I have a brief look at the dataset.\nGiven the diverse photo background from different designers, I think if we train a model with high accuracy, maybe the neural networks are classifying the overall photo statistics (e.g. background/light condition/...), rather than the clothes designs themselves.\nThis may make the trained model meaningless.
", "messages": [ { "id": 529113, "postDate": "2019-05-09T08:12:45.273Z", "content": "\"I think if we train a model with high accuracy, maybe the neural networks are classifying the overall photo statistics\"
\n\ni also note that the same model will wear the clothes by the same designer in the same fashion show. We know it is the same model because she can wear the same shoes
", "rawMarkdown": "\"I think if we train a model with high accuracy, maybe the neural networks are classifying the overall photo statistics\"\n\ni also note that the same model will wear the clothes by the same designer in the same fashion show. We know it is the same model because she can wear the same shoes", "replies": [ { "id": 530065, "postDate": "2019-05-11T16:02:52.310Z", "content": "Thank you, Heng, a nice idea. Although it is possible to classify style based on the model, I think this is not the original requirement of this competition?
", "rawMarkdown": "Thank you, Heng, a nice idea. Although it is possible to classify style based on the model, I think this is not the original requirement of this competition?" } ] }, { "id": 526020, "postDate": "2019-05-02T07:43:20.333Z", "content": "Is there a way to get version 1 of the dataset now?
", "rawMarkdown": "Is there a way to get version 1 of the dataset now?" }, { "id": 523613, "postDate": "2019-04-26T16:11:05.740Z", "content": "What image shape are u feeding in ur model??
", "rawMarkdown": "What image shape are u feeding in ur model??" }, { "id": 514981, "postDate": "2019-04-12T04:58:17.880Z", "content": "I saw the new test set, and found a few pictures are ruined.\nFor example, 98.png, 173.png, 233.png, 252.png, ...\nIs this intended, or it's a mistake?\nThanks
", "rawMarkdown": "I saw the new test set, and found a few pictures are ruined.\nFor example, 98.png, 173.png, 233.png, 252.png, ...\nIs this intended, or it's a mistake?\nThanks", "replies": [ { "id": 516277, "postDate": "2019-04-13T23:18:52.143Z", "content": "We applied post-processing to a the test set to attempt to remove the background from around the fashion models / apparel, in order to better reward approaches that are identifying features in the fashion design itself (rather than, for example) the color of the background or pattern of the floor. In some cases that algorithm removed some portions of the apparel as well. We’ve corrected most of these, but some remain. We do not presently intend to further update the test set.
", "rawMarkdown": "We applied post-processing to a the test set to attempt to remove the background from around the fashion models / apparel, in order to better reward approaches that are identifying features in the fashion design itself (rather than, for example) the color of the background or pattern of the floor. In some cases that algorithm removed some portions of the apparel as well. We’ve corrected most of these, but some remain. We do not presently intend to further update the test set." } ] }, { "id": 509618, "postDate": "2019-04-08T03:29:52.003Z", "content": "I have a brief look at the dataset.\nGiven the diverse photo background from different designers, I think if we train a model with high accuracy, maybe the neural networks are classifying the overall photo statistics (e.g. background/light condition/...), rather than the clothes designs themselves.\nThis may make the trained model meaningless.
", "rawMarkdown": "I have a brief look at the dataset.\nGiven the diverse photo background from different designers, I think if we train a model with high accuracy, maybe the neural networks are classifying the overall photo statistics (e.g. background/light condition/...), rather than the clothes designs themselves.\nThis may make the trained model meaningless." }, { "id": 514085, "postDate": "2019-04-11T12:47:55.540Z", "rawMarkdown": "", "isDeleted": true }, { "id": 513384, "postDate": "2019-04-11T05:31:26.647Z", "rawMarkdown": "", "isDeleted": true, "replies": [ { "id": 513450, "postDate": "2019-04-11T06:08:12.050Z", "rawMarkdown": "", "isDeleted": true }, { "id": 513509, "postDate": "2019-04-11T06:50:59.023Z", "content": "The test set has been changed to version2. Why not clear the LB? Besides, I found the new test set removed the background information, so why not do the same operation to the train set , and also provide a new train set ? Otherwise, the two sets are quite different.
", "rawMarkdown": "The test set has been changed to version2. Why not clear the LB? Besides, I found the new test set removed the background information, so why not do the same operation to the train set , and also provide a new train set ? Otherwise, the two sets are quite different. ", "votes": 2 } ] }, { "id": 510325, "postDate": "2019-04-09T00:54:04.967Z", "rawMarkdown": "", "votes": 1, "isDeleted": true }, { "id": 509666, "postDate": "2019-04-08T05:41:53.107Z", "rawMarkdown": "", "votes": 1, "isDeleted": true } ], "comments": [ { "id": 529113, "author_name": "hengck23", "author_url": "", "post_date": "2019-05-09T08:12:45.273000", "content": "\"I think if we train a model with high accuracy, maybe the neural networks are classifying the overall photo statistics\"
\n\ni also note that the same model will wear the clothes by the same designer in the same fashion show. We know it is the same model because she can wear the same shoes
", "votes": 0, "replies": [ { "id": 530065, "author_name": "Gary", "author_url": "", "post_date": "2019-05-11T16:02:52.310000", "content": "Thank you, Heng, a nice idea. Although it is possible to classify style based on the model, I think this is not the original requirement of this competition?
", "votes": 0, "replies": [] } ] }, { "id": 526020, "author_name": "Appian", "author_url": "", "post_date": "2019-05-02T07:43:20.333000", "content": "Is there a way to get version 1 of the dataset now?
", "votes": 0, "replies": [] }, { "id": 523613, "author_name": "Appy Patel", "author_url": "", "post_date": "2019-04-26T16:11:05.740000", "content": "What image shape are u feeding in ur model??
", "votes": 0, "replies": [] }, { "id": 514981, "author_name": "lzhbrian", "author_url": "", "post_date": "2019-04-12T04:58:17.880000", "content": "I saw the new test set, and found a few pictures are ruined.\nFor example, 98.png, 173.png, 233.png, 252.png, ...\nIs this intended, or it's a mistake?\nThanks
", "votes": 0, "replies": [ { "id": 516277, "author_name": "Hearst Machine Learning", "author_url": "", "post_date": "2019-04-13T23:18:52.143000", "content": "We applied post-processing to a the test set to attempt to remove the background from around the fashion models / apparel, in order to better reward approaches that are identifying features in the fashion design itself (rather than, for example) the color of the background or pattern of the floor. In some cases that algorithm removed some portions of the apparel as well. We’ve corrected most of these, but some remain. We do not presently intend to further update the test set.
", "votes": 0, "replies": [] } ] }, { "id": 514085, "author_name": "", "author_url": "", "post_date": "2019-04-11T12:47:55.540000", "content": "", "votes": 0, "replies": [] }, { "id": 513384, "author_name": "", "author_url": "", "post_date": "2019-04-11T05:31:26.647000", "content": "", "votes": 0, "replies": [ { "id": 513450, "author_name": "", "author_url": "", "post_date": "2019-04-11T06:08:12.050000", "content": "", "votes": 0, "replies": [] }, { "id": 513509, "author_name": "Lei", "author_url": "", "post_date": "2019-04-11T06:50:59.023000", "content": "The test set has been changed to version2. Why not clear the LB? Besides, I found the new test set removed the background information, so why not do the same operation to the train set , and also provide a new train set ? Otherwise, the two sets are quite different.
", "votes": 2, "replies": [] } ] }, { "id": 510325, "author_name": "", "author_url": "", "post_date": "2019-04-09T00:54:04.967000", "content": "", "votes": 1, "replies": [] }, { "id": 509666, "author_name": "", "author_url": "", "post_date": "2019-04-08T05:41:53.107000", "content": "", "votes": 1, "replies": [] } ], "raw_markdown_by_id": { "529113": "\"I think if we train a model with high accuracy, maybe the neural networks are classifying the overall photo statistics\"\n\ni also note that the same model will wear the clothes by the same designer in the same fashion show. We know it is the same model because she can wear the same shoes", "526020": "Is there a way to get version 1 of the dataset now?", "523613": "What image shape are u feeding in ur model??", "514981": "I saw the new test set, and found a few pictures are ruined.\nFor example, 98.png, 173.png, 233.png, 252.png, ...\nIs this intended, or it's a mistake?\nThanks", "509618": "I have a brief look at the dataset.\nGiven the diverse photo background from different designers, I think if we train a model with high accuracy, maybe the neural networks are classifying the overall photo statistics (e.g. background/light condition/...), rather than the clothes designs themselves.\nThis may make the trained model meaningless.", "514085": "", "513384": "", "510325": "", "509666": "" } }