![](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F12870466%2Fcd4079e4c550c311d7a36b823d1dfd17%2FScreenshot%20from%202024-02-07%2000-59-23.png?generation=1707264535063295&alt=media\")
Thanks to the organisers for the learning opportunities and thanks @shashwatraman for helping me in this comp.
\n
Submission Picking rule:
\nWe played against the shakeup by keeping the CV simple and the one that made the most sense for us: validation on kidney_3 dense. We didn't trust the LB at all in this competition and decided to pick the submission that followed the following criteria:
After that, the training protocol is very simple and there is no post/preprocessing done.
\nTakeaways:
\nMost of the things we tried in the comp didn't work, the things that worked best on CV was the 2.5d and 3d approaches, but I didn't want to trust that the z axis resolution was going to allow such models.
\nI thought about distilling those preds to a 2D model but i didn't have the time for it.
\nI would love to participate in another 3D comp where everything is more stable and where I can use a bit more imagination than this one, where most of the score came from having the discipline of not looking at the LB.
\nEven if I'm proud of our CV, our best submission scored 0.615, and if I really stuck to my rules, I would have picked it, but the LB for it was so low I did coward away from taking it, lesson learned.
Thanks to the organisers for the learning opportunities and thanks @shashwatraman for helping me in this comp.
\n
Submission Picking rule:
\nWe played against the shakeup by keeping the CV simple and the one that made the most sense for us: validation on kidney_3 dense. We didn't trust the LB at all in this competition and decided to pick the submission that followed the following criteria:
After that, the training protocol is very simple and there is no post/preprocessing done.
\nTakeaways:
\nMost of the things we tried in the comp didn't work, the things that worked best on CV was the 2.5d and 3d approaches, but I didn't want to trust that the z axis resolution was going to allow such models.
\nI thought about distilling those preds to a 2D model but i didn't have the time for it.
\nI would love to participate in another 3D comp where everything is more stable and where I can use a bit more imagination than this one, where most of the score came from having the discipline of not looking at the LB.
\nEven if I'm proud of our CV, our best submission scored 0.615, and if I really stuck to my rules, I would have picked it, but the LB for it was so low I did coward away from taking it, lesson learned.
In a similar boat myself, I think I trusted LB a bit too much though. I had my best as 0.61 but hadnt selected it, it was a solo model too which is interesting. Glad to see you had success by playing the long game!
", "rawMarkdown": "In a similar boat myself, I think I trusted LB a bit too much though. I had my best as 0.61 but hadnt selected it, it was a solo model too which is interesting. Glad to see you had success by playing the long game!", "votes": null }, { "id": "2640458", "postDate": "02/07/2024 00:28:25", "content": "congratulations！
", "rawMarkdown": "congratulations！", "votes": null }, { "id": "2640459", "postDate": "02/07/2024 00:29:20", "content": "\"and there is no post/preprocessing done\" I was worried about haven't used cc3d dust cleansing. But seems hasn't been my big mistake. Sadly one of the rejected submissions based on standardization on the volume values of the target subspace of first prediction was a silver medal.
\nWell, congratulations.
I think top 100 team fall in the same category, if we didn't join the comp so late (3weeks before end), we would probably have ended up over fitting due to the pressure of LB :)
", "rawMarkdown": "I think top 100 team fall in the same category, if we didn't join the comp so late (3weeks before end), we would probably have ended up over fitting due to the pressure of LB :)", "votes": null }, { "id": "2640475", "postDate": "02/07/2024 00:33:27", "content": "thanks @pkyangno1 !
", "rawMarkdown": "thanks @pkyangno1 !", "votes": null }, { "id": "2640478", "postDate": "02/07/2024 00:33:48", "content": "Similarly, our best score of 0.621 was not selected either.
", "rawMarkdown": "Similarly, our best score of 0.621 was not selected either.", "votes": null }, { "id": "2640479", "postDate": "02/07/2024 00:36:28", "content": "https://www.kaggle.com/competitions/blood-vessel-segmentation/discussion/473097
\nI think you were aware of the kidney_4 (5/6?) being online, when I saw it, seeing how noisy it was made me reconsider using anything that I couldn't test (on CV) (thrsholding, removing small objects etc). I think I could have taken more risks in this comp but I'm content with a silver
", "rawMarkdown": "https://www.kaggle.com/competitions/blood-vessel-segmentation/discussion/473097\n\nI think you were aware of the kidney_4 (5/6?) being online, when I saw it, seeing how noisy it was made me reconsider using anything that I couldn't test (on CV) (thrsholding, removing small objects etc). I think I could have taken more risks in this comp but I'm content with a silver", "votes": null }, { "id": "2640480", "postDate": "02/07/2024 00:37:16", "content": "I'm guessing that will be the case for a lot of people ! maybe some golds not being selected here and there as well
", "rawMarkdown": "I'm guessing that will be the case for a lot of people ! maybe some golds not being selected here and there as well", "votes": null }, { "id": "2642183", "postDate": "02/08/2024 01:58:37", "content": "Congratulations! @janmpia @shashwatraman
", "rawMarkdown": "Congratulations! @janmpia @shashwatraman", "votes": null } ], "comments": [ { "id": 2640457, "author_name": "cody11null", "author_url": "", "post_date": "02/07/2024 00:28:17", "content": "In a similar boat myself, I think I trusted LB a bit too much though. I had my best as 0.61 but hadnt selected it, it was a solo model too which is interesting. Glad to see you had success by playing the long game!
", "votes": null, "replies": [ { "id": 2640472, "author_name": "janmpia", "author_url": "", "post_date": "02/07/2024 00:33:09", "content": "I think top 100 team fall in the same category, if we didn't join the comp so late (3weeks before end), we would probably have ended up over fitting due to the pressure of LB :)
", "votes": null, "replies": [] } ] }, { "id": 2640458, "author_name": "pkyangno1", "author_url": "", "post_date": "02/07/2024 00:28:25", "content": "congratulations！
", "votes": null, "replies": [ { "id": 2640475, "author_name": "janmpia", "author_url": "", "post_date": "02/07/2024 00:33:27", "content": "thanks @pkyangno1 !
", "votes": null, "replies": [] } ] }, { "id": 2640459, "author_name": "sacuscreed", "author_url": "", "post_date": "02/07/2024 00:29:20", "content": "\"and there is no post/preprocessing done\" I was worried about haven't used cc3d dust cleansing. But seems hasn't been my big mistake. Sadly one of the rejected submissions based on standardization on the volume values of the target subspace of first prediction was a silver medal.
\nWell, congratulations.
https://www.kaggle.com/competitions/blood-vessel-segmentation/discussion/473097
\nI think you were aware of the kidney_4 (5/6?) being online, when I saw it, seeing how noisy it was made me reconsider using anything that I couldn't test (on CV) (thrsholding, removing small objects etc). I think I could have taken more risks in this comp but I'm content with a silver
", "votes": null, "replies": [] } ] }, { "id": 2640478, "author_name": "", "author_url": "", "post_date": "02/07/2024 00:33:48", "content": "Similarly, our best score of 0.621 was not selected either.
", "votes": null, "replies": [ { "id": 2640480, "author_name": "janmpia", "author_url": "", "post_date": "02/07/2024 00:37:16", "content": "I'm guessing that will be the case for a lot of people ! maybe some golds not being selected here and there as well
", "votes": null, "replies": [] } ] }, { "id": 2642183, "author_name": "kforkaran", "author_url": "", "post_date": "02/08/2024 01:58:37", "content": "Congratulations! @janmpia @shashwatraman
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "2640452": "## Our solution\n\nThanks to the organisers for the learning opportunities and thanks @shashwatraman for helping me in this comp.\n\n\n\n**Submission Picking rule:**\nWe played against the shakeup by keeping the CV simple and the one that made the most sense for us: validation on kidney_3 dense. We didn't trust the LB at all in this competition and decided to pick the submission that followed the following criteria:\n- Tiling Model - avoid weird shapes of the private LB\n- Used interpolation augs (resizing and random resize crop) - resolution shift\n- Last epoch weights (no checkpointing) - avoid overconfident CV\n\n\n**After that, the training protocol is very simple and there is no post/preprocessing done.**\n- 2D unet\n- Tiling as mentionned before (384x384)\n- maxvit_small_384 single model (no ensembling) (we didn't have time xd)\n- 15 epochs\n- volume norm\n- light augs \n- Low threshold: 0.1\n\n\n**Takeaways:**\nMost of the things we tried in the comp didn't work, the things that worked best on CV was the 2.5d and 3d approaches, but I didn't want to trust that the z axis resolution was going to allow such models.\nI thought about distilling those preds to a 2D model but i didn't have the time for it.\nI would love to participate in another 3D comp where everything is more stable and where I can use a bit more imagination than this one, where most of the score came from having the discipline of not looking at the LB. \nEven if I'm proud of our CV, our best submission scored 0.615, and if I really stuck to my rules, I would have picked it, but the LB for it was so low I did coward away from taking it, lesson learned.", "2640457": "In a similar boat myself, I think I trusted LB a bit too much though. I had my best as 0.61 but hadnt selected it, it was a solo model too which is interesting. Glad to see you had success by playing the long game!", "2640458": "congratulations！", "2640459": "\"and there is no post/preprocessing done\" I was worried about haven't used cc3d dust cleansing. But seems hasn't been my big mistake. Sadly one of the rejected submissions based on standardization on the volume values of the target subspace of first prediction was a silver medal.\nWell, congratulations.", "2640472": "I think top 100 team fall in the same category, if we didn't join the comp so late (3weeks before end), we would probably have ended up over fitting due to the pressure of LB :)", "2640475": "thanks @pkyangno1 !", "2640478": "Similarly, our best score of 0.621 was not selected either.", "2640479": "https://www.kaggle.com/competitions/blood-vessel-segmentation/discussion/473097\n\nI think you were aware of the kidney_4 (5/6?) being online, when I saw it, seeing how noisy it was made me reconsider using anything that I couldn't test (on CV) (thrsholding, removing small objects etc). I think I could have taken more risks in this comp but I'm content with a silver", "2640480": "I'm guessing that will be the case for a lot of people ! maybe some golds not being selected here and there as well", "2642183": "Congratulations! @janmpia @shashwatraman" }, "source": "meta" }