https://chrome.google.com/webstore/detail/dark-reader/eimadpbcbfnmbkopoojfekhnkhdbieeh?hl=en)." }, { "id": 943611, "postDate": "2020-07-24T13:21:26.563Z", "content": "That screenshot is actually Kaggle's default Dark Mode for kernals, but I am using that extension currently.
", "rawMarkdown": "That screenshot is actually Kaggle's default Dark Mode for kernals, but I am using that extension currently.", "votes": 1 } ] }, { "id": 897993, "postDate": "2020-06-23T08:03:11.943Z", "content": "![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2033774%2Fc81bd14c72aff677d961eddc28a85489%2F1234.PNG?generation=1592899558943065&alt=media\")
\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2033774%2Fc361b7838cde8f3042e7742e4bdd0e5b%2F123.PNG?generation=1592899559731908&alt=media\")
Dear @vadbeg
\n\nSorry for the inconvenience. Some answers to the bug here: https://www.kaggle.com/c/global-wheat-detection/discussion/159585
\n\nI will try to be more present.
", "rawMarkdown": "Dear @vadbeg \n\nSorry for the inconvenience. Some answers to the bug here: https://www.kaggle.com/c/global-wheat-detection/discussion/159585\n\nI will try to be more present. ", "votes": 4 }, { "id": 910602, "postDate": "2020-07-01T08:34:13.517Z", "content": "this is so useful. Thank you for sharing. It help me a lot :D
", "rawMarkdown": "this is so useful. Thank you for sharing. It help me a lot :D", "votes": 1 }, { "id": 891681, "postDate": "2020-06-18T11:34:53.590Z", "content": "Wired，after I remove these boxes,I got a lower lb score
", "rawMarkdown": "Wired，after I remove these boxes,I got a lower lb score", "replies": [ { "id": 891746, "postDate": "2020-06-18T12:36:07.527Z", "rawMarkdown": "", "votes": -2, "isDeleted": true }, { "id": 892348, "postDate": "2020-06-18T20:17:09.113Z", "content": "Did you remove just the boxes or the entire images from the training set?\nIt could also just be randomness depending on how good your CV is
", "rawMarkdown": "Did you remove just the boxes or the entire images from the training set?\nIt could also just be randomness depending on how good your CV is", "votes": 1 }, { "id": 893196, "postDate": "2020-06-19T12:57:08.783Z", "content": "I just remove the wrong boxes,after I remove these boxes,CV is improved,but lb get lower
", "rawMarkdown": "I just remove the wrong boxes,after I remove these boxes,CV is improved,but lb get lower" }, { "id": 893208, "postDate": "2020-06-19T13:07:34.633Z", "content": "Maybe LB dataset partly consists of such corrupted bboxes. Or it is just near-random score changes.
", "rawMarkdown": "Maybe LB dataset partly consists of such corrupted bboxes. Or it is just near-random score changes. ", "votes": 2 }, { "id": 911758, "postDate": "2020-07-02T02:17:29.573Z", "content": "If we delete the bad big boxes,will lead to some wheat have no box.Maybe we need to remark the big wrong box or delete the images contains wrong boxes. What do you think?
", "rawMarkdown": "If we delete the bad big boxes,will lead to some wheat have no box.Maybe we need to remark the big wrong box or delete the images contains wrong boxes. What do you think?" }, { "id": 911810, "postDate": "2020-07-02T03:20:20.207Z", "content": "That is correct. I have manually added boxes in this case.
", "rawMarkdown": "That is correct. I have manually added boxes in this case." }, { "id": 911937, "postDate": "2020-07-02T05:43:10.377Z", "content": "Not all bbox you need to redraw. There are genuine noise boxes in the training set.\n![\"\"](\"https://www.kaggleusercontent.com/kf/37408450/eyJhbGciOiJkaXIiLCJlbmMiOiJBMTI4Q0JDLUhTMjU2In0..SJtds68J362bnkrESasEVg.Xv7WLyE99ZQMcp4RHgq57Mu8oLDK4ogGlIhhsq8K1w5YEYU5oNzKsIe7iEYJDVjOARYbE_Pfa7P-eIH36FteWw9I3LEuPzE3_l6VfJ2PG60ET03c3Dircu2MwZR30ibD68ILWdyQ8cHeA4MYZkRUVoQiNK-hka3KDv33e9Dt_mgZv1-3a1DCL4nUIqqxflCu08QIvlt6MNTEIvVN_yfibApeTFT4OVSydG5xywXwOJSlXWovXT1BvgwJV1gWtA86eCayis1eGjZVjE4vEqS4mcZRtdeTSTCGiwg_E9UKLve3cE-3y6hDRIwDFXwrCqsbGA7p5GTMpm6dl5zjTazWcmUqyxXBgrjQsasgLas4T4Cy_-GBCLa1gryrlCZa20PjU68fnmU2tEyUrx-y54AZcpa00uw5KvF6s_ACX9amYlrqzg_hmdFpmjAPRZwp7_iJP19PMCpyAyliQN1rf_vvGmPrAH_bWU63THT4RwOKsFCIv1wbkDSCiakzQEldAsyjgDpoI3cUzHkPiUWpqodong_eLoudKx--CuOETODRsJXxwdWSvSrvRafMjXNPlBBrSPw2MHkO-Ir_FbWv8BuBoB8moeGInaLiqivNXrtPi_vIna1O_zwnVJpjMHg6uOFR_o3UEjdmQfJhQWle-ulvGLr3WfqFXl1RhL2ilWr-yxU.a8vv4Z-1mSP_82YP3oB9tQ/__results___files/__results___17_0.png\")
I bet gt in private test set is messed up. My cv score and LB is like far apart.
", "rawMarkdown": "Not all bbox you need to redraw. There are genuine noise boxes in the training set.\n\n\nI bet gt in private test set is messed up. My cv score and LB is like far apart.", "votes": 1 }, { "id": 912465, "postDate": "2020-07-02T13:47:24.517Z", "content": "What I meant is with the specific bug mentioned above, where two boxes are given the same ID and are merged, you must delete the large box and place two smaller boxes to correct it.
\n\nAnd yes, the ground truth of the private test set most certainly contains incorrect boxes. In the recent update the organizers say they will fix this in the final private test set, though I doubt the labels will still be perfect.
", "rawMarkdown": "What I meant is with the specific bug mentioned above, where two boxes are given the same ID and are merged, you must delete the large box and place two smaller boxes to correct it.\n\nAnd yes, the ground truth of the private test set most certainly contains incorrect boxes. In the recent update the organizers say they will fix this in the final private test set, though I doubt the labels will still be perfect." }, { "id": 915899, "postDate": "2020-07-05T07:26:50.687Z", "content": "@ahmmadrashid hello, may I ask question? What is CV score?
", "rawMarkdown": "@ahmmadrashid hello, may I ask question? What is CV score?", "votes": 1 }, { "id": 934058, "postDate": "2020-07-18T07:56:24.207Z", "content": "CV score is your local cross-validation score
", "rawMarkdown": "CV score is your local cross-validation score" }, { "id": 941643, "postDate": "2020-07-23T10:35:39.637Z", "content": "@matthewmasters what does it means by manually adding of boxes ?
", "rawMarkdown": "@matthewmasters what does it means by manually adding of boxes ?\n" }, { "id": 942835, "postDate": "2020-07-24T03:11:44.367Z", "content": "It means that I reviewed images by eye and added boxes when none was present. I had to convert the csv into xml and used this tool to do the annotations. I plan on releasing my dataset after the competition deadline.
", "rawMarkdown": "It means that I reviewed images by eye and added boxes when none was present. I had to convert the csv into xml and used [this tool](https://github.com/tzutalin/labelImg) to do the annotations. I plan on releasing my dataset after the competition deadline." }, { "id": 942939, "postDate": "2020-07-24T04:28:59.143Z", "content": "does it gives in xywh format ?
", "rawMarkdown": "does it gives in xywh format ?" } ] }, { "id": 914610, "postDate": "2020-07-04T04:56:35.837Z", "content": "Thanks for sharing this.
", "rawMarkdown": "Thanks for sharing this.", "votes": 1 } ], "comments": [ { "id": 891179, "author_name": "GreatGameDota", "author_url": "", "post_date": "2020-06-18T01:44:05.070000", "content": "Quick comb through and you missed a few images
\n\n9a30dd802.jpg
\n\n
d7a02151d.jpg
\n\n
dcafcae79.jpg
\n\n
Also here is a list of the indices of the boxes in train.csv so you don't have to remove the entire image from training
\n\n\n3687\n117344\n173\n113947\n52868\n2159\n2169\n121633\n121634\n147504\n118211\n52727\n147552\n\n
@greatgamedota Cool!
", "votes": 0, "replies": [] }, { "id": 901315, "author_name": "rosuluc", "author_url": "", "post_date": "2020-06-25T11:39:50.867000", "content": "bad_boxes = [3687, 117344,173,113947,52868,2159,2169,121633,121634,147504,118211,52727,147552]
", "votes": 3, "replies": [] }, { "id": 942803, "author_name": "عثمان", "author_url": "", "post_date": "2020-07-24T02:26:58.713000", "content": "Might I inquire as to how you're styling that notebook—I'm assuming some extension?
", "votes": 0, "replies": [] }, { "id": 942836, "author_name": "Matt", "author_url": "", "post_date": "2020-07-24T03:12:57.260000", "content": "I would assume they are using an extension like this.
", "votes": 0, "replies": [] }, { "id": 943611, "author_name": "GreatGameDota", "author_url": "", "post_date": "2020-07-24T13:21:26.563000", "content": "That screenshot is actually Kaggle's default Dark Mode for kernals, but I am using that extension currently.
", "votes": 1, "replies": [] } ] }, { "id": 897993, "author_name": "Chanhu", "author_url": "", "post_date": "2020-06-23T08:03:11.943000", "content": "\n
Dear @vadbeg
\n\nSorry for the inconvenience. Some answers to the bug here: https://www.kaggle.com/c/global-wheat-detection/discussion/159585
\n\nI will try to be more present.
", "votes": 4, "replies": [] }, { "id": 910602, "author_name": "Ahmmad Rashid", "author_url": "", "post_date": "2020-07-01T08:34:13.517000", "content": "this is so useful. Thank you for sharing. It help me a lot :D
", "votes": 1, "replies": [] }, { "id": 891681, "author_name": "thefatcat", "author_url": "", "post_date": "2020-06-18T11:34:53.590000", "content": "Wired，after I remove these boxes,I got a lower lb score
", "votes": 0, "replies": [ { "id": 891746, "author_name": "", "author_url": "", "post_date": "2020-06-18T12:36:07.527000", "content": "", "votes": -2, "replies": [] }, { "id": 892348, "author_name": "Matt", "author_url": "", "post_date": "2020-06-18T20:17:09.113000", "content": "Did you remove just the boxes or the entire images from the training set?\nIt could also just be randomness depending on how good your CV is
", "votes": 1, "replies": [] }, { "id": 893196, "author_name": "thefatcat", "author_url": "", "post_date": "2020-06-19T12:57:08.783000", "content": "I just remove the wrong boxes,after I remove these boxes,CV is improved,but lb get lower
", "votes": 0, "replies": [] }, { "id": 893208, "author_name": "Vadzim Tsitko", "author_url": "", "post_date": "2020-06-19T13:07:34.633000", "content": "Maybe LB dataset partly consists of such corrupted bboxes. Or it is just near-random score changes.
", "votes": 2, "replies": [] }, { "id": 911758, "author_name": "xiaopeng", "author_url": "", "post_date": "2020-07-02T02:17:29.573000", "content": "If we delete the bad big boxes,will lead to some wheat have no box.Maybe we need to remark the big wrong box or delete the images contains wrong boxes. What do you think?
", "votes": 0, "replies": [] }, { "id": 911810, "author_name": "Matt", "author_url": "", "post_date": "2020-07-02T03:20:20.207000", "content": "That is correct. I have manually added boxes in this case.
", "votes": 0, "replies": [] }, { "id": 911937, "author_name": "Ahmmad Rashid", "author_url": "", "post_date": "2020-07-02T05:43:10.377000", "content": "Not all bbox you need to redraw. There are genuine noise boxes in the training set.\n
I bet gt in private test set is messed up. My cv score and LB is like far apart.
", "votes": 1, "replies": [] }, { "id": 912465, "author_name": "Matt", "author_url": "", "post_date": "2020-07-02T13:47:24.517000", "content": "What I meant is with the specific bug mentioned above, where two boxes are given the same ID and are merged, you must delete the large box and place two smaller boxes to correct it.
\n\nAnd yes, the ground truth of the private test set most certainly contains incorrect boxes. In the recent update the organizers say they will fix this in the final private test set, though I doubt the labels will still be perfect.
", "votes": 0, "replies": [] }, { "id": 915899, "author_name": "yangmanyao", "author_url": "", "post_date": "2020-07-05T07:26:50.687000", "content": "@ahmmadrashid hello, may I ask question? What is CV score?
", "votes": 1, "replies": [] }, { "id": 934058, "author_name": "Haley", "author_url": "", "post_date": "2020-07-18T07:56:24.207000", "content": "CV score is your local cross-validation score
", "votes": 0, "replies": [] }, { "id": 941643, "author_name": "Jaideep", "author_url": "", "post_date": "2020-07-23T10:35:39.637000", "content": "@matthewmasters what does it means by manually adding of boxes ?
", "votes": 0, "replies": [] }, { "id": 942835, "author_name": "Matt", "author_url": "", "post_date": "2020-07-24T03:11:44.367000", "content": "It means that I reviewed images by eye and added boxes when none was present. I had to convert the csv into xml and used this tool to do the annotations. I plan on releasing my dataset after the competition deadline.
", "votes": 0, "replies": [] }, { "id": 942939, "author_name": "Jaideep", "author_url": "", "post_date": "2020-07-24T04:28:59.143000", "content": "does it gives in xywh format ?
", "votes": 0, "replies": [] } ] }, { "id": 914610, "author_name": "Kang", "author_url": "", "post_date": "2020-07-04T04:56:35.837000", "content": "Thanks for sharing this.
", "votes": 1, "replies": [] } ], "raw_markdown_by_id": { "891066": "So, I've looked at images with bboxes. And created a list of huge bboxes without wheat in them. There it is:\n``` \n41c0123cc.jpg\na1321ca95.jpg\n2cc75e9f5.jpg\n42e6efaaa.jpg\n409a8490c.jpg\nd067ac2b1.jpg\nd60e832a5.jpg\n893938464.jpg\n```\n\nIf somebody will find more of them, please, add in comments. Thanks to @artgor for mentioning this issue.\nImages:\n\n\n\n\n\n\n\n\n\n", "891179": "Quick comb through and you missed a few images\n\n9a30dd802.jpg\n\n\n\nd7a02151d.jpg\n\n\n\ndcafcae79.jpg\n\n\n\n\nAlso here is a list of the indices of the boxes in train.csv so you don't have to remove the entire image from training\n\n```\n3687\n117344\n173\n113947\n52868\n2159\n2169\n121633\n121634\n147504\n118211\n52727\n147552\n```\n\n", "897993": "\n\n", "896949": "Dear @vadbeg \n\nSorry for the inconvenience. Some answers to the bug here: https://www.kaggle.com/c/global-wheat-detection/discussion/159585\n\nI will try to be more present. ", "910602": "this is so useful. Thank you for sharing. It help me a lot :D", "891681": "Wired，after I remove these boxes,I got a lower lb score", "914610": "Thanks for sharing this." } }