{ "id": 286368, "title": "[Up-to-date Summary] We passed 30% of the competition, These are the stories up to this point!", "url": "/competitions/sartorius-cell-instance-segmentation/discussion/286368", "author_name": "", "post_date": "2021-11-08T20:46:48.820000", "votes": 154, "comment_count": 11, "views": 0, "content": "

Ladies and Gentlemen may I have your attention, please
\nWe are right about to pass the 1/3 mark of the competition timeline.
\nA Good time to look back and summarize what we had up to this point.

\n
\n

Starting from the beginning,

\n\n

As of this moment, we are still at the early stages of the competition and we should expect a much more intense discussions going forward.

\n

Those were the top stories for this hour.
\nI’m your host Yam, Reporting live once again from the TPU waiting queue 😭.
\nHave a great day.

", "messages": [ { "id": 1576054, "postDate": "2021-11-08T20:46:48.820Z", "content": "

Ladies and Gentlemen may I have your attention, please
\nWe are right about to pass the 1/3 mark of the competition timeline.
\nA Good time to look back and summarize what we had up to this point.

\n
\n

Starting from the beginning,

\n\n

As of this moment, we are still at the early stages of the competition and we should expect a much more intense discussions going forward.

\n

Those were the top stories for this hour.
\nI’m your host Yam, Reporting live once again from the TPU waiting queue 😭.
\nHave a great day.

", "rawMarkdown": "Ladies and Gentlemen may I have your attention, please\nWe are right about to pass the 1/3 mark of the competition timeline. \nA Good time to look back and summarize what we had up to this point.\n\n____\n\nStarting from the beginning, \n\n- Shortly after the kickoff we got Yaroslav Isaienkov [sharing](https://www.kaggle.com/c/sartorius-cell-instance-segmentation/discussion/278801) that some of the masks are broken!\n\n- During this time, M.Innat [detailed](https://www.kaggle.com/c/sartorius-cell-instance-segmentation/discussion/278883) a plan for action consists of some promising models for this competition, such as: [SOLOv2: Dynamic and Fast Instance Segmentation](https://arxiv.org/pdf/2003.10152.pdf), [ CenterMask : Real-Time Anchor-Free Instance Segmentation](https://arxiv.org/pdf/1911.06667.pdf), [Mask R-CNN for Object Detection and Segmentation](https://arxiv.org/pdf/1703.06870.pdf), [You Only Look At CoefficienTs](https://arxiv.org/pdf/1912.06218.pdf),\n[PolarMask: Single Shot Instance Segmentation with Polar Representation](https://arxiv.org/pdf/1909.13226.pdf). \nAlso includes a [survey of methods.](https://arxiv.org/pdf/2001.05566.pdf)\n\n- Couple of days later Slawek Biel [9th!] [released](https://www.kaggle.com/c/sartorius-cell-instance-segmentation/discussion/280137) an incredible notebook tutorial series for decetron 2, [here](https://www.kaggle.com/slawekbiel/positive-score-with-detectron-1-3-input-data), [here](https://www.kaggle.com/slawekbiel/positive-score-with-detectron-2-3-training) and [here](https://www.kaggle.com/slawekbiel/positive-score-with-detectron-3-3-inference). Couple of days layer we also got a [method](https://www.kaggle.com/c/sartorius-cell-instance-segmentation/discussion/282788) for showing the validation score in decatron.\n\n\n- And by the same time Ishan Dutta (who had just become a Notebooks Master! 🎉 Congrats!) releasing [his great EDA notebook](https://www.kaggle.com/ishandutta/sartorius-indepth-eda-explanation-model/) that makes the whole problem seems easy. \n\n- The following week we got Inoichan [17th!] releasing a [great post](https://www.kaggle.com/c/sartorius-cell-instance-segmentation/discussion/279790) about general tips to take in consideration from the beginning of the competition. \n\n- And at the same time Awsaf released some [code](https://www.kaggle.com/c/sartorius-cell-instance-segmentation/discussion/279995) to deal with the competition rule saying that each predicted mask must not be allowed to overlap the other mask in the same image. While also [releasing](https://www.kaggle.com/c/sartorius-cell-instance-segmentation/discussion/279996) a great Mask-RCNN starter pack you should visit if you didn't already. \n\n- On the topic of the broken annotations we had some Theo Viel [showing](https://www.kaggle.com/c/sartorius-cell-instance-segmentation/discussion/281205) us all that because the cells are smalls and labels are not pixel perfect, the metric fluctuates a lot. And so it makes no sense to evaluate the IoU at high thresholds for small cells.\n\n- And just when you thought you have seen it all Gunes Evitan [visualized every training image so you don't have to](https://www.kaggle.com/c/sartorius-cell-instance-segmentation/discussion/284057). \n\n- This is while Slawek Biel shared some [post-processing /threshold-optimization](https://www.kaggle.com/c/sartorius-cell-instance-segmentation/discussion/284215) experiment that gave him a small LB score boost. \n\n- While also summarizing the [LIVECell paper](https://www.kaggle.com/c/sartorius-cell-instance-segmentation/discussion/285384).\n\n- Getting to last week, Heng joined the party! [starting](https://www.kaggle.com/c/sartorius-cell-instance-segmentation/discussion/285516) the usual research diary discussion post that is already full of important details. \n\n- While Gunes Evitan [shared](https://www.kaggle.com/c/sartorius-cell-instance-segmentation/discussion/285546) a way to split folds for cross-validation, detailing that the data should be split into folds on image level, not annotation level.\n\n- Closing in with [a post](https://www.kaggle.com/c/sartorius-cell-instance-segmentation/discussion/285623) from Slawek Biel about getting per pixel scores from the detectron.\n\n\nAs of this moment, we are still at the early stages of the competition and we should expect a much more intense discussions going forward.\n\nThose were the top stories for this hour.\nI’m your host Yam, Reporting live once again from the TPU waiting queue 😭.\nHave a great day.\n\n\n\n\n\n\n\n", "votes": 152 }, { "id": 1616508, "postDate": "2021-12-13T13:13:29.540Z", "content": "

Great Work. Keep going.

", "rawMarkdown": "Great Work. Keep going." }, { "id": 1616491, "postDate": "2021-12-13T12:56:19.980Z", "content": "

Good work keep it up stay blessed thanks dear.

", "rawMarkdown": "Good work keep it up stay blessed thanks dear." }, { "id": 1584421, "postDate": "2021-11-16T14:16:30.117Z", "content": "

Thanks for the awesome summary. Appreciate it!

", "rawMarkdown": "Thanks for the awesome summary. Appreciate it!" }, { "id": 1582539, "postDate": "2021-11-15T02:02:27.417Z", "content": "

Thanks for great summary! This is good point to start 👍

", "rawMarkdown": "Thanks for great summary! This is good point to start 👍" }, { "id": 1576475, "postDate": "2021-11-09T08:02:08.513Z", "content": "

Very Comprehensive and meaningful summery! I will learn it , thanks to excellent works!

", "rawMarkdown": "Very Comprehensive and meaningful summery! I will learn it , thanks to excellent works!" }, { "id": 1576472, "postDate": "2021-11-09T07:58:00.183Z", "content": "

Niu Beeeeeeee! In Chinese word, what an excellent summary!👍

", "rawMarkdown": "Niu Beeeeeeee! In Chinese word, what an excellent summary!👍" }, { "id": 1576195, "postDate": "2021-11-09T01:26:48.997Z", "content": "
\n

I’m your host Yam, Reporting live once again from the TPU waiting queue 😭

\n
\n

Thanks for a good laugh early in the morning 😂😂😂
\nalso for the excellent summary !

", "rawMarkdown": "> I’m your host Yam, Reporting live once again from the TPU waiting queue 😭\n\nThanks for a good laugh early in the morning 😂😂😂\nalso for the excellent summary !" }, { "id": 1576175, "postDate": "2021-11-09T00:54:40.203Z", "content": "

nice summary. TPU is hard to get now. I suggest that you use GPU asap.

\n

below is worth reading:
\nhttps://www.kaggle.com/c/sartorius-cell-instance-segmentation/discussion/286058

", "rawMarkdown": "nice summary. TPU is hard to get now. I suggest that you use GPU asap.\n\nbelow is worth reading:\nhttps://www.kaggle.com/c/sartorius-cell-instance-segmentation/discussion/286058" }, { "id": 1591534, "postDate": "2021-11-22T12:46:25.040Z", "rawMarkdown": "", "isDeleted": true }, { "id": 1632842, "postDate": "2021-12-30T06:30:57.593Z", "content": "

Great work! Thanks for sharing.

", "rawMarkdown": "Great work! Thanks for sharing." }, { "id": 1577736, "postDate": "2021-11-10T11:04:14.420Z", "content": "

Thank you Yam, excellent work!

", "rawMarkdown": "Thank you Yam, excellent work!\n\n" } ], "comments": [ { "id": 1616508, "author_name": "Moaz Mohammed Elesawey", "author_url": "", "post_date": "2021-12-13T13:13:29.540000", "content": "

Great Work. Keep going.

", "votes": 0, "replies": [] }, { "id": 1616491, "author_name": "waseem sajjad", "author_url": "", "post_date": "2021-12-13T12:56:19.980000", "content": "

Good work keep it up stay blessed thanks dear.

", "votes": 0, "replies": [] }, { "id": 1584421, "author_name": "AshwinVP", "author_url": "", "post_date": "2021-11-16T14:16:30.117000", "content": "

Thanks for the awesome summary. Appreciate it!

", "votes": 0, "replies": [] }, { "id": 1582539, "author_name": "Wonho Song", "author_url": "", "post_date": "2021-11-15T02:02:27.417000", "content": "

Thanks for great summary! This is good point to start 👍

", "votes": 0, "replies": [] }, { "id": 1576475, "author_name": "Ctrl_CV", "author_url": "", "post_date": "2021-11-09T08:02:08.513000", "content": "

Very Comprehensive and meaningful summery! I will learn it , thanks to excellent works!

", "votes": 0, "replies": [] }, { "id": 1576472, "author_name": "ForcewithMe", "author_url": "", "post_date": "2021-11-09T07:58:00.183000", "content": "

Niu Beeeeeeee! In Chinese word, what an excellent summary!👍

", "votes": 0, "replies": [] }, { "id": 1576195, "author_name": "yukiya", "author_url": "", "post_date": "2021-11-09T01:26:48.997000", "content": "
\n

I’m your host Yam, Reporting live once again from the TPU waiting queue 😭

\n
\n

Thanks for a good laugh early in the morning 😂😂😂
\nalso for the excellent summary !

", "votes": 0, "replies": [] }, { "id": 1576175, "author_name": "dragon zhang", "author_url": "", "post_date": "2021-11-09T00:54:40.203000", "content": "

nice summary. TPU is hard to get now. I suggest that you use GPU asap.

\n

below is worth reading:
\nhttps://www.kaggle.com/c/sartorius-cell-instance-segmentation/discussion/286058

", "votes": 0, "replies": [] }, { "id": 1591534, "author_name": "", "author_url": "", "post_date": "2021-11-22T12:46:25.040000", "content": "", "votes": 0, "replies": [] }, { "id": 1632842, "author_name": "chiba1sonny", "author_url": "", "post_date": "2021-12-30T06:30:57.593000", "content": "

Great work! Thanks for sharing.

", "votes": 0, "replies": [] }, { "id": 1577736, "author_name": "NotFound404", "author_url": "", "post_date": "2021-11-10T11:04:14.420000", "content": "

Thank you Yam, excellent work!

", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "1576054": "Ladies and Gentlemen may I have your attention, please\nWe are right about to pass the 1/3 mark of the competition timeline. \nA Good time to look back and summarize what we had up to this point.\n\n____\n\nStarting from the beginning, \n\n- Shortly after the kickoff we got Yaroslav Isaienkov [sharing](https://www.kaggle.com/c/sartorius-cell-instance-segmentation/discussion/278801) that some of the masks are broken!\n\n- During this time, M.Innat [detailed](https://www.kaggle.com/c/sartorius-cell-instance-segmentation/discussion/278883) a plan for action consists of some promising models for this competition, such as: [SOLOv2: Dynamic and Fast Instance Segmentation](https://arxiv.org/pdf/2003.10152.pdf), [ CenterMask : Real-Time Anchor-Free Instance Segmentation](https://arxiv.org/pdf/1911.06667.pdf), [Mask R-CNN for Object Detection and Segmentation](https://arxiv.org/pdf/1703.06870.pdf), [You Only Look At CoefficienTs](https://arxiv.org/pdf/1912.06218.pdf),\n[PolarMask: Single Shot Instance Segmentation with Polar Representation](https://arxiv.org/pdf/1909.13226.pdf). \nAlso includes a [survey of methods.](https://arxiv.org/pdf/2001.05566.pdf)\n\n- Couple of days later Slawek Biel [9th!] [released](https://www.kaggle.com/c/sartorius-cell-instance-segmentation/discussion/280137) an incredible notebook tutorial series for decetron 2, [here](https://www.kaggle.com/slawekbiel/positive-score-with-detectron-1-3-input-data), [here](https://www.kaggle.com/slawekbiel/positive-score-with-detectron-2-3-training) and [here](https://www.kaggle.com/slawekbiel/positive-score-with-detectron-3-3-inference). Couple of days layer we also got a [method](https://www.kaggle.com/c/sartorius-cell-instance-segmentation/discussion/282788) for showing the validation score in decatron.\n\n\n- And by the same time Ishan Dutta (who had just become a Notebooks Master! 🎉 Congrats!) releasing [his great EDA notebook](https://www.kaggle.com/ishandutta/sartorius-indepth-eda-explanation-model/) that makes the whole problem seems easy. \n\n- The following week we got Inoichan [17th!] releasing a [great post](https://www.kaggle.com/c/sartorius-cell-instance-segmentation/discussion/279790) about general tips to take in consideration from the beginning of the competition. \n\n- And at the same time Awsaf released some [code](https://www.kaggle.com/c/sartorius-cell-instance-segmentation/discussion/279995) to deal with the competition rule saying that each predicted mask must not be allowed to overlap the other mask in the same image. While also [releasing](https://www.kaggle.com/c/sartorius-cell-instance-segmentation/discussion/279996) a great Mask-RCNN starter pack you should visit if you didn't already. \n\n- On the topic of the broken annotations we had some Theo Viel [showing](https://www.kaggle.com/c/sartorius-cell-instance-segmentation/discussion/281205) us all that because the cells are smalls and labels are not pixel perfect, the metric fluctuates a lot. And so it makes no sense to evaluate the IoU at high thresholds for small cells.\n\n- And just when you thought you have seen it all Gunes Evitan [visualized every training image so you don't have to](https://www.kaggle.com/c/sartorius-cell-instance-segmentation/discussion/284057). \n\n- This is while Slawek Biel shared some [post-processing /threshold-optimization](https://www.kaggle.com/c/sartorius-cell-instance-segmentation/discussion/284215) experiment that gave him a small LB score boost. \n\n- While also summarizing the [LIVECell paper](https://www.kaggle.com/c/sartorius-cell-instance-segmentation/discussion/285384).\n\n- Getting to last week, Heng joined the party! [starting](https://www.kaggle.com/c/sartorius-cell-instance-segmentation/discussion/285516) the usual research diary discussion post that is already full of important details. \n\n- While Gunes Evitan [shared](https://www.kaggle.com/c/sartorius-cell-instance-segmentation/discussion/285546) a way to split folds for cross-validation, detailing that the data should be split into folds on image level, not annotation level.\n\n- Closing in with [a post](https://www.kaggle.com/c/sartorius-cell-instance-segmentation/discussion/285623) from Slawek Biel about getting per pixel scores from the detectron.\n\n\nAs of this moment, we are still at the early stages of the competition and we should expect a much more intense discussions going forward.\n\nThose were the top stories for this hour.\nI’m your host Yam, Reporting live once again from the TPU waiting queue 😭.\nHave a great day.\n\n\n\n\n\n\n\n", "1616508": "Great Work. Keep going.", "1616491": "Good work keep it up stay blessed thanks dear.", "1584421": "Thanks for the awesome summary. Appreciate it!", "1582539": "Thanks for great summary! This is good point to start 👍", "1576475": "Very Comprehensive and meaningful summery! I will learn it , thanks to excellent works!", "1576472": "Niu Beeeeeeee! In Chinese word, what an excellent summary!👍", "1576195": "> I’m your host Yam, Reporting live once again from the TPU waiting queue 😭\n\nThanks for a good laugh early in the morning 😂😂😂\nalso for the excellent summary !", "1576175": "nice summary. TPU is hard to get now. I suggest that you use GPU asap.\n\nbelow is worth reading:\nhttps://www.kaggle.com/c/sartorius-cell-instance-segmentation/discussion/286058", "1591534": "", "1632842": "Great work! Thanks for sharing.", "1577736": "Thank you Yam, excellent work!\n\n" } }