![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F159865%2Fc9190636c532a69b22b9409619b169a7%2F1e40a05.jpg?generation=1569820375741148&alt=media\")
I went through all the images manually and compiled a list of potentially problematic images. Some images are comprised of multiple images which may not be a problem. Others are clearly mislabeled as portions of the black stripes have labels.
\n\nHere are a few examples. From the training set:
\n\n1e40a05.jpg is comprised of multiple images
\n\n
8bd81ce.jpg is comprised of multiple images, one is washed out
\n\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F159865%2F13f5bb8a157c8530f04910068d5ee95d%2F8bd81ce.jpg?generation=1569820433046924&alt=media\")
563fc48.jpg has labels that extend into the black stripe in the lower left corner
\n\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F159865%2F122cf7763bdf6ff2413836572b2c8d27%2Finvalid.png?generation=1569820529205239&alt=media\")
Here are my notes for the training set:
\n\nHere are my notes for the test set:
\n\nI went through all the images manually and compiled a list of potentially problematic images. Some images are comprised of multiple images which may not be a problem. Others are clearly mislabeled as portions of the black stripes have labels.
\n\nHere are a few examples. From the training set:
\n\n1e40a05.jpg is comprised of multiple images
\n\n
8bd81ce.jpg is comprised of multiple images, one is washed out
\n\n
563fc48.jpg has labels that extend into the black stripe in the lower left corner
\n\n
Here are my notes for the training set:
\n\nHere are my notes for the test set:
\n\nLooks like you found a couple more than me at https://www.kaggle.com/c/understanding_cloud_organization/discussion/105359#latest-629844 !
", "rawMarkdown": "Looks like you found a couple more than me at https://www.kaggle.com/c/understanding_cloud_organization/discussion/105359#latest-629844 !", "votes": null }, { "id": "637046", "postDate": "09/30/2019 15:21:13", "content": "Sorry, I missed your original thread. Feel free to add these to your original thread if you'd like.
", "rawMarkdown": "Sorry, I missed your original thread. Feel free to add these to your original thread if you'd like.", "votes": null }, { "id": "637617", "postDate": "10/01/2019 06:59:50", "content": "there could be much more than that,i am sure those are the pictures that hurting our models and preventing from learning useful features,i said it on the other day but no one seems to believe,thanks for this post @joshvarty
", "rawMarkdown": "there could be much more than that,i am sure those are the pictures that hurting our models and preventing from learning useful features,i said it on the other day but no one seems to believe,thanks for this post @joshvarty", "votes": null }, { "id": "639010", "postDate": "10/02/2019 17:07:42", "content": "Good work, I doubt that those few can have any impact on the model though.
", "rawMarkdown": "Good work, I doubt that those few can have any impact on the model though.", "votes": null }, { "id": "668275", "postDate": "11/08/2019 08:03:37", "content": "How many score increase on LB for del these bad images ?
", "rawMarkdown": "How many score increase on LB for del these bad images ?", "votes": null }, { "id": "668281", "postDate": "11/08/2019 08:10:16", "content": "0.0000 :)
", "rawMarkdown": "0.0000 :)", "votes": null } ], "comments": [ { "id": 636762, "author_name": "robga", "author_url": "", "post_date": "09/30/2019 07:00:37", "content": "Looks like you found a couple more than me at https://www.kaggle.com/c/understanding_cloud_organization/discussion/105359#latest-629844 !
", "votes": null, "replies": [ { "id": 637046, "author_name": "joshvarty", "author_url": "", "post_date": "09/30/2019 15:21:13", "content": "Sorry, I missed your original thread. Feel free to add these to your original thread if you'd like.
", "votes": null, "replies": [] } ] }, { "id": 637617, "author_name": "mobassir", "author_url": "", "post_date": "10/01/2019 06:59:50", "content": "there could be much more than that,i am sure those are the pictures that hurting our models and preventing from learning useful features,i said it on the other day but no one seems to believe,thanks for this post @joshvarty
", "votes": null, "replies": [] }, { "id": 639010, "author_name": "philippsinger", "author_url": "", "post_date": "10/02/2019 17:07:42", "content": "Good work, I doubt that those few can have any impact on the model though.
", "votes": null, "replies": [] }, { "id": 668275, "author_name": "", "author_url": "", "post_date": "11/08/2019 08:03:37", "content": "How many score increase on LB for del these bad images ?
", "votes": null, "replies": [ { "id": 668281, "author_name": "robga", "author_url": "", "post_date": "11/08/2019 08:10:16", "content": "0.0000 :)
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "636721": "I went through all the images manually and compiled a list of potentially problematic images. Some images are comprised of multiple images which may not be a problem. Others are clearly mislabeled as portions of the black stripes have labels.\n\nHere are a few examples. From the training set:\n\n1e40a05.jpg is comprised of multiple images\n\n\n\n8bd81ce.jpg is comprised of multiple images, one is washed out\n\n\n\n563fc48.jpg has labels that extend into the black stripe in the lower left corner\n\n\n\nHere are my notes for the training set:\n\n - 5e70931.jpg Has smaller black streaks in it\n - 7ca1d0b.jpg Has a small black streak in it\n - 1e40a05.jpg is comprised of 3 images\n - 8bd81ce.jpg is comprised of 2 images and one is washed out\n - 41f92e5.jpg is comprised of 2 images and one is washed out\n - 449b792.jpg is comprised of 3 images and one is washed out (may have bad labels)\n - 563fc48.jpg is comprised of 3 images with a bad label\n - 1588d4c.jpg is comprised of 3 images \n - 046586a.jpg is comprised of multiple images\n - b092cc1.jpg is partially corrupted\n - c26c635.jpg is partially corrupted\n - c0306e5.jpg is partially corrupted with a bad label\n - e04fea3.jpg is partially corrupted with a bad label\n - e5f2f24.jpg is partially corrupted \n - ee0ba55.jpg seems weirdly labelled\n - fa645da is partially corrupted with a bad label\n\nHere are my notes for the test set:\n\n- 3cb4ce5.jpg is corrupt\n- 18c7c27.jpg is multiple images\n- 25c5403.jpg is multiple images\n- 43d43a8.jpg is multiple images\n- 4375f93.jpg is multiple images\n- 638185f.jpg is multiple images\n- a4da892.jpg has a weird line\n- d1378d2.jpg has a small distortion at the bottom", "636762": "Looks like you found a couple more than me at https://www.kaggle.com/c/understanding_cloud_organization/discussion/105359#latest-629844 !", "637046": "Sorry, I missed your original thread. Feel free to add these to your original thread if you'd like.", "637617": "there could be much more than that,i am sure those are the pictures that hurting our models and preventing from learning useful features,i said it on the other day but no one seems to believe,thanks for this post @joshvarty", "639010": "Good work, I doubt that those few can have any impact on the model though.", "668275": "How many score increase on LB for del these bad images ?", "668281": "0.0000 :)" }, "source": "meta" }