I read the competition rule and it says: \"Submissions may not use or incorporate information from hand labeling or human prediction of the validation dataset or test data records.\"
\nMy question is that can I label some segmentations like vertebrae, disc or neural sac as assistance to help building and training my model?
", "messages": [ { "id": 2866656, "postDate": "2024-06-11T12:43:28.293Z", "content": "I read the competition rule and it says: \"Submissions may not use or incorporate information from hand labeling or human prediction of the validation dataset or test data records.\"
\nMy question is that can I label some segmentations like vertebrae, disc or neural sac as assistance to help building and training my model?
", "rawMarkdown": "I read the competition rule and it says: \"Submissions may not use or incorporate information from hand labeling or human prediction of the validation dataset or test data records.\"\n\nMy question is that can I label some segmentations like vertebrae, disc or neural sac as assistance to help building and training my model?", "votes": 4 }, { "id": 2923547, "postDate": "2024-07-16T01:23:48.820Z", "content": "Maybe yes, because I remember that there are cases in which the coordinates do not correspond when viewing the image. In other cases it may be better to use techniques to impute data and in other cases it may be advisable to train a model to predict missing data.
", "rawMarkdown": "Maybe yes, because I remember that there are cases in which the coordinates do not correspond when viewing the image. In other cases it may be better to use techniques to impute data and in other cases it may be advisable to train a model to predict missing data." }, { "id": 2874243, "postDate": "2024-06-16T06:10:53.863Z", "content": "I think that the rules mean that you can not hand-label the final predictions with the intent of a better score. However, I think that hand-labelled datasets, when made freely and publicly available, can be used to train the model. I am not sure.
", "rawMarkdown": "I think that the rules mean that you can not hand-label the final predictions with the intent of a better score. However, I think that hand-labelled datasets, when made freely and publicly available, can be used to train the model. I am not sure." }, { "id": 2867741, "postDate": "2024-06-12T04:24:37.963Z", "content": "I'm reading that as you can't use ground truth labels from the secret test or validation set? Normally, I've seen people correct labels in other competitions so I wouldn't see why you couldn't do that here on the train data.
", "rawMarkdown": "I'm reading that as you can't use ground truth labels from the secret test or validation set? Normally, I've seen people correct labels in other competitions so I wouldn't see why you couldn't do that here on the train data." }, { "id": 2866834, "postDate": "2024-06-11T14:30:34.640Z", "content": "I don't think it's a good idea to label the dataset yourself.
", "rawMarkdown": " I don't think it's a good idea to label the dataset yourself.\n" }, { "id": 2866703, "postDate": "2024-06-11T13:05:18.380Z", "content": "@sohier , dear host, is it OK to label training dataset by myself?
", "rawMarkdown": "@sohier , dear host, is it OK to label training dataset by myself?" }, { "id": 2866698, "postDate": "2024-06-11T13:03:52.680Z", "rawMarkdown": "", "isDeleted": true } ], "comments": [ { "id": 2923547, "author_name": "LUIS ADRIAN MARTINEZ PEREZ", "author_url": "", "post_date": "2024-07-16T01:23:48.820000", "content": "Maybe yes, because I remember that there are cases in which the coordinates do not correspond when viewing the image. In other cases it may be better to use techniques to impute data and in other cases it may be advisable to train a model to predict missing data.
", "votes": 0, "replies": [] }, { "id": 2874243, "author_name": "Devsya ", "author_url": "", "post_date": "2024-06-16T06:10:53.863000", "content": "I think that the rules mean that you can not hand-label the final predictions with the intent of a better score. However, I think that hand-labelled datasets, when made freely and publicly available, can be used to train the model. I am not sure.
", "votes": 0, "replies": [] }, { "id": 2867741, "author_name": "Matt S.", "author_url": "", "post_date": "2024-06-12T04:24:37.963000", "content": "I'm reading that as you can't use ground truth labels from the secret test or validation set? Normally, I've seen people correct labels in other competitions so I wouldn't see why you couldn't do that here on the train data.
", "votes": 0, "replies": [] }, { "id": 2866834, "author_name": "Staru09", "author_url": "", "post_date": "2024-06-11T14:30:34.640000", "content": "I don't think it's a good idea to label the dataset yourself.
", "votes": 0, "replies": [] }, { "id": 2866703, "author_name": "Hajimi Nanbeilvdou", "author_url": "", "post_date": "2024-06-11T13:05:18.380000", "content": "@sohier , dear host, is it OK to label training dataset by myself?
", "votes": 0, "replies": [] }, { "id": 2866698, "author_name": "", "author_url": "", "post_date": "2024-06-11T13:03:52.680000", "content": "", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "2866656": "I read the competition rule and it says: \"Submissions may not use or incorporate information from hand labeling or human prediction of the validation dataset or test data records.\"\n\nMy question is that can I label some segmentations like vertebrae, disc or neural sac as assistance to help building and training my model?", "2923547": "Maybe yes, because I remember that there are cases in which the coordinates do not correspond when viewing the image. In other cases it may be better to use techniques to impute data and in other cases it may be advisable to train a model to predict missing data.", "2874243": "I think that the rules mean that you can not hand-label the final predictions with the intent of a better score. However, I think that hand-labelled datasets, when made freely and publicly available, can be used to train the model. I am not sure.", "2867741": "I'm reading that as you can't use ground truth labels from the secret test or validation set? Normally, I've seen people correct labels in other competitions so I wouldn't see why you couldn't do that here on the train data.", "2866834": " I don't think it's a good idea to label the dataset yourself.\n", "2866703": "@sohier , dear host, is it OK to label training dataset by myself?", "2866698": "" } }