The initial leaderboard refresh last week did not properly reflect the new labels. I've patched the problem and initiated a fresh rescore, which should take roughly a day to complete. Apologies for the confusion.
\nEdit: the rescore is now complete.
", "messages": [ { "id": 3383356, "postDate": "2025-12-30T00:38:25.613Z", "content": "The initial leaderboard refresh last week did not properly reflect the new labels. I've patched the problem and initiated a fresh rescore, which should take roughly a day to complete. Apologies for the confusion.
\nEdit: the rescore is now complete.
", "rawMarkdown": "The initial leaderboard refresh last week [did not properly reflect the new labels](https://www.kaggle.com/competitions/vesuvius-challenge-surface-detection/discussion/664184#3383353). I've patched the problem and initiated a fresh rescore, which should take roughly a day to complete. Apologies for the confusion.\n\nEdit: the rescore is now complete.", "votes": 19 }, { "id": 3383603, "postDate": "2025-12-30T13:30:02.487Z", "content": "I am not your bro, and no, I do not have 16GB card… @ggayoayogg, can you ship me one if it's so trivial?
", "rawMarkdown": "I am not your bro, and no, I do not have 16GB card... @ggayoayogg, can you ship me one if it's so trivial?", "votes": 2 }, { "id": 3383477, "postDate": "2025-12-30T08:37:52.583Z", "content": "@sohier Only the score in LB will be updated and the dataset that you previously updated will remain same, right? I mean, there will be no further update on the competition dataset, is it correct?
", "rawMarkdown": "@sohier Only the score in LB will be updated and the dataset that you previously updated will remain same, right? I mean, there will be no further update on the competition dataset, is it correct?", "votes": 2, "replies": [ { "id": 3383486, "postDate": "2025-12-30T09:20:26.187Z", "content": "Yess, I don't think public dataset will have any changes, the problem was with the test dataset.
", "rawMarkdown": "Yess, I don't think public dataset will have any changes, the problem was with the test dataset.", "votes": 1 } ] }, { "id": 3383361, "postDate": "2025-12-30T00:42:37.217Z", "content": "Major shake-up
", "rawMarkdown": "Major shake-up", "votes": 2, "replies": [ { "id": 3383454, "postDate": "2025-12-30T07:30:16.037Z", "content": "hmm, It's mainly cause people were working with old dataset. In a few days after people retrain their old pipelines based on newer data, i feel the rankings will return to normal.
", "rawMarkdown": "hmm, It's mainly cause people were working with old dataset. In a few days after people retrain their old pipelines based on newer data, i feel the rankings will return to normal.", "votes": 1, "replies": [ { "id": 3383484, "postDate": "2025-12-30T09:12:53.480Z", "content": "I imagine this will be the case. Our baseline, which is a simple nnUNet with some modified augmentations, reaches approx 0.59 LB on the updated dataset. The baseline LB on the old dataset was a bit lower than the top submissions on the old leaderboard. I am very curious to see how the leaderboard will adjust when the top teams will have updated their training runs :)
", "rawMarkdown": "I imagine this will be the case. Our baseline, which is a simple nnUNet with some modified augmentations, reaches approx 0.59 LB on the updated dataset. The baseline LB on the old dataset was a bit lower than the top submissions on the old leaderboard. I am very curious to see how the leaderboard will adjust when the top teams will have updated their training runs :)", "votes": 6 }, { "id": 3383488, "postDate": "2025-12-30T09:24:00.733Z", "rawMarkdown": "", "votes": 1, "isDeleted": true }, { "id": 3383737, "postDate": "2025-12-30T18:35:10.777Z", "content": "Agreed, I have no idea why so many were using the old data after over 2 weeks of new data discussions
", "rawMarkdown": "Agreed, I have no idea why so many were using the old data after over 2 weeks of new data discussions", "votes": 4 }, { "id": 3383753, "postDate": "2025-12-30T19:08:08.417Z", "content": "maybe because until last night the LB was computed with old data… 🤔
", "rawMarkdown": "maybe because until last night the LB was computed with old data... 🤔", "votes": -2 }, { "id": 3383765, "postDate": "2025-12-30T19:29:06.713Z", "content": "LB being computed with older data was bug, and I don't think when it's garaunteed that final results and eventually the leaderboard will be judged on the newer and the better dataset, then trying to boost temporary results is the wisest choice.
", "rawMarkdown": "LB being computed with older data was bug, and I don't think when it's garaunteed that final results and eventually the leaderboard will be judged on the newer and the better dataset, then trying to boost temporary results is the wisest choice. " }, { "id": 3383767, "postDate": "2025-12-30T19:30:53.553Z", "content": "I think the only mistake on host's side was not properly updating the test data with the train data on the first attempt, but that too has been fixed well in time, all the top teams will reproduce their results on the new dataset within week or two.
", "rawMarkdown": "I think the only mistake on host's side was not properly updating the test data with the train data on the first attempt, but that too has been fixed well in time, all the top teams will reproduce their results on the new dataset within week or two." }, { "id": 3384830, "postDate": "2026-01-02T04:50:18.593Z", "rawMarkdown": "", "isDeleted": true }, { "id": 3384831, "postDate": "2026-01-02T04:50:42.163Z", "content": "@giorgioangelotti Thanks for the info. I’m not sure if our execution is off, but my local results are consistent with the baseline. Did you run any comparative experiments back then?
", "rawMarkdown": "@giorgioangelotti Thanks for the info. I’m not sure if our execution is off, but my local results are consistent with the baseline. Did you run any comparative experiments back then?\n\n" } ] } ] }, { "id": 3383504, "postDate": "2025-12-30T09:56:55.567Z", "content": "Is there a plan to extend the competition? It would be very much appreciated considering there is only 6 weeks left and there was a non-negligible shake-up which highlights that the dataset was significantly modified.
", "rawMarkdown": "Is there a plan to extend the competition? It would be very much appreciated considering there is only 6 weeks left and there was a non-negligible shake-up which highlights that the dataset was significantly modified.", "replies": [ { "id": 3383519, "postDate": "2025-12-30T10:33:16.210Z", "content": "Thanks for the suggestion. At the moment we’re not planning to extend the competition.\nThe recent shake-up is mainly due to the delayed rescore being applied (rather than additional dataset changes). That said, we’re listening to the community: if many participants feel an extension is needed, we can consider a short one and we’ll communicate any decision clearly.
", "rawMarkdown": "Thanks for the suggestion. At the moment we’re not planning to extend the competition.\nThe recent shake-up is mainly due to the delayed rescore being applied (rather than additional dataset changes). That said, we’re listening to the community: if many participants feel an extension is needed, we can consider a short one and we’ll communicate any decision clearly.", "votes": 2 } ] }, { "id": 3383581, "postDate": "2025-12-30T12:01:49.367Z", "content": "maybe he means his LLM credits :D
", "rawMarkdown": "maybe he means his LLM credits :D", "votes": -5, "replies": [ { "id": 3383604, "postDate": "2025-12-30T13:31:45.577Z", "content": "@aaronhance1 nice try, but I still believe that Kaggle community is polite, of course, with some exceptions, right?
", "rawMarkdown": "@aaronhance1 nice try, but I still believe that Kaggle community is polite, of course, with some exceptions, right?", "votes": 2 }, { "id": 3383663, "postDate": "2025-12-30T15:55:19.050Z", "content": "Sorry, sometimes I'm a bit silly. But on a real note, I'm training locally on a 3080 TI 12GB, around £400 on ebay, also the P100 16GB is around £100, a bit slower though, might be worth a look if you don't have a local GPU.
", "rawMarkdown": "Sorry, sometimes I'm a bit silly. But on a real note, I'm training locally on a 3080 TI 12GB, around £400 on ebay, also the P100 16GB is around £100, a bit slower though, might be worth a look if you don't have a local GPU. " }, { "id": 3383679, "postDate": "2025-12-30T16:25:08.533Z", "content": "I'm a 4070 TiS, around $600.
", "rawMarkdown": "I'm a 4070 TiS, around $600." }, { "id": 3383699, "postDate": "2025-12-30T17:07:51.683Z", "content": "@ggayoayogg how much time does it take you to train let's say 100 epochs nnUNet for this competitions dataset??
", "rawMarkdown": "@ggayoayogg how much time does it take you to train let's say 100 epochs nnUNet for this competitions dataset??" }, { "id": 3383719, "postDate": "2025-12-30T17:57:43.550Z", "content": "1 epoch takes approximately 80 seconds and requires about 9 GB of VRAM.
", "rawMarkdown": "1 epoch takes approximately 80 seconds and requires about 9 GB of VRAM.", "votes": 1 }, { "id": 3383902, "postDate": "2025-12-31T04:35:49.300Z", "content": "I believe anyone whose bathroom is full of NVIDIA H200s and likes brute-force solutions must be exceptionally polite.
", "rawMarkdown": "I believe anyone whose bathroom is full of NVIDIA H200s and likes brute-force solutions must be exceptionally polite.", "votes": 1 }, { "id": 3383925, "postDate": "2025-12-31T05:42:58.173Z", "content": "有H200是很不錯 但我連5090都買不起
", "rawMarkdown": "有H200是很不錯 但我連5090都買不起" } ] }, { "id": 3383508, "postDate": "2025-12-30T10:06:57.940Z", "content": "How many pivotal changes shall we expect, for example a week before deadline would be great timing for another shake… Just make me think is it worth to continue in this unpredictable competition? Not sure if rise price justify this actions…
", "rawMarkdown": "How many pivotal changes shall we expect, for example a week before deadline would be great timing for another shake... Just make me think is it worth to continue in this unpredictable competition? Not sure if rise price justify this actions... ", "votes": -4, "replies": [ { "id": 3383514, "postDate": "2025-12-30T10:28:24.613Z", "content": "We understand the rescore timing has been frustrating, especially over the holiday period.
\nTo clarify: we released a single dataset update based on community feedback (fixing label topology issues and making sheet thickness uniform). A rescore was intended at that time, but due to an issue on Kaggle’s side the correct test set was not applied, so the “first rescore” effectively didn’t happen. Kaggle support has now corrected this, and the rescore happening now is simply the delayed application of that same update.
\nAt this point, we do not plan any further dataset changes before the deadline. If a major data issue is discovered, we would consider addressing it, but we would communicate it clearly and as early as possible. Our goal is to keep the competition fair and stable for everyone.
", "rawMarkdown": "We understand the rescore timing has been frustrating, especially over the holiday period.\n\nTo clarify: we released a **single** dataset update based on community feedback (fixing label topology issues and making sheet thickness uniform). A rescore was intended at that time, but due to an issue on Kaggle’s side the correct test set was not applied, so the “first rescore” effectively didn’t happen. Kaggle support has now corrected this, and the rescore happening now is simply the delayed application of that same update.\n\nAt this point, we do not plan any further dataset changes before the deadline. If a major data issue is discovered, we would consider addressing it, but we would communicate it clearly and as early as possible. Our goal is to keep the competition fair and stable for everyone.", "votes": 4 }, { "id": 3383551, "postDate": "2025-12-30T11:05:15.273Z", "content": "But till now only one change has been made which is the dataset change, and I think after 3-4 days when you've retrained your pipeline on new dataset, your score will be surely higher than the last time.
", "rawMarkdown": "But till now only one change has been made which is the dataset change, and I think after 3-4 days when you've retrained your pipeline on new dataset, your score will be surely higher than the last time.", "votes": 2, "replies": [ { "id": 3383559, "postDate": "2025-12-30T11:27:48.630Z", "content": "Could be but I already used most of my resources and credits I have accumulated so this major change pushes me out of the game as many other participants and only \"rich\" will compete for even higher price…
", "rawMarkdown": "Could be but I already used most of my resources and credits I have accumulated so this major change pushes me out of the game as many other participants and only \"rich\" will compete for even higher price... " }, { "id": 3383593, "postDate": "2025-12-30T12:45:54.677Z", "content": "Yes. Especially if you spend time on preprocessing dataset and training.
", "rawMarkdown": "Yes. Especially if you spend time on preprocessing dataset and training." }, { "id": 3383605, "postDate": "2025-12-30T13:32:38.997Z", "content": "@dragonzhang yes, and making it public for anyone to use…
", "rawMarkdown": "@dragonzhang yes, and making it public for anyone to use..." }, { "id": 3383607, "postDate": "2025-12-30T13:34:32.877Z", "content": "Regarding the credits, I utilized Colab Pro's three-month compute credits.
", "rawMarkdown": "Regarding the credits, I utilized Colab Pro's three-month compute credits.", "votes": 1 } ] } ] }, { "id": 3384677, "postDate": "2026-01-01T17:22:21.253Z", "content": "i cant view the updated score in (private) notebooks in the Submissions tab
", "rawMarkdown": "i cant view the updated score in (private) notebooks in the Submissions tab" } ], "comments": [ { "id": 3383603, "author_name": "Jirka", "author_url": "", "post_date": "2025-12-30T13:30:02.487000", "content": "I am not your bro, and no, I do not have 16GB card… @ggayoayogg, can you ship me one if it's so trivial?
", "votes": 2, "replies": [] }, { "id": 3383477, "author_name": "Simon Alerdic", "author_url": "", "post_date": "2025-12-30T08:37:52.583000", "content": "@sohier Only the score in LB will be updated and the dataset that you previously updated will remain same, right? I mean, there will be no further update on the competition dataset, is it correct?
", "votes": 2, "replies": [ { "id": 3383486, "author_name": "Manas Choudhary", "author_url": "", "post_date": "2025-12-30T09:20:26.187000", "content": "Yess, I don't think public dataset will have any changes, the problem was with the test dataset.
", "votes": 1, "replies": [] } ] }, { "id": 3383361, "author_name": "GG Ayo (AyoGG)", "author_url": "", "post_date": "2025-12-30T00:42:37.217000", "content": "Major shake-up
", "votes": 2, "replies": [ { "id": 3383454, "author_name": "Manas Choudhary", "author_url": "", "post_date": "2025-12-30T07:30:16.037000", "content": "hmm, It's mainly cause people were working with old dataset. In a few days after people retrain their old pipelines based on newer data, i feel the rankings will return to normal.
", "votes": 1, "replies": [ { "id": 3383484, "author_name": "Giorgio Angelotti", "author_url": "", "post_date": "2025-12-30T09:12:53.480000", "content": "I imagine this will be the case. Our baseline, which is a simple nnUNet with some modified augmentations, reaches approx 0.59 LB on the updated dataset. The baseline LB on the old dataset was a bit lower than the top submissions on the old leaderboard. I am very curious to see how the leaderboard will adjust when the top teams will have updated their training runs :)
", "votes": 6, "replies": [] }, { "id": 3383488, "author_name": "", "author_url": "", "post_date": "2025-12-30T09:24:00.733000", "content": "", "votes": 1, "replies": [] }, { "id": 3383737, "author_name": "Cody_Null", "author_url": "", "post_date": "2025-12-30T18:35:10.777000", "content": "Agreed, I have no idea why so many were using the old data after over 2 weeks of new data discussions
", "votes": 4, "replies": [] }, { "id": 3383753, "author_name": "Jirka", "author_url": "", "post_date": "2025-12-30T19:08:08.417000", "content": "maybe because until last night the LB was computed with old data… 🤔
", "votes": -2, "replies": [] }, { "id": 3383765, "author_name": "Manas Choudhary", "author_url": "", "post_date": "2025-12-30T19:29:06.713000", "content": "LB being computed with older data was bug, and I don't think when it's garaunteed that final results and eventually the leaderboard will be judged on the newer and the better dataset, then trying to boost temporary results is the wisest choice.
", "votes": 0, "replies": [] }, { "id": 3383767, "author_name": "Manas Choudhary", "author_url": "", "post_date": "2025-12-30T19:30:53.553000", "content": "I think the only mistake on host's side was not properly updating the test data with the train data on the first attempt, but that too has been fixed well in time, all the top teams will reproduce their results on the new dataset within week or two.
", "votes": 0, "replies": [] }, { "id": 3384830, "author_name": "", "author_url": "", "post_date": "2026-01-02T04:50:18.593000", "content": "", "votes": 0, "replies": [] }, { "id": 3384831, "author_name": "tingyi", "author_url": "", "post_date": "2026-01-02T04:50:42.163000", "content": "@giorgioangelotti Thanks for the info. I’m not sure if our execution is off, but my local results are consistent with the baseline. Did you run any comparative experiments back then?
", "votes": 0, "replies": [] } ] } ] }, { "id": 3383504, "author_name": "Francois Lemarchand", "author_url": "", "post_date": "2025-12-30T09:56:55.567000", "content": "Is there a plan to extend the competition? It would be very much appreciated considering there is only 6 weeks left and there was a non-negligible shake-up which highlights that the dataset was significantly modified.
", "votes": 0, "replies": [ { "id": 3383519, "author_name": "Giorgio Angelotti", "author_url": "", "post_date": "2025-12-30T10:33:16.210000", "content": "Thanks for the suggestion. At the moment we’re not planning to extend the competition.\nThe recent shake-up is mainly due to the delayed rescore being applied (rather than additional dataset changes). That said, we’re listening to the community: if many participants feel an extension is needed, we can consider a short one and we’ll communicate any decision clearly.
", "votes": 2, "replies": [] } ] }, { "id": 3383581, "author_name": "Aaron Hance", "author_url": "", "post_date": "2025-12-30T12:01:49.367000", "content": "maybe he means his LLM credits :D
", "votes": -5, "replies": [ { "id": 3383604, "author_name": "Jirka", "author_url": "", "post_date": "2025-12-30T13:31:45.577000", "content": "@aaronhance1 nice try, but I still believe that Kaggle community is polite, of course, with some exceptions, right?
", "votes": 2, "replies": [] }, { "id": 3383663, "author_name": "Aaron Hance", "author_url": "", "post_date": "2025-12-30T15:55:19.050000", "content": "Sorry, sometimes I'm a bit silly. But on a real note, I'm training locally on a 3080 TI 12GB, around £400 on ebay, also the P100 16GB is around £100, a bit slower though, might be worth a look if you don't have a local GPU.
", "votes": 0, "replies": [] }, { "id": 3383679, "author_name": "GG Ayo (AyoGG)", "author_url": "", "post_date": "2025-12-30T16:25:08.533000", "content": "I'm a 4070 TiS, around $600.
", "votes": 0, "replies": [] }, { "id": 3383699, "author_name": "Manas Choudhary", "author_url": "", "post_date": "2025-12-30T17:07:51.683000", "content": "@ggayoayogg how much time does it take you to train let's say 100 epochs nnUNet for this competitions dataset??
", "votes": 0, "replies": [] }, { "id": 3383719, "author_name": "GG Ayo (AyoGG)", "author_url": "", "post_date": "2025-12-30T17:57:43.550000", "content": "1 epoch takes approximately 80 seconds and requires about 9 GB of VRAM.
", "votes": 1, "replies": [] }, { "id": 3383902, "author_name": "Tom", "author_url": "", "post_date": "2025-12-31T04:35:49.300000", "content": "I believe anyone whose bathroom is full of NVIDIA H200s and likes brute-force solutions must be exceptionally polite.
", "votes": 1, "replies": [] }, { "id": 3383925, "author_name": "GG Ayo (AyoGG)", "author_url": "", "post_date": "2025-12-31T05:42:58.173000", "content": "有H200是很不錯 但我連5090都買不起
", "votes": 0, "replies": [] } ] }, { "id": 3383508, "author_name": "Jirka", "author_url": "", "post_date": "2025-12-30T10:06:57.940000", "content": "How many pivotal changes shall we expect, for example a week before deadline would be great timing for another shake… Just make me think is it worth to continue in this unpredictable competition? Not sure if rise price justify this actions…
", "votes": -4, "replies": [ { "id": 3383514, "author_name": "Giorgio Angelotti", "author_url": "", "post_date": "2025-12-30T10:28:24.613000", "content": "We understand the rescore timing has been frustrating, especially over the holiday period.
\nTo clarify: we released a single dataset update based on community feedback (fixing label topology issues and making sheet thickness uniform). A rescore was intended at that time, but due to an issue on Kaggle’s side the correct test set was not applied, so the “first rescore” effectively didn’t happen. Kaggle support has now corrected this, and the rescore happening now is simply the delayed application of that same update.
\nAt this point, we do not plan any further dataset changes before the deadline. If a major data issue is discovered, we would consider addressing it, but we would communicate it clearly and as early as possible. Our goal is to keep the competition fair and stable for everyone.
", "votes": 4, "replies": [] }, { "id": 3383551, "author_name": "Manas Choudhary", "author_url": "", "post_date": "2025-12-30T11:05:15.273000", "content": "But till now only one change has been made which is the dataset change, and I think after 3-4 days when you've retrained your pipeline on new dataset, your score will be surely higher than the last time.
", "votes": 2, "replies": [ { "id": 3383559, "author_name": "Jirka", "author_url": "", "post_date": "2025-12-30T11:27:48.630000", "content": "Could be but I already used most of my resources and credits I have accumulated so this major change pushes me out of the game as many other participants and only \"rich\" will compete for even higher price…
", "votes": 0, "replies": [] }, { "id": 3383593, "author_name": "dragon zhang", "author_url": "", "post_date": "2025-12-30T12:45:54.677000", "content": "Yes. Especially if you spend time on preprocessing dataset and training.
", "votes": 0, "replies": [] }, { "id": 3383605, "author_name": "Jirka", "author_url": "", "post_date": "2025-12-30T13:32:38.997000", "content": "@dragonzhang yes, and making it public for anyone to use…
", "votes": 0, "replies": [] }, { "id": 3383607, "author_name": "Jirka", "author_url": "", "post_date": "2025-12-30T13:34:32.877000", "content": "Regarding the credits, I utilized Colab Pro's three-month compute credits.
", "votes": 1, "replies": [] } ] } ] }, { "id": 3384677, "author_name": "parthenos", "author_url": "", "post_date": "2026-01-01T17:22:21.253000", "content": "i cant view the updated score in (private) notebooks in the Submissions tab
", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "3383356": "The initial leaderboard refresh last week [did not properly reflect the new labels](https://www.kaggle.com/competitions/vesuvius-challenge-surface-detection/discussion/664184#3383353). I've patched the problem and initiated a fresh rescore, which should take roughly a day to complete. Apologies for the confusion.\n\nEdit: the rescore is now complete.", "3383603": "I am not your bro, and no, I do not have 16GB card... @ggayoayogg, can you ship me one if it's so trivial?", "3383477": "@sohier Only the score in LB will be updated and the dataset that you previously updated will remain same, right? I mean, there will be no further update on the competition dataset, is it correct?", "3383361": "Major shake-up", "3383504": "Is there a plan to extend the competition? It would be very much appreciated considering there is only 6 weeks left and there was a non-negligible shake-up which highlights that the dataset was significantly modified.", "3383581": "maybe he means his LLM credits :D", "3383508": "How many pivotal changes shall we expect, for example a week before deadline would be great timing for another shake... Just make me think is it worth to continue in this unpredictable competition? Not sure if rise price justify this actions... ", "3384677": "i cant view the updated score in (private) notebooks in the Submissions tab" } }