Hi,
\nI have a question about classes 6 (Endoplasmic reticulum) and 10 (Microtubules). If my understanding is correct, we must identify various organelles present in cells in the test set. To identify regions occupied by the ER and Microtubules, can't we just look at the yellow and red channels of the image? Does this mean that for images in the test set, I can predict the location of the 6th and 10th class with confidence 1 since they must correspond with the yellow and red channels without any model?
This seems to be discussed a little here, but I'm still confused. Any help would be appreciated.
", "messages": [ { "id": 1227510, "postDate": "2021-03-05T15:52:16.703Z", "content": "You don't predict the organelles but in which organelle the proteines (green channel) are located.
\nYou can assume, that if the proteines for example only overlap with the nucleus (blue channel), they won't be localized in the ER or Microtubuli. But you cannot assume, that if for example the proteines overlap with the ER or Microtubuli, they are actually located in the ER or Microtubuli (because there are organelles that also overlap with ER and Microtubuli).
", "rawMarkdown": "You don't predict the organelles but in which organelle the proteines (green channel) are located. \n\nYou can assume, that if the proteines for example only overlap with the nucleus (blue channel), they won't be localized in the ER or Microtubuli. But you cannot assume, that if for example the proteines overlap with the ER or Microtubuli, they are actually located in the ER or Microtubuli (because there are organelles that also overlap with ER and Microtubuli).", "votes": 1, "replies": [ { "id": 1227558, "postDate": "2021-03-05T16:44:26.933Z", "content": "Thanks for your reply. Just to confirm, for the test images, we look at each cell and what proteins they have and then predict which organelles they are located in + generate masks for the organelles, right?
", "rawMarkdown": "Thanks for your reply. Just to confirm, for the test images, we look at each cell and what proteins they have and then predict which organelles they are located in + generate masks for the organelles, right?", "replies": [ { "id": 1227576, "postDate": "2021-03-05T16:53:16.533Z", "content": "no, you're supposed to look at each cell, generate a mask for the cell and predict, in which organelle(s) the proteine(s) are located in every cell (the proteines can be located in multiple organelles)
", "rawMarkdown": "no, you're supposed to look at each cell, generate a mask for the cell and predict, in which organelle(s) the proteine(s) are located in every cell (the proteines can be located in multiple organelles)", "votes": 1 } ] }, { "id": 1227600, "postDate": "2021-03-05T17:04:54.877Z", "content": "Ohh, I understand it now. Thank you so much 🙂
", "rawMarkdown": "Ohh, I understand it now. Thank you so much 🙂", "votes": 1 } ] }, { "id": 1227470, "postDate": "2021-03-05T15:11:58.223Z", "content": "Hi,
\nI have a question about classes 6 (Endoplasmic reticulum) and 10 (Microtubules). If my understanding is correct, we must identify various organelles present in cells in the test set. To identify regions occupied by the ER and Microtubules, can't we just look at the yellow and red channels of the image? Does this mean that for images in the test set, I can predict the location of the 6th and 10th class with confidence 1 since they must correspond with the yellow and red channels without any model?
This seems to be discussed a little here, but I'm still confused. Any help would be appreciated.
", "rawMarkdown": "Hi,\nI have a question about classes 6 (Endoplasmic reticulum) and 10 (Microtubules). If my understanding is correct, we must identify various organelles present in cells in the test set. To identify regions occupied by the ER and Microtubules, can't we just look at the yellow and red channels of the image? Does this mean that for images in the test set, I can predict the location of the 6th and 10th class with confidence 1 since they must correspond with the yellow and red channels without any model?\n\nThis seems to be discussed a little [here](https://www.kaggle.com/c/hpa-single-cell-image-classification/discussion/215736), but I'm still confused. Any help would be appreciated.", "votes": 1 } ], "comments": [ { "id": 1227510, "author_name": "Alexander Riedel", "author_url": "", "post_date": "2021-03-05T15:52:16.703000", "content": "You don't predict the organelles but in which organelle the proteines (green channel) are located.
\nYou can assume, that if the proteines for example only overlap with the nucleus (blue channel), they won't be localized in the ER or Microtubuli. But you cannot assume, that if for example the proteines overlap with the ER or Microtubuli, they are actually located in the ER or Microtubuli (because there are organelles that also overlap with ER and Microtubuli).
", "votes": 1, "replies": [ { "id": 1227558, "author_name": "novice03", "author_url": "", "post_date": "2021-03-05T16:44:26.933000", "content": "Thanks for your reply. Just to confirm, for the test images, we look at each cell and what proteins they have and then predict which organelles they are located in + generate masks for the organelles, right?
", "votes": 0, "replies": [ { "id": 1227576, "author_name": "Alexander Riedel", "author_url": "", "post_date": "2021-03-05T16:53:16.533000", "content": "no, you're supposed to look at each cell, generate a mask for the cell and predict, in which organelle(s) the proteine(s) are located in every cell (the proteines can be located in multiple organelles)
", "votes": 1, "replies": [] } ] }, { "id": 1227600, "author_name": "novice03", "author_url": "", "post_date": "2021-03-05T17:04:54.877000", "content": "Ohh, I understand it now. Thank you so much 🙂
", "votes": 1, "replies": [] } ] } ], "raw_markdown_by_id": { "1227510": "You don't predict the organelles but in which organelle the proteines (green channel) are located. \n\nYou can assume, that if the proteines for example only overlap with the nucleus (blue channel), they won't be localized in the ER or Microtubuli. But you cannot assume, that if for example the proteines overlap with the ER or Microtubuli, they are actually located in the ER or Microtubuli (because there are organelles that also overlap with ER and Microtubuli).", "1227470": "Hi,\nI have a question about classes 6 (Endoplasmic reticulum) and 10 (Microtubules). If my understanding is correct, we must identify various organelles present in cells in the test set. To identify regions occupied by the ER and Microtubules, can't we just look at the yellow and red channels of the image? Does this mean that for images in the test set, I can predict the location of the 6th and 10th class with confidence 1 since they must correspond with the yellow and red channels without any model?\n\nThis seems to be discussed a little [here](https://www.kaggle.com/c/hpa-single-cell-image-classification/discussion/215736), but I'm still confused. Any help would be appreciated." } }