{ "id": 428289, "title": "So the answer was No dilate?", "url": "/competitions/hubmap-hacking-the-human-vasculature/discussion/428289", "author_name": "", "post_date": "2023-08-01T00:11:21.986992800Z", "votes": 1, "comment_count": 11, "views": 0, "content": "

In my case, the difference between dilate or no dilate was very large.
\n(Dilate 555->454, No dilate 541 -> 544)

\n

How were the others?

", "messages": [ { "id": "2367927", "postDate": "08/01/2023 00:11:21", "content": "

In my case, the difference between dilate or no dilate was very large.
\n(Dilate 555->454, No dilate 541 -> 544)

\n

How were the others?

", "rawMarkdown": "In my case, the difference between dilate or no dilate was very large.\n(Dilate 555->454, No dilate 541 -> 544)\n\n\nHow were the others?", "votes": null }, { "id": "2367932", "postDate": "08/01/2023 00:14:11", "content": "

yes
\nFinally I did not select no dilate submissions then….

\n

best single model

\n

lb556 pb 461 --> dilate
\nlb495 pb 550 --> no dilate

", "rawMarkdown": "yes\nFinally I did not select no dilate submissions then....\n\nbest single model\n \nlb556 pb 461 --> dilate\nlb495 pb 550 --> no dilate", "votes": null }, { "id": "2367935", "postDate": "08/01/2023 00:17:11", "content": "

yes, also for me,
\nprivate LB: no dilate 0.561, with dilate 0.455. Unfortunately, we didn't select it.

", "rawMarkdown": "yes, also for me, \nprivate LB: no dilate 0.561, with dilate 0.455. Unfortunately, we didn't select it.", "votes": null }, { "id": "2367937", "postDate": "08/01/2023 00:19:25", "content": "

i'm doing a sub without dilate, I'll let you know in a couple of hours :D

", "rawMarkdown": "i'm doing a sub without dilate, I'll let you know in a couple of hours :D", "votes": null }, { "id": "2367968", "postDate": "08/01/2023 00:38:10", "content": "

😢 Thank you for sharing !

", "rawMarkdown": "😢 Thank you for sharing !", "votes": null }, { "id": "2367969", "postDate": "08/01/2023 00:38:47", "content": "

😢 Please share your results

", "rawMarkdown": "😢 Please share your results", "votes": null }, { "id": "2367975", "postDate": "08/01/2023 00:41:10", "content": "

Answer is Dataset 1. My single model with dataset 1 without dialate is Private 0.543, Public 0.51. My best private is 0.597 :(( My final submission is combine model train by dataset 1 and model train by dataset 2 -_- \"\"

", "rawMarkdown": "Answer is Dataset 1. My single model with dataset 1 without dialate is Private 0.543, Public 0.51. My best private is 0.597 :(( My final submission is combine model train by dataset 1 and model train by dataset 2 -_- ![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F2832287%2Fe1c7ada4ec5f11c3470b459e7e225c29%2FScreenshot%20from%202023-08-01%2007-15-12.png?generation=1690850101951840&alt=media)", "votes": null }, { "id": "2367977", "postDate": "08/01/2023 00:42:46", "content": "

My best private and last submission -_-

", "rawMarkdown": "My best private and last submission -_-", "votes": null }, { "id": "2367978", "postDate": "08/01/2023 00:43:18", "content": "

private 0.597 is amazaing… Did you use MaskDino single Model?

", "rawMarkdown": "private 0.597 is amazaing... Did you use MaskDino single Model?", "votes": null }, { "id": "2368115", "postDate": "08/01/2023 02:40:13", "content": "

My mistake is not use only dataset1 for ensemble :(( https://www.kaggle.com/competitions/hubmap-hacking-the-human-vasculature/discussion/428315

", "rawMarkdown": "My mistake is not use only dataset1 for ensemble :(( https://www.kaggle.com/competitions/hubmap-hacking-the-human-vasculature/discussion/428315", "votes": null }, { "id": "2384026", "postDate": "08/10/2023 18:28:49", "content": "

For me no dilate doesn't help [the score doesn't improve on private LB when we drop dilate],
\nwe blame the second dataset, but we're not sure

\n

it's even weirder that our single model scores 0.525 on private, but ensemble scores ~0.39…? something is going on with distributions of masks that neither reflected on public LB or on our CV, but we didn't find whats wrong

", "rawMarkdown": "For me no dilate doesn't help [the score doesn't improve on private LB when we drop dilate], \nwe blame the second dataset, but we're not sure\n\nit's even weirder that our single model scores 0.525 on private, but ensemble scores ~0.39...? something is going on with distributions of masks that neither reflected on public LB or on our CV, but we didn't find whats wrong", "votes": null }, { "id": "2384318", "postDate": "08/11/2023 00:33:41", "content": "

check if mask threshold per pixel correct or not.
\nin my case, averaging all masks and set
\nmask[mask<=0.5]=0

\n

then everythnig is ok

", "rawMarkdown": "check if mask threshold per pixel correct or not.\nin my case, averaging all masks and set \nmask[mask<=0.5]=0 \n\nthen everythnig is ok", "votes": null } ], "comments": [ { "id": 2367932, "author_name": "atom1231", "author_url": "", "post_date": "08/01/2023 00:14:11", "content": "

yes
\nFinally I did not select no dilate submissions then….

\n

best single model

\n

lb556 pb 461 --> dilate
\nlb495 pb 550 --> no dilate

", "votes": null, "replies": [ { "id": 2367968, "author_name": "devchopin", "author_url": "", "post_date": "08/01/2023 00:38:10", "content": "

😢 Thank you for sharing !

", "votes": null, "replies": [] } ] }, { "id": 2367935, "author_name": "tanjiroll", "author_url": "", "post_date": "08/01/2023 00:17:11", "content": "

yes, also for me,
\nprivate LB: no dilate 0.561, with dilate 0.455. Unfortunately, we didn't select it.

", "votes": null, "replies": [] }, { "id": 2367937, "author_name": "martynoveduard", "author_url": "", "post_date": "08/01/2023 00:19:25", "content": "

i'm doing a sub without dilate, I'll let you know in a couple of hours :D

", "votes": null, "replies": [ { "id": 2367969, "author_name": "devchopin", "author_url": "", "post_date": "08/01/2023 00:38:47", "content": "

😢 Please share your results

", "votes": null, "replies": [ { "id": 2384026, "author_name": "martynoveduard", "author_url": "", "post_date": "08/10/2023 18:28:49", "content": "

For me no dilate doesn't help [the score doesn't improve on private LB when we drop dilate],
\nwe blame the second dataset, but we're not sure

\n

it's even weirder that our single model scores 0.525 on private, but ensemble scores ~0.39…? something is going on with distributions of masks that neither reflected on public LB or on our CV, but we didn't find whats wrong

", "votes": null, "replies": [ { "id": 2384318, "author_name": "atom1231", "author_url": "", "post_date": "08/11/2023 00:33:41", "content": "

check if mask threshold per pixel correct or not.
\nin my case, averaging all masks and set
\nmask[mask<=0.5]=0

\n

then everythnig is ok

", "votes": null, "replies": [] } ] } ] } ] }, { "id": 2367975, "author_name": "quan0095", "author_url": "", "post_date": "08/01/2023 00:41:10", "content": "

Answer is Dataset 1. My single model with dataset 1 without dialate is Private 0.543, Public 0.51. My best private is 0.597 :(( My final submission is combine model train by dataset 1 and model train by dataset 2 -_- \"\"

", "votes": null, "replies": [ { "id": 2367977, "author_name": "quan0095", "author_url": "", "post_date": "08/01/2023 00:42:46", "content": "

My best private and last submission -_-

", "votes": null, "replies": [] }, { "id": 2367978, "author_name": "devchopin", "author_url": "", "post_date": "08/01/2023 00:43:18", "content": "

private 0.597 is amazaing… Did you use MaskDino single Model?

", "votes": null, "replies": [ { "id": 2368115, "author_name": "quan0095", "author_url": "", "post_date": "08/01/2023 02:40:13", "content": "

My mistake is not use only dataset1 for ensemble :(( https://www.kaggle.com/competitions/hubmap-hacking-the-human-vasculature/discussion/428315

", "votes": null, "replies": [] } ] } ] } ], "raw_markdown_by_id": { "2367927": "In my case, the difference between dilate or no dilate was very large.\n(Dilate 555->454, No dilate 541 -> 544)\n\n\nHow were the others?", "2367932": "yes\nFinally I did not select no dilate submissions then....\n\nbest single model\n \nlb556 pb 461 --> dilate\nlb495 pb 550 --> no dilate", "2367935": "yes, also for me, \nprivate LB: no dilate 0.561, with dilate 0.455. Unfortunately, we didn't select it.", "2367937": "i'm doing a sub without dilate, I'll let you know in a couple of hours :D", "2367968": "😢 Thank you for sharing !", "2367969": "😢 Please share your results", "2367975": "Answer is Dataset 1. My single model with dataset 1 without dialate is Private 0.543, Public 0.51. My best private is 0.597 :(( My final submission is combine model train by dataset 1 and model train by dataset 2 -_- ![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F2832287%2Fe1c7ada4ec5f11c3470b459e7e225c29%2FScreenshot%20from%202023-08-01%2007-15-12.png?generation=1690850101951840&alt=media)", "2367977": "My best private and last submission -_-", "2367978": "private 0.597 is amazaing... Did you use MaskDino single Model?", "2368115": "My mistake is not use only dataset1 for ensemble :(( https://www.kaggle.com/competitions/hubmap-hacking-the-human-vasculature/discussion/428315", "2384026": "For me no dilate doesn't help [the score doesn't improve on private LB when we drop dilate], \nwe blame the second dataset, but we're not sure\n\nit's even weirder that our single model scores 0.525 on private, but ensemble scores ~0.39...? something is going on with distributions of masks that neither reflected on public LB or on our CV, but we didn't find whats wrong", "2384318": "check if mask threshold per pixel correct or not.\nin my case, averaging all masks and set \nmask[mask<=0.5]=0 \n\nthen everythnig is ok" }, "source": "meta" }