{ "id": 573200, "title": "What if more than one flagellar is found for the same tomography?", "url": "/competitions/byu-locating-bacterial-flagellar-motors-2025/discussion/573200", "author_name": "", "post_date": "2025-04-14T07:16:41.406410500Z", "votes": 1, "comment_count": 4, "views": 0, "content": "

Hello,
\nfrom the output instructions I can see:

\n
\n

Your submission should be a CSV file with one row per tomogram found in the test set.

\n
\n

But what if for the same tomography I find 2 different flagellars?

", "messages": [ { "id": "3178437", "postDate": "04/14/2025 07:16:41", "content": "

Hello,
\nfrom the output instructions I can see:

\n
\n

Your submission should be a CSV file with one row per tomogram found in the test set.

\n
\n

But what if for the same tomography I find 2 different flagellars?

", "rawMarkdown": "Hello,\nfrom the output instructions I can see:\n>Your submission should be a CSV file with one row per tomogram found in the test set.\n\nBut what if for the same tomography I find 2 different flagellars?", "votes": null }, { "id": "3178455", "postDate": "04/14/2025 07:58:30", "content": "

submit the more/most likely one

\n

alternatively, if you train a softmax selector

", "rawMarkdown": "submit the more/most likely one\n\nalternatively, if you train a softmax selector", "votes": null }, { "id": "3178693", "postDate": "04/14/2025 13:32:53", "content": "

I'd compare the probabilities of the two detections and take the one with the higher. You could also consider other probabilistic ideas such as the normal location of predictions (i.e. identify hotspots for miss detections such as edges).

", "rawMarkdown": "I'd compare the probabilities of the two detections and take the one with the higher. You could also consider other probabilistic ideas such as the normal location of predictions (i.e. identify hotspots for miss detections such as edges).", "votes": null }, { "id": "3178803", "postDate": "04/14/2025 16:21:58", "content": "

From the data page:

\n
\n

The test data only contain tomograms with one or zero motors.

\n
\n

So that shouldn't be an issue.

\n

Edit: thanks for correcting my understanding

", "rawMarkdown": "From the data page:\n\n>The test data only contain tomograms with one or zero motors.\n\nSo that shouldn't be an issue.\n\nEdit: thanks for correcting my understanding", "votes": null }, { "id": "3178933", "postDate": "04/14/2025 19:46:22", "content": "
\n

So that shouldn't be an issue.

\n
\n

I think you misunderstood. Just because there is at most one motor per private test tomograms doesn't mean that one can't predict more than one motor. The original poster was asking what to do in such cases, and one option is to only submit the prediction for the motor with highest confidence.

", "rawMarkdown": "> So that shouldn't be an issue.\n\nI think you misunderstood. Just because there is at most one motor per private test tomograms doesn't mean that one can't predict more than one motor. The original poster was asking what to do in such cases, and one option is to only submit the prediction for the motor with highest confidence.", "votes": null } ], "comments": [ { "id": 3178455, "author_name": "hengck23", "author_url": "", "post_date": "04/14/2025 07:58:30", "content": "

submit the more/most likely one

\n

alternatively, if you train a softmax selector

", "votes": null, "replies": [] }, { "id": 3178693, "author_name": "connorjd", "author_url": "", "post_date": "04/14/2025 13:32:53", "content": "

I'd compare the probabilities of the two detections and take the one with the higher. You could also consider other probabilistic ideas such as the normal location of predictions (i.e. identify hotspots for miss detections such as edges).

", "votes": null, "replies": [] }, { "id": 3178803, "author_name": "andrewjdarley", "author_url": "", "post_date": "04/14/2025 16:21:58", "content": "

From the data page:

\n
\n

The test data only contain tomograms with one or zero motors.

\n
\n

So that shouldn't be an issue.

\n

Edit: thanks for correcting my understanding

", "votes": null, "replies": [ { "id": 3178933, "author_name": "tilii7", "author_url": "", "post_date": "04/14/2025 19:46:22", "content": "
\n

So that shouldn't be an issue.

\n
\n

I think you misunderstood. Just because there is at most one motor per private test tomograms doesn't mean that one can't predict more than one motor. The original poster was asking what to do in such cases, and one option is to only submit the prediction for the motor with highest confidence.

", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "3178437": "Hello,\nfrom the output instructions I can see:\n>Your submission should be a CSV file with one row per tomogram found in the test set.\n\nBut what if for the same tomography I find 2 different flagellars?", "3178455": "submit the more/most likely one\n\nalternatively, if you train a softmax selector", "3178693": "I'd compare the probabilities of the two detections and take the one with the higher. You could also consider other probabilistic ideas such as the normal location of predictions (i.e. identify hotspots for miss detections such as edges).", "3178803": "From the data page:\n\n>The test data only contain tomograms with one or zero motors.\n\nSo that shouldn't be an issue.\n\nEdit: thanks for correcting my understanding", "3178933": "> So that shouldn't be an issue.\n\nI think you misunderstood. Just because there is at most one motor per private test tomograms doesn't mean that one can't predict more than one motor. The original poster was asking what to do in such cases, and one option is to only submit the prediction for the motor with highest confidence." }, "source": "meta" }