This moment seems to be magical, a 4th place finish. In my first two competitions(Trends and SIIM Melanoma), I ended with top 12% and 8% rank on final LB. I did not win big in those competitions but learnt a lot.
\nIn this competition, I believe that the tissue segmentation and other features extracted from CT scans made the difference. So what worked for us?
\nTissue segmentation can be found here
\nFeature Extraction and the entire data processing and inference pipeline can be found here
It would be totally unfair if I dont thank the Kaggle community for selfless sharing of ideas and approaches on discussion forums and in the form of notebooks. Our solution is built on the amazing work done by other kagglers in this competition.
\nI'd specially like to thank:
\nThanks Everyone!
", "messages": [ { "id": "1040155", "postDate": "10/07/2020 02:13:31", "content": "This moment seems to be magical, a 4th place finish. In my first two competitions(Trends and SIIM Melanoma), I ended with top 12% and 8% rank on final LB. I did not win big in those competitions but learnt a lot.
\nIn this competition, I believe that the tissue segmentation and other features extracted from CT scans made the difference. So what worked for us?
\nTissue segmentation can be found here
\nFeature Extraction and the entire data processing and inference pipeline can be found here
It would be totally unfair if I dont thank the Kaggle community for selfless sharing of ideas and approaches on discussion forums and in the form of notebooks. Our solution is built on the amazing work done by other kagglers in this competition.
\nI'd specially like to thank:
\nThanks Everyone!
", "rawMarkdown": "This moment seems to be magical, a 4th place finish. In my first two competitions(Trends and SIIM Melanoma), I ended with top 12% and 8% rank on final LB. I did not win big in those competitions but learnt a lot. \n\nIn this competition, I believe that the tissue segmentation and other features extracted from CT scans made the difference. So what worked for us?\n- Robust CV\n- Good quality lung and tissue segmentation\n- Rich features extracted from CT scans\n- Ensemble of 10 models\n- A lot of courage to not get baffeled by LB scores\n\nTissue segmentation can be found [here](https://www.kaggle.com/abhishekgbhat/tissue-segmentation-used-in-4th-place-solution)\nFeature Extraction and the entire data processing and inference pipeline can be found [here](https://www.kaggle.com/abhishekgbhat/quantreg-linear-decay-efficientnet-b1-su)\n\n\nIt would be totally unfair if I dont thank the Kaggle community for selfless sharing of ideas and approaches on discussion forums and in the form of notebooks. Our solution is built on the amazing work done by other kagglers in this competition.\n\nI'd specially like to thank:\n- @andradaolteanu for her lung segmentaion notebook. This served as a base for our lung segmentation. With a few tweaks we could improve the lung segmentation and further add tissue segmentation as well.\n- @miklgr500 for sharing Linear Decay notebook\n- @ulrich07 for the Quant Reg notebook\n\nThanks Everyone!", "votes": null }, { "id": "1040161", "postDate": "10/07/2020 02:18:44", "content": "Congratulations @abhishekgbhat! Waiting for your solution, especially waiting for the work done on the tissue segmentation!
", "rawMarkdown": "Congratulations @abhishekgbhat! Waiting for your solution, especially waiting for the work done on the tissue segmentation!", "votes": null }, { "id": "1040206", "postDate": "10/07/2020 03:01:49", "content": "@aadhavvignesh I have updated my post with the solution.
", "rawMarkdown": "aadhavvignesh I have updated my post with the solution.", "votes": null }, { "id": "1040432", "postDate": "10/07/2020 06:15:51", "content": "Congratulations Abhishek on the win , Looking forward to your solution
", "rawMarkdown": "Congratulations Abhishek on the win , Looking forward to your solution", "votes": null }, { "id": "1042086", "postDate": "10/08/2020 03:49:59", "content": "Hey @tanulsingh077, I have updated my post to include solution notebooks. Hope you find it helpful.
", "rawMarkdown": "Hey @tanulsingh077, I have updated my post to include solution notebooks. Hope you find it helpful.", "votes": null } ], "comments": [ { "id": 1040161, "author_name": "aadhavvignesh", "author_url": "", "post_date": "10/07/2020 02:18:44", "content": "Congratulations @abhishekgbhat! Waiting for your solution, especially waiting for the work done on the tissue segmentation!
", "votes": null, "replies": [ { "id": 1040206, "author_name": "abhishekgbhat", "author_url": "", "post_date": "10/07/2020 03:01:49", "content": "@aadhavvignesh I have updated my post with the solution.
", "votes": null, "replies": [] } ] }, { "id": 1040432, "author_name": "tanulsingh077", "author_url": "", "post_date": "10/07/2020 06:15:51", "content": "Congratulations Abhishek on the win , Looking forward to your solution
", "votes": null, "replies": [ { "id": 1042086, "author_name": "abhishekgbhat", "author_url": "", "post_date": "10/08/2020 03:49:59", "content": "Hey @tanulsingh077, I have updated my post to include solution notebooks. Hope you find it helpful.
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "1040155": "This moment seems to be magical, a 4th place finish. In my first two competitions(Trends and SIIM Melanoma), I ended with top 12% and 8% rank on final LB. I did not win big in those competitions but learnt a lot. \n\nIn this competition, I believe that the tissue segmentation and other features extracted from CT scans made the difference. So what worked for us?\n- Robust CV\n- Good quality lung and tissue segmentation\n- Rich features extracted from CT scans\n- Ensemble of 10 models\n- A lot of courage to not get baffeled by LB scores\n\nTissue segmentation can be found [here](https://www.kaggle.com/abhishekgbhat/tissue-segmentation-used-in-4th-place-solution)\nFeature Extraction and the entire data processing and inference pipeline can be found [here](https://www.kaggle.com/abhishekgbhat/quantreg-linear-decay-efficientnet-b1-su)\n\n\nIt would be totally unfair if I dont thank the Kaggle community for selfless sharing of ideas and approaches on discussion forums and in the form of notebooks. Our solution is built on the amazing work done by other kagglers in this competition.\n\nI'd specially like to thank:\n- @andradaolteanu for her lung segmentaion notebook. This served as a base for our lung segmentation. With a few tweaks we could improve the lung segmentation and further add tissue segmentation as well.\n- @miklgr500 for sharing Linear Decay notebook\n- @ulrich07 for the Quant Reg notebook\n\nThanks Everyone!", "1040161": "Congratulations @abhishekgbhat! Waiting for your solution, especially waiting for the work done on the tissue segmentation!", "1040206": "aadhavvignesh I have updated my post with the solution.", "1040432": "Congratulations Abhishek on the win , Looking forward to your solution", "1042086": "Hey @tanulsingh077, I have updated my post to include solution notebooks. Hope you find it helpful." }, "source": "meta" }