{ "id": 547091, "title": "Particles Overlapping", "url": "/competitions/czii-cryo-et-object-identification/discussion/547091", "author_name": "", "post_date": "2024-11-19T18:08:24.126430200Z", "votes": 12, "comment_count": 11, "views": 0, "content": "

I noticed some particles are overlapping in the first training dataset that I am looking. For instance, at z=56, the beta-amylase and beta-galactosidase (blue-green):

\n

\n

When using a network with semantic segmentation UNET, how can I deal with this, removing overlaps? I assume using an instance segmentation method like Mask RCNN would be more beneficial?

", "messages": [ { "id": "3050020", "postDate": "11/19/2024 18:08:24", "content": "

I noticed some particles are overlapping in the first training dataset that I am looking. For instance, at z=56, the beta-amylase and beta-galactosidase (blue-green):

\n

\n

When using a network with semantic segmentation UNET, how can I deal with this, removing overlaps? I assume using an instance segmentation method like Mask RCNN would be more beneficial?

", "rawMarkdown": "I noticed some particles are overlapping in the first training dataset that I am looking. For instance, at z=56, the beta-amylase and beta-galactosidase (blue-green):\n\n\n\nWhen using a network with semantic segmentation UNET, how can I deal with this, removing overlaps? I assume using an instance segmentation method like Mask RCNN would be more beneficial?", "votes": null }, { "id": "3050031", "postDate": "11/19/2024 18:36:35", "content": "

Hi Ignasi, can you please let me know which tomogram this is? We dealt with all the overlaps one-by-one. The overlaps were a result of having multiple annotation workflows. But this one must have slipped or it's a result of some error.

", "rawMarkdown": "Hi Ignasi, can you please let me know which tomogram this is? We dealt with all the overlaps one-by-one. The overlaps were a result of having multiple annotation workflows. But this one must have slipped or it's a result of some error.", "votes": null }, { "id": "3050181", "postDate": "11/19/2024 22:42:10", "content": "

That's TS_5_4. The particle locations - \"x\": 5417.091, \"y\": 1120.574, \"z\": 567.527' (beta-amylase) and \"x\": 5421.651, \"y\": 1158.868, \"z\": 563.302 (beta-galactosidase)

", "rawMarkdown": "That's TS_5_4. The particle locations - \"x\": 5417.091, \"y\": 1120.574, \"z\": 567.527' (beta-amylase) and \"x\": 5421.651, \"y\": 1158.868, \"z\": 563.302 (beta-galactosidase)", "votes": null }, { "id": "3050199", "postDate": "11/19/2024 23:14:07", "content": "

\" But this one must have slipped or it's a result of some error….\"

\n

this means that none of the annotations are supposed to overlap
\n(At least, the physical particles in the phantom sample don't overlap)

", "rawMarkdown": "\" But this one must have slipped or it's a result of some error....\"\n\nthis means that none of the annotations are supposed to overlap\n(At least, the physical particles in the phantom sample don't overlap)", "votes": null }, { "id": "3050221", "postDate": "11/20/2024 00:33:54", "content": "

Thanks for bringing this to our attention. This was indeed human error. The overlaps that were resolved were not propagated. We will work to update the coordinates across all 492 tomograms. The change will be minimal. Here are the number of affected particles:
\napo-ferritin 40
\nvirus-like-particle 0
\nribosome 196
\nthyroglobulin 113
\nbeta-amylase 65
\nbeta-galactosidase 186

", "rawMarkdown": "Thanks for bringing this to our attention. This was indeed human error. The overlaps that were resolved were not propagated. We will work to update the coordinates across all 492 tomograms. The change will be minimal. Here are the number of affected particles:\napo-ferritin 40\nvirus-like-particle 0\nribosome 196\nthyroglobulin 113\nbeta-amylase 65\nbeta-galactosidase 186", "votes": null }, { "id": "3050615", "postDate": "11/20/2024 11:33:56", "content": "

Thank you for replying so fast! 🚀

", "rawMarkdown": "Thank you for replying so fast! 🚀", "votes": null }, { "id": "3052712", "postDate": "11/22/2024 18:22:53", "content": "

Howdy folks,

\n

Just to follow up on this, we've tried rescoring with a version of the dataset that is corrected for these overlaps and have found that the score changes are minimal and ranking remained consistent through our test.

\n

Therefore we do not plan to make any updates.

\n

Cheers and thanks all,
\nKyle

", "rawMarkdown": "Howdy folks,\n\nJust to follow up on this, we've tried rescoring with a version of the dataset that is corrected for these overlaps and have found that the score changes are minimal and ranking remained consistent through our test. \n\nTherefore we do not plan to make any updates. \n\nCheers and thanks all,\nKyle", "votes": null }, { "id": "3053438", "postDate": "11/23/2024 13:37:06", "content": "

So now train and test dataset are updated or all left as it was?

", "rawMarkdown": "So now train and test dataset are updated or all left as it was?", "votes": null }, { "id": "3055035", "postDate": "11/25/2024 13:04:58", "content": "

The dataset was not updated. We wanted to minimize disruption and found that score changes were minimal/ranking was consistent for our sample.

", "rawMarkdown": "The dataset was not updated. We wanted to minimize disruption and found that score changes were minimal/ranking was consistent for our sample.", "votes": null }, { "id": "3058204", "postDate": "11/29/2024 07:55:43", "content": "

Please apologize if it's a dump or inapropiate question. But how the overlap of two particles on train (one of them not asked in submission) can affect test samples?

", "rawMarkdown": "Please apologize if it's a dump or inapropiate question. But how the overlap of two particles on train (one of them not asked in submission) can affect test samples?", "votes": null }, { "id": "3058236", "postDate": "11/29/2024 08:50:06", "content": "

It's not just two particles, read above there were multiple overlaps in train and test set.

\n

'apo-ferritin 40
\nvirus-like-particle 0
\nribosome 196
\nthyroglobulin 113
\nbeta-amylase 65
\nbeta-galactosidase 186'

\n

This has been solved but they have realised that the effects on LB were minimal, so they decided not to update it.

", "rawMarkdown": "It's not just two particles, read above there were multiple overlaps in train and test set.\n\n 'apo-ferritin 40\nvirus-like-particle 0\nribosome 196\nthyroglobulin 113\nbeta-amylase 65\nbeta-galactosidase 186'\n\nThis has been solved but they have realised that the effects on LB were minimal, so they decided not to update it.", "votes": null }, { "id": "3058272", "postDate": "11/29/2024 09:19:42", "content": "

Thanks. I I thought those particles were all in test and consequence of this single two in train. Ribosomes and apo-ferritin are very stacked so I wouldn't tag them as annotation human error. Now makes sense.

", "rawMarkdown": "Thanks. I I thought those particles were all in test and consequence of this single two in train. Ribosomes and apo-ferritin are very stacked so I wouldn't tag them as annotation human error. Now makes sense.", "votes": null } ], "comments": [ { "id": 3050031, "author_name": "rezaparaan", "author_url": "", "post_date": "11/19/2024 18:36:35", "content": "

Hi Ignasi, can you please let me know which tomogram this is? We dealt with all the overlaps one-by-one. The overlaps were a result of having multiple annotation workflows. But this one must have slipped or it's a result of some error.

", "votes": null, "replies": [ { "id": 3050181, "author_name": "ignasialemany", "author_url": "", "post_date": "11/19/2024 22:42:10", "content": "

That's TS_5_4. The particle locations - \"x\": 5417.091, \"y\": 1120.574, \"z\": 567.527' (beta-amylase) and \"x\": 5421.651, \"y\": 1158.868, \"z\": 563.302 (beta-galactosidase)

", "votes": null, "replies": [ { "id": 3050221, "author_name": "rezaparaan", "author_url": "", "post_date": "11/20/2024 00:33:54", "content": "

Thanks for bringing this to our attention. This was indeed human error. The overlaps that were resolved were not propagated. We will work to update the coordinates across all 492 tomograms. The change will be minimal. Here are the number of affected particles:
\napo-ferritin 40
\nvirus-like-particle 0
\nribosome 196
\nthyroglobulin 113
\nbeta-amylase 65
\nbeta-galactosidase 186

", "votes": null, "replies": [ { "id": 3050615, "author_name": "ignasialemany", "author_url": "", "post_date": "11/20/2024 11:33:56", "content": "

Thank you for replying so fast! 🚀

", "votes": null, "replies": [] }, { "id": 3052712, "author_name": "kharrington", "author_url": "", "post_date": "11/22/2024 18:22:53", "content": "

Howdy folks,

\n

Just to follow up on this, we've tried rescoring with a version of the dataset that is corrected for these overlaps and have found that the score changes are minimal and ranking remained consistent through our test.

\n

Therefore we do not plan to make any updates.

\n

Cheers and thanks all,
\nKyle

", "votes": null, "replies": [ { "id": 3053438, "author_name": "hypocrites", "author_url": "", "post_date": "11/23/2024 13:37:06", "content": "

So now train and test dataset are updated or all left as it was?

", "votes": null, "replies": [ { "id": 3055035, "author_name": "kharrington", "author_url": "", "post_date": "11/25/2024 13:04:58", "content": "

The dataset was not updated. We wanted to minimize disruption and found that score changes were minimal/ranking was consistent for our sample.

", "votes": null, "replies": [] } ] } ] }, { "id": 3058204, "author_name": "sacuscreed", "author_url": "", "post_date": "11/29/2024 07:55:43", "content": "

Please apologize if it's a dump or inapropiate question. But how the overlap of two particles on train (one of them not asked in submission) can affect test samples?

", "votes": null, "replies": [ { "id": 3058236, "author_name": "ignasialemany", "author_url": "", "post_date": "11/29/2024 08:50:06", "content": "

It's not just two particles, read above there were multiple overlaps in train and test set.

\n

'apo-ferritin 40
\nvirus-like-particle 0
\nribosome 196
\nthyroglobulin 113
\nbeta-amylase 65
\nbeta-galactosidase 186'

\n

This has been solved but they have realised that the effects on LB were minimal, so they decided not to update it.

", "votes": null, "replies": [ { "id": 3058272, "author_name": "sacuscreed", "author_url": "", "post_date": "11/29/2024 09:19:42", "content": "

Thanks. I I thought those particles were all in test and consequence of this single two in train. Ribosomes and apo-ferritin are very stacked so I wouldn't tag them as annotation human error. Now makes sense.

", "votes": null, "replies": [] } ] } ] } ] } ] }, { "id": 3050199, "author_name": "hengck23", "author_url": "", "post_date": "11/19/2024 23:14:07", "content": "

\" But this one must have slipped or it's a result of some error….\"

\n

this means that none of the annotations are supposed to overlap
\n(At least, the physical particles in the phantom sample don't overlap)

", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "3050020": "I noticed some particles are overlapping in the first training dataset that I am looking. For instance, at z=56, the beta-amylase and beta-galactosidase (blue-green):\n\n\n\nWhen using a network with semantic segmentation UNET, how can I deal with this, removing overlaps? I assume using an instance segmentation method like Mask RCNN would be more beneficial?", "3050031": "Hi Ignasi, can you please let me know which tomogram this is? We dealt with all the overlaps one-by-one. The overlaps were a result of having multiple annotation workflows. But this one must have slipped or it's a result of some error.", "3050181": "That's TS_5_4. The particle locations - \"x\": 5417.091, \"y\": 1120.574, \"z\": 567.527' (beta-amylase) and \"x\": 5421.651, \"y\": 1158.868, \"z\": 563.302 (beta-galactosidase)", "3050199": "\" But this one must have slipped or it's a result of some error....\"\n\nthis means that none of the annotations are supposed to overlap\n(At least, the physical particles in the phantom sample don't overlap)", "3050221": "Thanks for bringing this to our attention. This was indeed human error. The overlaps that were resolved were not propagated. We will work to update the coordinates across all 492 tomograms. The change will be minimal. Here are the number of affected particles:\napo-ferritin 40\nvirus-like-particle 0\nribosome 196\nthyroglobulin 113\nbeta-amylase 65\nbeta-galactosidase 186", "3050615": "Thank you for replying so fast! 🚀", "3052712": "Howdy folks,\n\nJust to follow up on this, we've tried rescoring with a version of the dataset that is corrected for these overlaps and have found that the score changes are minimal and ranking remained consistent through our test. \n\nTherefore we do not plan to make any updates. \n\nCheers and thanks all,\nKyle", "3053438": "So now train and test dataset are updated or all left as it was?", "3055035": "The dataset was not updated. We wanted to minimize disruption and found that score changes were minimal/ranking was consistent for our sample.", "3058204": "Please apologize if it's a dump or inapropiate question. But how the overlap of two particles on train (one of them not asked in submission) can affect test samples?", "3058236": "It's not just two particles, read above there were multiple overlaps in train and test set.\n\n 'apo-ferritin 40\nvirus-like-particle 0\nribosome 196\nthyroglobulin 113\nbeta-amylase 65\nbeta-galactosidase 186'\n\nThis has been solved but they have realised that the effects on LB were minimal, so they decided not to update it.", "3058272": "Thanks. I I thought those particles were all in test and consequence of this single two in train. Ribosomes and apo-ferritin are very stacked so I wouldn't tag them as annotation human error. Now makes sense." }, "source": "meta" }