![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2164736%2F0a7cdcb22e1b586730984c7cead2dd4c%2Fe8d19d9f029904e2561b3b2a220a301.png?generation=1576342183730083&alt=media\")
\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2164736%2Fd34adfe483aa0258a3c5f6f4d7e3c97f%2Fd98153e74ea343832ba652295782abf.png?generation=1576342183674857&alt=media\")
Overview:\n\n
And you can find that some of the images are manually flipped horizontally.\nBecause in China, the opposite lanes are always on your left, obviously...\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2164736%2Fd251e07d87036e3d9c2ddf5706ef2419%2FID_2fee670a1.jpg?generation=1576342348946754&alt=media\")
At last,there's an interesting picture......๐ Is this picture taken by mobile phone....\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2164736%2F10aedd56c80b21cb3511996441f15166%2FID_0a2c7cb83.jpg?generation=1576342541620503&alt=media\")
Overview:\n\n
And you can find that some of the images are manually flipped horizontally.\nBecause in China, the opposite lanes are always on your left, obviously...\n
At last,there's an interesting picture......๐ Is this picture taken by mobile phone....\n
The distribution of training sets and test sets is very different (it seems that these differences were made intentionally)\nSome augmentations maybe important for model generalization
", "rawMarkdown": "The distribution of training sets and test sets is very different (it seems that these differences were made intentionally)\nSome augmentations maybe important for model generalization", "votes": null }, { "id": "695400", "postDate": "12/15/2019 04:26:05", "content": "some images have intentional gaussian noise too..\ni guess the host wants to see how we handle the image augumentations. (photography noise etc..)
", "rawMarkdown": "some images have intentional gaussian noise too..\ni guess the host wants to see how we handle the image augumentations. (photography noise etc..)", "votes": null }, { "id": "695483", "postDate": "12/15/2019 08:11:17", "content": "intersting
", "rawMarkdown": "intersting", "votes": null }, { "id": "695758", "postDate": "12/15/2019 14:22:25", "content": "good find๐
", "rawMarkdown": "good find๐", "votes": null }, { "id": "696064", "postDate": "12/16/2019 04:36:04", "content": "Noticed that too... I added some augmentations brightness, blur, flip... local cv higher but lb unchanged... Do you have any success with augmentations?
", "rawMarkdown": "Noticed that too... I added some augmentations brightness, blur, flip... local cv higher but lb unchanged... Do you have any success with augmentations?", "votes": null }, { "id": "700042", "postDate": "12/21/2019 10:46:45", "content": "Interesting ..
", "rawMarkdown": "Interesting ..", "votes": null }, { "id": "700642", "postDate": "12/22/2019 10:57:56", "content": "Also add guassian , multitrandomfloat, random brightness, motion blur augment, but not change lb
", "rawMarkdown": "Also add guassian , multitrandomfloat, random brightness, motion blur augment, but not change lb", "votes": null }, { "id": "700644", "postDate": "12/22/2019 10:58:39", "content": "Have you tried filter the heat map confidence with threshold?
", "rawMarkdown": "Have you tried filter the heat map confidence with threshold?", "votes": null } ], "comments": [ { "id": 695151, "author_name": "changewow", "author_url": "", "post_date": "12/14/2019 17:02:10", "content": "The distribution of training sets and test sets is very different (it seems that these differences were made intentionally)\nSome augmentations maybe important for model generalization
", "votes": null, "replies": [] }, { "id": 695400, "author_name": "kyoshioka47", "author_url": "", "post_date": "12/15/2019 04:26:05", "content": "some images have intentional gaussian noise too..\ni guess the host wants to see how we handle the image augumentations. (photography noise etc..)
", "votes": null, "replies": [] }, { "id": 695483, "author_name": "haoqiu111", "author_url": "", "post_date": "12/15/2019 08:11:17", "content": "intersting
", "votes": null, "replies": [] }, { "id": 695758, "author_name": "diegojohnson", "author_url": "", "post_date": "12/15/2019 14:22:25", "content": "good find๐
", "votes": null, "replies": [] }, { "id": 696064, "author_name": "niuddd", "author_url": "", "post_date": "12/16/2019 04:36:04", "content": "Noticed that too... I added some augmentations brightness, blur, flip... local cv higher but lb unchanged... Do you have any success with augmentations?
", "votes": null, "replies": [ { "id": 700644, "author_name": "cswwp347724", "author_url": "", "post_date": "12/22/2019 10:58:39", "content": "Have you tried filter the heat map confidence with threshold?
", "votes": null, "replies": [] } ] }, { "id": 700042, "author_name": "swathym", "author_url": "", "post_date": "12/21/2019 10:46:45", "content": "Interesting ..
", "votes": null, "replies": [] }, { "id": 700642, "author_name": "cswwp347724", "author_url": "", "post_date": "12/22/2019 10:57:56", "content": "Also add guassian , multitrandomfloat, random brightness, motion blur augment, but not change lb
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "695149": "**Overview:**\n\n\n\n**And you can find that some of the images are manually flipped horizontally.\nBecause in China, the opposite lanes are always on your left, obviously...**\n\n\n**At last,there's an interesting picture......๐ Is this picture taken by mobile phone....**\n", "695151": "The distribution of training sets and test sets is very different (it seems that these differences were made intentionally)\nSome augmentations maybe important for model generalization", "695400": "some images have intentional gaussian noise too..\ni guess the host wants to see how we handle the image augumentations. (photography noise etc..)", "695483": "intersting", "695758": "good find๐", "696064": "Noticed that too... I added some augmentations brightness, blur, flip... local cv higher but lb unchanged... Do you have any success with augmentations?", "700042": "Interesting ..", "700642": "Also add guassian , multitrandomfloat, random brightness, motion blur augment, but not change lb", "700644": "Have you tried filter the heat map confidence with threshold?" }, "source": "meta" }