I checked the labels and it seems like none of the motors span across multiple slices. I haven't verified this visually yet, so please correct me if I'm wrong.
\nI did some research and found that flagellar motors are typically around 30–50 nm in diameter, depending on the species. The minimum voxel spacing is 6.5 angstrom (0.65 nm) and maximum voxel spacing is 19.7 angstrom (1.97 nm). Each voxel represents a chunk of space that is much smaller than the motor itself.
\nFor example if voxel spacing is 15.6 angstrom, then a 30 nm motor would be seen on 19 slices. A 50 nm motor would be seen on 32 slices. If my calculations are correct, labels are too weak in my opinion. Besides, it's not very clear what labels refer to. Are they the center coordinates of the motors?
", "messages": [ { "id": "3144307", "postDate": "03/08/2025 07:16:44", "content": "I checked the labels and it seems like none of the motors span across multiple slices. I haven't verified this visually yet, so please correct me if I'm wrong.
\nI did some research and found that flagellar motors are typically around 30–50 nm in diameter, depending on the species. The minimum voxel spacing is 6.5 angstrom (0.65 nm) and maximum voxel spacing is 19.7 angstrom (1.97 nm). Each voxel represents a chunk of space that is much smaller than the motor itself.
\nFor example if voxel spacing is 15.6 angstrom, then a 30 nm motor would be seen on 19 slices. A 50 nm motor would be seen on 32 slices. If my calculations are correct, labels are too weak in my opinion. Besides, it's not very clear what labels refer to. Are they the center coordinates of the motors?
", "rawMarkdown": "I checked the labels and it seems like none of the motors span across multiple slices. I haven't verified this visually yet, so please correct me if I'm wrong.\n\nI did some research and found that flagellar motors are typically around 30–50 nm in diameter, depending on the species. The minimum voxel spacing is 6.5 angstrom (0.65 nm) and maximum voxel spacing is 19.7 angstrom (1.97 nm). Each voxel represents a chunk of space that is much smaller than the motor itself.\n\nFor example if voxel spacing is 15.6 angstrom, then a 30 nm motor would be seen on 19 slices. A 50 nm motor would be seen on 32 slices. If my calculations are correct, labels are too weak in my opinion. Besides, it's not very clear what labels refer to. Are they the center coordinates of the motors?", "votes": null }, { "id": "3144358", "postDate": "03/08/2025 08:51:33", "content": "Isn't it written in the Metric of the Overview?
\nIf the Euclidean distance from y_true is within 1000 Angstroms, it is considered True Positive.
\nIn other words, aren't the positive examples forming a sphere in three-dimensional space, with the given coordinates indicating its center?
Yeah, I missed that part. Predictions have to be within 1000 angstroms probably compensates labels being weak. You are right, they form a sphere, so it's kinda like a regression problem. However, I couldn't find anything about coordinates are the center of the motors. Is it written anywhere?
\nIf coordinates are centers of flagellar motors, then it raises the question \"What is considered the center\" which is asked here.
", "rawMarkdown": "Yeah, I missed that part. Predictions have to be within 1000 angstroms probably compensates labels being weak. You are right, they form a sphere, so it's kinda like a regression problem. However, I couldn't find anything about coordinates are the center of the motors. Is it written anywhere?\n\nIf coordinates are centers of flagellar motors, then it raises the question \"What is considered the center\" which is asked [here](https://www.kaggle.com/competitions/byu-locating-bacterial-flagellar-motors-2025/discussion/566137#3142843).", "votes": null }, { "id": "3144428", "postDate": "03/08/2025 10:58:25", "content": "\n\nyou are tasked with finding flagellar motor centers in 3D tomograms
\n
The dataset description says something like this, so it seems certain that the coordinates are \"the center of the motor.\"
\nHowever, You're right.
\nThe annotator's criteria are not specified. I expect the host's response to clarify \"What is considered the center.\"
That's the case if we approximate them as spheres. I am not knowledgeable in the domain, but I found on the internet that they typically are cylinders with a height/depth less than or equal to the radius, so orientation (unknown to us) would influence this as well.
", "rawMarkdown": "That's the case if we approximate them as spheres. I am not knowledgeable in the domain, but I found on the internet that they typically are cylinders with a height/depth less than or equal to the radius, so orientation (unknown to us) would influence this as well.", "votes": null }, { "id": "3144640", "postDate": "03/08/2025 16:44:38", "content": "Thank you all for pointing this out—we will consider clarifying the description of the annotations in the data description. The labels are the coordinates of the center of the motor, not a mask of the motor, hence why the label is on only one slice.
", "rawMarkdown": "Thank you all for pointing this out—we will consider clarifying the description of the annotations in the data description. The labels are the coordinates of the center of the motor, not a mask of the motor, hence why the label is on only one slice.", "votes": null }, { "id": "3145009", "postDate": "03/09/2025 07:22:01", "content": "I'm not sure but a 3D CNN may implicitly generalize to different orientations and metric should be enough to compensate for label noise.
", "rawMarkdown": "I'm not sure but a 3D CNN may implicitly generalize to different orientations and metric should be enough to compensate for label noise.", "votes": null } ], "comments": [ { "id": 3144358, "author_name": "tomokihirose", "author_url": "", "post_date": "03/08/2025 08:51:33", "content": "Isn't it written in the Metric of the Overview?
\nIf the Euclidean distance from y_true is within 1000 Angstroms, it is considered True Positive.
\nIn other words, aren't the positive examples forming a sphere in three-dimensional space, with the given coordinates indicating its center?
Yeah, I missed that part. Predictions have to be within 1000 angstroms probably compensates labels being weak. You are right, they form a sphere, so it's kinda like a regression problem. However, I couldn't find anything about coordinates are the center of the motors. Is it written anywhere?
\nIf coordinates are centers of flagellar motors, then it raises the question \"What is considered the center\" which is asked here.
", "votes": null, "replies": [ { "id": 3144428, "author_name": "tomokihirose", "author_url": "", "post_date": "03/08/2025 10:58:25", "content": "\n\nyou are tasked with finding flagellar motor centers in 3D tomograms
\n
The dataset description says something like this, so it seems certain that the coordinates are \"the center of the motor.\"
\nHowever, You're right.
\nThe annotator's criteria are not specified. I expect the host's response to clarify \"What is considered the center.\"
That's the case if we approximate them as spheres. I am not knowledgeable in the domain, but I found on the internet that they typically are cylinders with a height/depth less than or equal to the radius, so orientation (unknown to us) would influence this as well.
", "votes": null, "replies": [ { "id": 3145009, "author_name": "gunesevitan", "author_url": "", "post_date": "03/09/2025 07:22:01", "content": "I'm not sure but a 3D CNN may implicitly generalize to different orientations and metric should be enough to compensate for label noise.
", "votes": null, "replies": [] } ] }, { "id": 3144640, "author_name": "jacksonpond", "author_url": "", "post_date": "03/08/2025 16:44:38", "content": "Thank you all for pointing this out—we will consider clarifying the description of the annotations in the data description. The labels are the coordinates of the center of the motor, not a mask of the motor, hence why the label is on only one slice.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "3144307": "I checked the labels and it seems like none of the motors span across multiple slices. I haven't verified this visually yet, so please correct me if I'm wrong.\n\nI did some research and found that flagellar motors are typically around 30–50 nm in diameter, depending on the species. The minimum voxel spacing is 6.5 angstrom (0.65 nm) and maximum voxel spacing is 19.7 angstrom (1.97 nm). Each voxel represents a chunk of space that is much smaller than the motor itself.\n\nFor example if voxel spacing is 15.6 angstrom, then a 30 nm motor would be seen on 19 slices. A 50 nm motor would be seen on 32 slices. If my calculations are correct, labels are too weak in my opinion. Besides, it's not very clear what labels refer to. Are they the center coordinates of the motors?", "3144358": "Isn't it written in the Metric of the Overview?\nIf the Euclidean distance from y_true is within 1000 Angstroms, it is considered True Positive.\nIn other words, aren't the positive examples forming a sphere in three-dimensional space, with the given coordinates indicating its center?", "3144378": "Yeah, I missed that part. Predictions have to be within 1000 angstroms probably compensates labels being weak. You are right, they form a sphere, so it's kinda like a regression problem. However, I couldn't find anything about coordinates are the center of the motors. Is it written anywhere?\n\nIf coordinates are centers of flagellar motors, then it raises the question \"What is considered the center\" which is asked [here](https://www.kaggle.com/competitions/byu-locating-bacterial-flagellar-motors-2025/discussion/566137#3142843).", "3144428": ">you are tasked with finding flagellar motor centers in 3D tomograms\n\nThe dataset description says something like this, so it seems certain that the coordinates are \"the center of the motor.\"\nHowever, You're right.\nThe annotator's criteria are not specified. I expect the host's response to clarify \"What is considered the center.\"", "3144519": "That's the case if we approximate them as spheres. I am not knowledgeable in the domain, but I found on the internet that they typically are cylinders with a height/depth less than or equal to the radius, so orientation (unknown to us) would influence this as well.", "3144640": "Thank you all for pointing this out—we will consider clarifying the description of the annotations in the data description. The labels are the coordinates of the center of the motor, not a mask of the motor, hence why the label is on only one slice.", "3145009": "I'm not sure but a 3D CNN may implicitly generalize to different orientations and metric should be enough to compensate for label noise." }, "source": "meta" }