Welcome to iMaterialist (Fashion) 2019 -FGVC6! We present a new clothing dataset with the goal of introducing a novel fine-grained segmentation task by joining forces between the fashion and computer vision communities.
\n\nIn this competition, we challenge you to develop algorithms that will help with an important step towards automatic product detection – to accurately assign segmentations and attribute labels for fashion images.
\n\nGLHF!
", "messages": [ { "id": "523278", "postDate": "04/25/2019 22:18:43", "content": "Welcome to iMaterialist (Fashion) 2019 -FGVC6! We present a new clothing dataset with the goal of introducing a novel fine-grained segmentation task by joining forces between the fashion and computer vision communities.
\n\nIn this competition, we challenge you to develop algorithms that will help with an important step towards automatic product detection – to accurately assign segmentations and attribute labels for fashion images.
\n\nGLHF!
", "rawMarkdown": "**Welcome to iMaterialist (Fashion) 2019 -FGVC6!** We present a new clothing dataset with the goal of introducing a novel **fine-grained segmentation** task by joining forces between the fashion and computer vision communities. \n\nIn this competition, we challenge you to develop algorithms that will help with an important step towards automatic product detection – to **accurately assign segmentations and attribute labels** for fashion images. \n\n\nGLHF!", "votes": null }, { "id": "523298", "postDate": "04/26/2019 00:56:10", "content": "And that is awesome! Thank you for this challenge, and for the medals :))
", "rawMarkdown": "And that is awesome! Thank you for this challenge, and for the medals :))", "votes": null }, { "id": "526239", "postDate": "05/02/2019 16:57:30", "content": "@makeitworkjml small question regarding evaluation metric. Evaluation page states:
\n\n\n\n\nThe average precision of a single ClassId and a single image is then calculated as the mean of the above precision values at each IoU threshold:\n The score returned by the competition metric is the mean taken over the individual average precisions of each image and each ClassId in the test dataset.
\n
But if some class is not present in an image (both in ground truth and in predictions), then average precision of this ClassId on this image is not defined (we have 0 divided by 0). Are such cases excluded when calculating the average overall all classes and all images?
", "rawMarkdown": "makeitworkjml small question regarding evaluation metric. Evaluation page states:\n\n> The average precision of a single ClassId and a single image is then calculated as the mean of the above precision values at each IoU threshold:\n> The score returned by the competition metric is the mean taken over the individual average precisions of each image and each ClassId in the test dataset.\n\nBut if some class is not present in an image (both in ground truth and in predictions), then average precision of this ClassId on this image is not defined (we have 0 divided by 0). Are such cases excluded when calculating the average overall all classes and all images?", "votes": null }, { "id": "526845", "postDate": "05/03/2019 23:14:13", "content": "Hi Konstantin, if a ClassId not present in the groundtruth, then this mask will not be counted as any true positive case. So you are right, such cases are excluded.
", "rawMarkdown": "Hi Konstantin, if a ClassId not present in the groundtruth, then this mask will not be counted as any true positive case. So you are right, such cases are excluded.", "votes": null }, { "id": "526943", "postDate": "05/04/2019 07:03:08", "content": "Great, thanks for clarification @makeitworkjml !
", "rawMarkdown": "Great, thanks for clarification @makeitworkjml !", "votes": null }, { "id": "546362", "postDate": "06/06/2019 14:10:18", "content": "@makeitworkjml attributes clarification is needed.\nThere are also attributes in classes 27, 28, 33. Are they labelled in the test for these classes, or it's just mislabeled?
", "rawMarkdown": "makeitworkjml attributes clarification is needed.\nThere are also attributes in classes 27, 28, 33. Are they labelled in the test for these classes, or it's just mislabeled?", "votes": null }, { "id": "546442", "postDate": "06/06/2019 15:35:43", "content": "Hi @blondinka , thank you for pointing that out. This is an annotation error (only a few images has this issue, and test images are not affected). We are planning to release the dataset in JSON format after the competition. The error will be removed by then. Let me know if you have any other questions!
", "rawMarkdown": "Hi @blondinka , thank you for pointing that out. This is an annotation error (only a few images has this issue, and test images are not affected). We are planning to release the dataset in JSON format after the competition. The error will be removed by then. Let me know if you have any other questions!", "votes": null }, { "id": "546562", "postDate": "06/06/2019 17:26:39", "content": "Hi @makeitworkjml actually, I have. The numer of attributes varies and some attributes that could be for an object are missing sometimes... If we get one extra or one missing it test -- it does not count the whole class, does not it? Is the amount of attributes varied in the test as well for the same class?
", "rawMarkdown": "Hi @makeitworkjml actually, I have. The numer of attributes varies and some attributes that could be for an object are missing sometimes... If we get one extra or one missing it test -- it does not count the whole class, does not it? Is the amount of attributes varied in the test as well for the same class?", "votes": null }, { "id": "546967", "postDate": "06/07/2019 04:24:37", "content": "Hi @blondinka , please refer to this post for the reason why some attributes are missing:\nhttps://www.kaggle.com/c/imaterialist-fashion-2019-FGVC6/discussion/94811
\n\nAnd yes, the number of attributes varies for each mask, in both training and test data.
", "rawMarkdown": "Hi @blondinka , please refer to this post for the reason why some attributes are missing:\nhttps://www.kaggle.com/c/imaterialist-fashion-2019-FGVC6/discussion/94811\n\nAnd yes, the number of attributes varies for each mask, in both training and test data.", "votes": null } ], "comments": [ { "id": 523298, "author_name": "blondinka", "author_url": "", "post_date": "04/26/2019 00:56:10", "content": "And that is awesome! Thank you for this challenge, and for the medals :))
", "votes": null, "replies": [] }, { "id": 526239, "author_name": "lopuhin", "author_url": "", "post_date": "05/02/2019 16:57:30", "content": "@makeitworkjml small question regarding evaluation metric. Evaluation page states:
\n\n\n\n\nThe average precision of a single ClassId and a single image is then calculated as the mean of the above precision values at each IoU threshold:\n The score returned by the competition metric is the mean taken over the individual average precisions of each image and each ClassId in the test dataset.
\n
But if some class is not present in an image (both in ground truth and in predictions), then average precision of this ClassId on this image is not defined (we have 0 divided by 0). Are such cases excluded when calculating the average overall all classes and all images?
", "votes": null, "replies": [ { "id": 526845, "author_name": "makeitworkjml", "author_url": "", "post_date": "05/03/2019 23:14:13", "content": "Hi Konstantin, if a ClassId not present in the groundtruth, then this mask will not be counted as any true positive case. So you are right, such cases are excluded.
", "votes": null, "replies": [] }, { "id": 526943, "author_name": "lopuhin", "author_url": "", "post_date": "05/04/2019 07:03:08", "content": "Great, thanks for clarification @makeitworkjml !
", "votes": null, "replies": [] } ] }, { "id": 546362, "author_name": "blondinka", "author_url": "", "post_date": "06/06/2019 14:10:18", "content": "@makeitworkjml attributes clarification is needed.\nThere are also attributes in classes 27, 28, 33. Are they labelled in the test for these classes, or it's just mislabeled?
", "votes": null, "replies": [ { "id": 546442, "author_name": "makeitworkjml", "author_url": "", "post_date": "06/06/2019 15:35:43", "content": "Hi @blondinka , thank you for pointing that out. This is an annotation error (only a few images has this issue, and test images are not affected). We are planning to release the dataset in JSON format after the competition. The error will be removed by then. Let me know if you have any other questions!
", "votes": null, "replies": [] }, { "id": 546562, "author_name": "blondinka", "author_url": "", "post_date": "06/06/2019 17:26:39", "content": "Hi @makeitworkjml actually, I have. The numer of attributes varies and some attributes that could be for an object are missing sometimes... If we get one extra or one missing it test -- it does not count the whole class, does not it? Is the amount of attributes varied in the test as well for the same class?
", "votes": null, "replies": [] }, { "id": 546967, "author_name": "makeitworkjml", "author_url": "", "post_date": "06/07/2019 04:24:37", "content": "Hi @blondinka , please refer to this post for the reason why some attributes are missing:\nhttps://www.kaggle.com/c/imaterialist-fashion-2019-FGVC6/discussion/94811
\n\nAnd yes, the number of attributes varies for each mask, in both training and test data.
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "523278": "**Welcome to iMaterialist (Fashion) 2019 -FGVC6!** We present a new clothing dataset with the goal of introducing a novel **fine-grained segmentation** task by joining forces between the fashion and computer vision communities. \n\nIn this competition, we challenge you to develop algorithms that will help with an important step towards automatic product detection – to **accurately assign segmentations and attribute labels** for fashion images. \n\n\nGLHF!", "523298": "And that is awesome! Thank you for this challenge, and for the medals :))", "526239": "makeitworkjml small question regarding evaluation metric. Evaluation page states:\n\n> The average precision of a single ClassId and a single image is then calculated as the mean of the above precision values at each IoU threshold:\n> The score returned by the competition metric is the mean taken over the individual average precisions of each image and each ClassId in the test dataset.\n\nBut if some class is not present in an image (both in ground truth and in predictions), then average precision of this ClassId on this image is not defined (we have 0 divided by 0). Are such cases excluded when calculating the average overall all classes and all images?", "526845": "Hi Konstantin, if a ClassId not present in the groundtruth, then this mask will not be counted as any true positive case. So you are right, such cases are excluded.", "526943": "Great, thanks for clarification @makeitworkjml !", "546362": "makeitworkjml attributes clarification is needed.\nThere are also attributes in classes 27, 28, 33. Are they labelled in the test for these classes, or it's just mislabeled?", "546442": "Hi @blondinka , thank you for pointing that out. This is an annotation error (only a few images has this issue, and test images are not affected). We are planning to release the dataset in JSON format after the competition. The error will be removed by then. Let me know if you have any other questions!", "546562": "Hi @makeitworkjml actually, I have. The numer of attributes varies and some attributes that could be for an object are missing sometimes... If we get one extra or one missing it test -- it does not count the whole class, does not it? Is the amount of attributes varied in the test as well for the same class?", "546967": "Hi @blondinka , please refer to this post for the reason why some attributes are missing:\nhttps://www.kaggle.com/c/imaterialist-fashion-2019-FGVC6/discussion/94811\n\nAnd yes, the number of attributes varies for each mask, in both training and test data." }, "source": "meta" }