This competition is described as \"Weak supervised learning\" because the labels are image-based, but we are supposed to output classes for each cell in the image. However, when looking at the data, many images have multiple labels, so would it be best to just treat this as a normal supervised learning task and see how it does, or is an entirely new approach needed?I would love to know what everybody else thinks would be the best place to start.
", "messages": [ { "id": 1171610, "postDate": "2021-01-27T01:45:42.533Z", "content": "Note that it is not guaranteed that all cells will exhibit all labels that are associated with the image. The labels are created by annotation over multiple images such that the antibody used for the image has been shown to show all those labels in images, but they may not all show in the specific image you are looking at.
", "rawMarkdown": "Note that it is not guaranteed that all cells will exhibit all labels that are associated with the image. The labels are created by annotation over multiple images such that the antibody used for the image has been shown to show all those labels in images, but they may not all show in the specific image you are looking at.", "votes": 6 }, { "id": 1171605, "postDate": "2021-01-27T01:37:47.873Z", "content": "This competition is described as \"Weak supervised learning\" because the labels are image-based, but we are supposed to output classes for each cell in the image. However, when looking at the data, many images have multiple labels, so would it be best to just treat this as a normal supervised learning task and see how it does, or is an entirely new approach needed?I would love to know what everybody else thinks would be the best place to start.
", "rawMarkdown": "This competition is described as \"Weak supervised learning\" because the labels are image-based, but we are supposed to output classes for each cell in the image. However, when looking at the data, many images have multiple labels, so would it be best to just treat this as a normal supervised learning task and see how it does, or is an entirely new approach needed?I would love to know what everybody else thinks would be the best place to start.", "votes": 1 }, { "id": 1171926, "postDate": "2021-01-27T07:16:08.107Z", "content": "Please check this (https://www.kaggle.com/lnhtrang/single-cell-patterns) to understand the patterns more, especially section \"SCV and multi-localization examples\"
", "rawMarkdown": "Please check this (https://www.kaggle.com/lnhtrang/single-cell-patterns) to understand the patterns more, especially section \"SCV and multi-localization examples\"" } ], "comments": [ { "id": 1171610, "author_name": "Casper Winsnes", "author_url": "", "post_date": "2021-01-27T01:45:42.533000", "content": "Note that it is not guaranteed that all cells will exhibit all labels that are associated with the image. The labels are created by annotation over multiple images such that the antibody used for the image has been shown to show all those labels in images, but they may not all show in the specific image you are looking at.
", "votes": 6, "replies": [] }, { "id": 1171926, "author_name": "Trang Le", "author_url": "", "post_date": "2021-01-27T07:16:08.107000", "content": "Please check this (https://www.kaggle.com/lnhtrang/single-cell-patterns) to understand the patterns more, especially section \"SCV and multi-localization examples\"
", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "1171610": "Note that it is not guaranteed that all cells will exhibit all labels that are associated with the image. The labels are created by annotation over multiple images such that the antibody used for the image has been shown to show all those labels in images, but they may not all show in the specific image you are looking at.", "1171605": "This competition is described as \"Weak supervised learning\" because the labels are image-based, but we are supposed to output classes for each cell in the image. However, when looking at the data, many images have multiple labels, so would it be best to just treat this as a normal supervised learning task and see how it does, or is an entirely new approach needed?I would love to know what everybody else thinks would be the best place to start.", "1171926": "Please check this (https://www.kaggle.com/lnhtrang/single-cell-patterns) to understand the patterns more, especially section \"SCV and multi-localization examples\"" } }