{ "id": 509696, "title": "Strange Behavior in IMC?", "url": "/competitions/image-matching-challenge-2024/discussion/509696", "author_name": "Cody_Null", "post_date": "2024-06-03T13:54:15.687000", "votes": 35, "comment_count": 44, "views": 0, "content": "

I have faith in the Kaggle staff to prevent any wrong doing that may or may not be going on here but I do see something strange I wanted to point out. In the last few weeks we always expect a pickup in competition but this has been special. Over the last 3 weeks we went from 12 people having a score of 0.16 to now approaching 100. And the number of 0.17 has gone from 10 to 43 in a week as well. What I hadn't noticed until recently is they are a large percentage of people in this group who have created their account during the competition. This competition is very challenging so I certainly did not expect so many to be able to make such a push in the endgame. I do hope that the incredible staff at Kaggle does not have their hands too full with this in their no doubt already busy day.

", "messages": [ { "id": 2852873, "postDate": "2024-06-03T13:54:15.687Z", "content": "

I have faith in the Kaggle staff to prevent any wrong doing that may or may not be going on here but I do see something strange I wanted to point out. In the last few weeks we always expect a pickup in competition but this has been special. Over the last 3 weeks we went from 12 people having a score of 0.16 to now approaching 100. And the number of 0.17 has gone from 10 to 43 in a week as well. What I hadn't noticed until recently is they are a large percentage of people in this group who have created their account during the competition. This competition is very challenging so I certainly did not expect so many to be able to make such a push in the endgame. I do hope that the incredible staff at Kaggle does not have their hands too full with this in their no doubt already busy day.

", "rawMarkdown": "I have faith in the Kaggle staff to prevent any wrong doing that may or may not be going on here but I do see something strange I wanted to point out. In the last few weeks we always expect a pickup in competition but this has been special. Over the last 3 weeks we went from 12 people having a score of 0.16 to now approaching 100. And the number of 0.17 has gone from 10 to 43 in a week as well. What I hadn't noticed until recently is they are a large percentage of people in this group who have created their account during the competition. This competition is very challenging so I certainly did not expect so many to be able to make such a push in the endgame. I do hope that the incredible staff at Kaggle does not have their hands too full with this in their no doubt already busy day. ", "votes": 35 }, { "id": 2854343, "postDate": "2024-06-04T09:05:10.937Z", "content": "

Unfortunately, medals were being sold in this competition.
\nI hope the organizers do something about it.

", "rawMarkdown": "Unfortunately, medals were being sold in this competition. \nI hope the organizers do something about it. ![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F3350264%2F32d51d9c38be0998964400a3f1bd6b74%2Fimg_0658.jpg?generation=1717491904290772&alt=media)", "votes": 17, "replies": [ { "id": 2854359, "postDate": "2024-06-04T09:18:13.170Z", "content": "

@oldufo @addisonhoward @ashleychow @wcukierski see above, time checking the top 100 codes extra and broadly.

", "rawMarkdown": "@oldufo @addisonhoward @ashleychow @wcukierski see above, time checking the top 100 codes extra and broadly.", "votes": 8 }, { "id": 2854420, "postDate": "2024-06-04T10:04:50.107Z", "rawMarkdown": "", "isDeleted": true }, { "id": 2854456, "postDate": "2024-06-04T10:26:29.740Z", "content": "

Most posts on Xiaohongshu (the Chinese social media platform shown above) are full of lies: they can't guarantee medals nor teach you basic skills. At the same time, there also well-established institution out there that have GMs mentor and sell medals (I know some of them personally), and I can attest that they know pretty well how to circumvent Kaggle's regulations, so Kaggle really should do something about it!

", "rawMarkdown": "Most posts on Xiaohongshu (the Chinese social media platform shown above) are full of lies: they can't guarantee medals nor teach you basic skills. At the same time, there also well-established institution out there that have GMs mentor and sell medals (I know some of them personally), and I can attest that they know pretty well how to circumvent Kaggle's regulations, so Kaggle really should do something about it!", "votes": 5 }, { "id": 2854592, "postDate": "2024-06-04T11:13:48.167Z", "content": "

The people that sell and buy solutions likely don't realize that they are the biggest losers in the universe.

\n

I propose we give those people the best places in the 'Losers Hall of Shame'….. I vote Yes.

", "rawMarkdown": "The people that sell and buy solutions likely don't realize that they are the biggest losers in the universe.\n\nI propose we give those people the best places in the 'Losers Hall of Shame'..... I vote Yes.", "votes": 5, "replies": [ { "id": 2854605, "postDate": "2024-06-04T11:19:22.587Z", "content": "

Would it not be likely that the seller is in the metal range for the public Leaderboard in each of these competitions? There couldn’t be many accounts with that But I guess it would be impossible to find which one exactly anyway most likely

", "rawMarkdown": "Would it not be likely that the seller is in the metal range for the public Leaderboard in each of these competitions? There couldn’t be many accounts with that But I guess it would be impossible to find which one exactly anyway most likely", "votes": 2 }, { "id": 2854624, "postDate": "2024-06-04T11:29:37.180Z", "content": "

Yes I'am afraid that is very much likely….you have to validate the solution you sell. Anyway let's hope that Kaggle can clean as much as possible…

", "rawMarkdown": "Yes I'am afraid that is very much likely....you have to validate the solution you sell. Anyway let's hope that Kaggle can clean as much as possible...", "votes": 3 } ] } ] }, { "id": 2854179, "postDate": "2024-06-04T06:45:14.363Z", "content": "

@oldufo @addisonhoward @ashleychow @wcukierski Maybe you can give the top 100 codes an extra check this time, many analyzes and reports from different users in this topic. It would help all contributors from doubts, reporters as well as top rankers, and also maintain the competition spirit and energy for coming competitions.

", "rawMarkdown": "@oldufo @addisonhoward @ashleychow @wcukierski Maybe you can give the top 100 codes an extra check this time, many analyzes and reports from different users in this topic. It would help all contributors from doubts, reporters as well as top rankers, and also maintain the competition spirit and energy for coming competitions.", "votes": 11 }, { "id": 2853791, "postDate": "2024-06-04T01:01:22.023Z", "content": "

@cody11null Many, many, many contributors 0.17

", "rawMarkdown": "@cody11null Many, many, many contributors 0.17", "votes": 9, "replies": [ { "id": 2853861, "postDate": "2024-06-04T02:27:38.920Z", "content": "

And majority of them are new accs with getting high scores in the last days of competition… I feel so sad for the fair teams who are losing their deserved medals because of this \"phenomenon\" =(

", "rawMarkdown": "And majority of them are new accs with getting high scores in the last days of competition... I feel so sad for the fair teams who are losing their deserved medals because of this \"phenomenon\" =(", "votes": 8, "replies": [ { "id": 2853934, "postDate": "2024-06-04T03:49:30.847Z", "content": "

Very true! I am not sure how “catchable” these potential cheaters would be but I am hoping that the Kaggle staff would find a handful of them but we will see!

", "rawMarkdown": "Very true! I am not sure how “catchable” these potential cheaters would be but I am hoping that the Kaggle staff would find a handful of them but we will see!", "votes": 2 }, { "id": 2853969, "postDate": "2024-06-04T04:13:02.063Z", "content": "

@clearwaterkzk @ayushkhaire @cody11null @zhonghaoyi @wolfmedal @vipin20 @vostankovich Two three contributors 0.17 May be a coincidence, it is unlikely that such a large scale 0.17, is there so many solutions for 0.17, I suggest that the kaggle team check from their submitted notes, I think what useful information can be found, such as whether the submitted code is the same open source code copy

", "rawMarkdown": "@clearwaterkzk @ayushkhaire @cody11null @zhonghaoyi @wolfmedal @vipin20 @vostankovich Two three contributors 0.17 May be a coincidence, it is unlikely that such a large scale 0.17, is there so many solutions for 0.17, I suggest that the kaggle team check from their submitted notes, I think what useful information can be found, such as whether the submitted code is the same open source code copy", "votes": 6 } ] } ] }, { "id": 2852892, "postDate": "2024-06-03T14:03:28.493Z", "content": "

this might be related to medals selling business again, but it's just an assumption

", "rawMarkdown": "this might be related to medals selling business again, but it's just an assumption", "votes": 10, "replies": [ { "id": 2852901, "postDate": "2024-06-03T14:07:21.823Z", "content": "

I have no clue but that would be a shame. I would think that it would be remarkably easy to catch no? Theoretically if it was true many would have very similar solutions? A little disappointed myself because this is the first time I have seen something like this on Kaggle. Of course it is only the public LB but 2 weeks ago I was in the fight for gold and now I am looking at near bronze hahaha

", "rawMarkdown": "I have no clue but that would be a shame. I would think that it would be remarkably easy to catch no? Theoretically if it was true many would have very similar solutions? A little disappointed myself because this is the first time I have seen something like this on Kaggle. Of course it is only the public LB but 2 weeks ago I was in the fight for gold and now I am looking at near bronze hahaha ", "votes": 4, "replies": [ { "id": 2852926, "postDate": "2024-06-03T14:17:35.723Z", "content": "

Unfortunately I've seen it in almost every comp. I am not sure, but I guess there is no automatic code checking for plagiarism etc. Kaggle staff have to manually check thousands of subs, and most probably they check only a small portion of subs, or those that are reported to them directly. I don't know the process =(

", "rawMarkdown": "Unfortunately I've seen it in almost every comp. I am not sure, but I guess there is no automatic code checking for plagiarism etc. Kaggle staff have to manually check thousands of subs, and most probably they check only a small portion of subs, or those that are reported to them directly. I don't know the process =(", "votes": 4 }, { "id": 2852950, "postDate": "2024-06-03T14:28:30.707Z", "content": "

Fair enough, either way it’s an unfortunate situation but I know the Kaggle staff will do their best! Hopefully there is some way to look at submitted subs only

", "rawMarkdown": "Fair enough, either way it’s an unfortunate situation but I know the Kaggle staff will do their best! Hopefully there is some way to look at submitted subs only", "votes": 3 } ] } ] }, { "id": 2854432, "postDate": "2024-06-04T10:13:56.153Z", "content": "

@oldufo @addisonhoward I want the kaggle team to check the open source traces of the code notebook copy they submitted, it is too coincidentally, so many 0.17 in two days, and then click on it to see, the activity has not been more than ten days, can ten days be on the list, I do not believe that there are so many 0.17 solutions, may need manual inspection

", "rawMarkdown": "@oldufo @addisonhoward I want the kaggle team to check the open source traces of the code notebook copy they submitted, it is too coincidentally, so many 0.17 in two days, and then click on it to see, the activity has not been more than ten days, can ten days be on the list, I do not believe that there are so many 0.17 solutions, may need manual inspection", "votes": 7, "replies": [ { "id": 2854612, "postDate": "2024-06-04T11:24:59.613Z", "content": "

", "rawMarkdown": "![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F17651006%2F5f5913207ae24f63dec55a29cc9e57fd%2F_20240604192433.png?generation=1717500297631060&alt=media)", "votes": 5, "replies": [ { "id": 2855748, "postDate": "2024-06-05T01:50:00.203Z", "content": "

@wolfmedal There are accounts that have almost zero submission throughout the comp and yet are members of the some top-scoring teams. Very suspicious

", "rawMarkdown": "@wolfmedal There are accounts that have almost zero submission throughout the comp and yet are members of the some top-scoring teams. Very suspicious", "votes": 3 }, { "id": 2855749, "postDate": "2024-06-05T01:54:08.390Z", "content": "

I only found one with 19 submissions

", "rawMarkdown": "I only found one with 19 submissions", "votes": 1 } ] } ] }, { "id": 2853802, "postDate": "2024-06-04T01:24:55.540Z", "content": "

I think the kaggle staff should focus on checking codes where the team size is 4 or 5 and the score is 0.16 and 0.17. Since I know people who have purchased medals for 2023 IMC, this year's setup should be similar.

", "rawMarkdown": "I think the kaggle staff should focus on checking codes where the team size is 4 or 5 and the score is 0.16 and 0.17. Since I know people who have purchased medals for 2023 IMC, this year's setup should be similar.", "votes": 7, "replies": [ { "id": 2853813, "postDate": "2024-06-04T01:35:53.153Z", "content": "

I think that is reasonable as well but really just the medal range in general. A difficult task but I have faith in the wonderful Kaggle staff to do everything in their power to make the right decisions!

", "rawMarkdown": "I think that is reasonable as well but really just the medal range in general. A difficult task but I have faith in the wonderful Kaggle staff to do everything in their power to make the right decisions! ", "votes": 3 }, { "id": 2854438, "postDate": "2024-06-04T10:17:18.440Z", "content": "

@zhonghaoyi Are there so many solutions for 0.17? It's my problem after all, I didn't find a solution for 0.17😂

", "rawMarkdown": "@zhonghaoyi Are there so many solutions for 0.17? It's my problem after all, I didn't find a solution for 0.17😂", "votes": 4 } ] }, { "id": 2854073, "postDate": "2024-06-04T05:56:23.060Z", "content": "

At this point, i too am losing the motivation to participate in kaggle challenges.While it remains an awesome place to test my knowledge and try out various problems, but as a competitive platform, Kaggle is losing its value in my opinion.
\nI just hope they genuinely fix the leaderboard

", "rawMarkdown": "At this point, i too am losing the motivation to participate in kaggle challenges.While it remains an awesome place to test my knowledge and try out various problems, but as a competitive platform, Kaggle is losing its value in my opinion.\nI just hope they genuinely fix the leaderboard", "votes": 8 }, { "id": 2857074, "postDate": "2024-06-05T17:25:31.083Z", "content": "

Last years competition was hard and there were like 500 teams. This one is way harder but somehow there are 1000 teams lol. I don't buy it.

", "rawMarkdown": "Last years competition was hard and there were like 500 teams. This one is way harder but somehow there are 1000 teams lol. I don't buy it.", "votes": 6 }, { "id": 2852955, "postDate": "2024-06-03T14:33:38.127Z", "content": "

Same feeling as me.
\nA weeks ago, the threshold between bronze and silver were ~0.15 but now over 0.16.

\n

This is just the information, but it happened last year and my colleague in China is observing that some participants in other competitions are selling solutions.
\nhttps://www.kaggle.com/competitions/image-matching-challenge-2023/discussion/416647

\n

Maybe Kaggle should host \"Detect Cheaters Challenge\", in which participants have timeseries user log in kaggle and predict cheaters probabilities…

", "rawMarkdown": "Same feeling as me.\nA weeks ago, the threshold between bronze and silver were ~0.15 but now over 0.16.\n\nThis is just the information, but it happened last year and my colleague in China is observing that some participants in other competitions are selling solutions. \nhttps://www.kaggle.com/competitions/image-matching-challenge-2023/discussion/416647\n\nMaybe Kaggle should host \"Detect Cheaters Challenge\", in which participants have timeseries user log in kaggle and predict cheaters probabilities...", "votes": 6, "replies": [ { "id": 2852986, "postDate": "2024-06-03T14:42:28.880Z", "content": "

That with submission code similarity would probably preform pretty well!

", "rawMarkdown": "That with submission code similarity would probably preform pretty well! ", "votes": 5 } ] }, { "id": 2860265, "postDate": "2024-06-07T14:05:14.240Z", "content": "

The distribution is certainly unnatural…

\n

", "rawMarkdown": "The distribution is certainly unnatural...\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F14688007%2F75e0baebd9a365ab2e9a968e39faf966%2FIMC2024_PLBhist.bmp?generation=1717805398550456&alt=media)", "votes": 5, "replies": [ { "id": 2873012, "postDate": "2024-06-15T08:51:03.110Z", "content": "

It seems that this competition is still ongoing…
\n2024/06/04 --- 81/1003
\n2024/06/10 --- 54/927
\n2024/06/15 --- 55/929

", "rawMarkdown": "It seems that this competition is still ongoing...\n2024/06/04 --- 81/1003\n2024/06/10 --- 54/927\n2024/06/15 --- 55/929", "replies": [ { "id": 2874426, "postDate": "2024-06-16T08:52:37.417Z", "content": "

@mitsuyasuhoshino https://www.kaggle.com/competitions/rsna-2024-lumbar-spine-degenerative-classification/discussion/511759
\nThis one's more interesting. Three newbie numbers registered on the same day

", "rawMarkdown": "@mitsuyasuhoshino https://www.kaggle.com/competitions/rsna-2024-lumbar-spine-degenerative-classification/discussion/511759\nThis one's more interesting. Three newbie numbers registered on the same day", "votes": 1 } ] } ] }, { "id": 2854466, "postDate": "2024-06-04T10:35:33.050Z", "content": "

Kaggle really needs to take a closer look into potential cheaters's solutions - manually if possible rather than relying on some pretrained detector. Medals sellers already get away with medals for too many times

", "rawMarkdown": "Kaggle really needs to take a closer look into potential cheaters's solutions - manually if possible rather than relying on some pretrained detector. Medals sellers already get away with medals for too many times", "votes": 3 }, { "id": 2854036, "postDate": "2024-06-04T05:19:34.157Z", "content": "

I posted some questionable LB results for a recently concluded competition. Even though there were obvious visual issues (not only public->private ranking changes, but also two teams with the exact same private LB scores), Kaggle did not remove these participants until I reported them. Therefore, I am questioning the validity of the LB cleanup that will be done in the next few days.

", "rawMarkdown": "I [posted](https://www.kaggle.com/competitions/home-credit-credit-risk-model-stability/discussion/509402) some questionable LB results for a recently concluded competition. Even though there were obvious visual issues (not only public->private ranking changes, but also two teams with the exact same private LB scores), Kaggle did not remove these participants until I reported them. Therefore, I am questioning the validity of the LB cleanup that will be done in the next few days.", "votes": 4, "replies": [ { "id": 2854474, "postDate": "2024-06-04T10:39:16.983Z", "content": "

Colmap reconstruction is plagued with randomness and thus it's unlikely for two teams to have same scores in this comp, but you are absolutely right: cleanup needs to be more transparent and trustworthy

", "rawMarkdown": "Colmap reconstruction is plagued with randomness and thus it's unlikely for two teams to have same scores in this comp, but you are absolutely right: cleanup needs to be more transparent and trustworthy", "votes": 4, "replies": [ { "id": 2854502, "postDate": "2024-06-04T10:47:34.323Z", "content": "

@renyiwei That's a good point, but I don't think there are so many solutions. A careful manual inspection will yield some results. If that doesn't work, you can set up a cheating query competition.🤣

", "rawMarkdown": "@renyiwei That's a good point, but I don't think there are so many solutions. A careful manual inspection will yield some results. If that doesn't work, you can set up a cheating query competition.🤣", "votes": 5 }, { "id": 2854544, "postDate": "2024-06-04T10:58:55.223Z", "content": "

You are right. I think most team has used SP+LG and hardnet affnet etc, so it will be a little tricky to detect. It feels really bad to drop suddenly from Top 20 to Bronze medal zone, especially when it might be attributed to some cheaters.

", "rawMarkdown": "You are right. I think most team has used SP+LG and hardnet affnet etc, so it will be a little tricky to detect. It feels really bad to drop suddenly from Top 20 to Bronze medal zone, especially when it might be attributed to some cheaters." }, { "id": 2854553, "postDate": "2024-06-04T11:01:32.510Z", "content": "

@renyiwei This is not very important. This is my first time to participate in 3D reconstruction. I am looking forward to the solutions and open source codes of the big guys in front.

", "rawMarkdown": "@renyiwei This is not very important. This is my first time to participate in 3D reconstruction. I am looking forward to the solutions and open source codes of the big guys in front." }, { "id": 2854620, "postDate": "2024-06-04T11:28:56.663Z", "rawMarkdown": "", "isDeleted": true }, { "id": 2856084, "postDate": "2024-06-05T06:25:24.087Z", "content": "

We dropped from the 12th to the 72nd position. The reason for our initial success, placing us at 12th, was our effective handling of the transparent categories. However, a mistake in the parameter settings for our method dealing with non-transparent categories led to our subsequent drop to the 72nd position.

", "rawMarkdown": "We dropped from the 12th to the 72nd position. The reason for our initial success, placing us at 12th, was our effective handling of the transparent categories. However, a mistake in the parameter settings for our method dealing with non-transparent categories led to our subsequent drop to the 72nd position.", "votes": 2 }, { "id": 2856097, "postDate": "2024-06-05T06:32:33.843Z", "content": "

@chenxupeng this is unfortunate, I am sorry for this as I understand that it must be painful. Have you tried to late submit after fixing the bug?

", "rawMarkdown": "@chenxupeng this is unfortunate, I am sorry for this as I understand that it must be painful. Have you tried to late submit after fixing the bug? " } ] } ] }, { "id": 2853084, "postDate": "2024-06-03T15:34:33.153Z", "content": "

Same feeling !
\nfew hour ago i have 135th rank now i am on 147 so something is going on or maybe they have computation power to experiment locally.

", "rawMarkdown": "Same feeling !\nfew hour ago i have 135th rank now i am on 147 so something is going on or maybe they have computation power to experiment locally.", "votes": 2 }, { "id": 2852987, "postDate": "2024-06-03T14:42:41.723Z", "content": "

@cody11null

\n

Sir I am not in this competition, but I wish to see topper solution code . Can you provide? Specifically 0.28 that make a difference.

\n

The situation you are saying I think can be a private sharing or a copy paste to improve code . Because scores are almost same .

", "rawMarkdown": "@cody11null \n\nSir I am not in this competition, but I wish to see topper solution code . Can you provide? Specifically 0.28 that make a difference. \n\nThe situation you are saying I think can be a private sharing or a copy paste to improve code . Because scores are almost same .", "votes": 2 }, { "id": 2854013, "postDate": "2024-06-04T04:51:18.333Z", "content": "

It happens every reward competition,eapecially last day.

", "rawMarkdown": "It happens every reward competition,eapecially last day.", "votes": 1 }, { "id": 2863532, "postDate": "2024-06-09T13:13:45.477Z", "content": "

The Kaggle staff should prioritize reviewing code submissions where teams have four or five members and scores around 0.16 and 0.17. Based on my observations, some participants acquired medals in the 2023 IMC, suggesting a similar situation could be occurring this year.

", "rawMarkdown": "The Kaggle staff should prioritize reviewing code submissions where teams have four or five members and scores around 0.16 and 0.17. Based on my observations, some participants acquired medals in the 2023 IMC, suggesting a similar situation could be occurring this year.", "votes": -1 }, { "id": 2863123, "postDate": "2024-06-09T08:32:43.933Z", "rawMarkdown": "", "isDeleted": true } ], "comments": [ { "id": 2854343, "author_name": "Muhammad Ahmed", "author_url": "", "post_date": "2024-06-04T09:05:10.937000", "content": "

Unfortunately, medals were being sold in this competition.
\nI hope the organizers do something about it.

", "votes": 17, "replies": [ { "id": 2854359, "author_name": "Kirderf", "author_url": "", "post_date": "2024-06-04T09:18:13.170000", "content": "

@oldufo @addisonhoward @ashleychow @wcukierski see above, time checking the top 100 codes extra and broadly.

", "votes": 8, "replies": [] }, { "id": 2854420, "author_name": "", "author_url": "", "post_date": "2024-06-04T10:04:50.107000", "content": "", "votes": 0, "replies": [] }, { "id": 2854456, "author_name": "Roy Wei", "author_url": "", "post_date": "2024-06-04T10:26:29.740000", "content": "

Most posts on Xiaohongshu (the Chinese social media platform shown above) are full of lies: they can't guarantee medals nor teach you basic skills. At the same time, there also well-established institution out there that have GMs mentor and sell medals (I know some of them personally), and I can attest that they know pretty well how to circumvent Kaggle's regulations, so Kaggle really should do something about it!

", "votes": 5, "replies": [] }, { "id": 2854592, "author_name": "Robin Smits", "author_url": "", "post_date": "2024-06-04T11:13:48.167000", "content": "

The people that sell and buy solutions likely don't realize that they are the biggest losers in the universe.

\n

I propose we give those people the best places in the 'Losers Hall of Shame'….. I vote Yes.

", "votes": 5, "replies": [ { "id": 2854605, "author_name": "Cody_Null", "author_url": "", "post_date": "2024-06-04T11:19:22.587000", "content": "

Would it not be likely that the seller is in the metal range for the public Leaderboard in each of these competitions? There couldn’t be many accounts with that But I guess it would be impossible to find which one exactly anyway most likely

", "votes": 2, "replies": [] }, { "id": 2854624, "author_name": "Robin Smits", "author_url": "", "post_date": "2024-06-04T11:29:37.180000", "content": "

Yes I'am afraid that is very much likely….you have to validate the solution you sell. Anyway let's hope that Kaggle can clean as much as possible…

", "votes": 3, "replies": [] } ] } ] }, { "id": 2854179, "author_name": "Kirderf", "author_url": "", "post_date": "2024-06-04T06:45:14.363000", "content": "

@oldufo @addisonhoward @ashleychow @wcukierski Maybe you can give the top 100 codes an extra check this time, many analyzes and reports from different users in this topic. It would help all contributors from doubts, reporters as well as top rankers, and also maintain the competition spirit and energy for coming competitions.

", "votes": 11, "replies": [] }, { "id": 2853791, "author_name": "", "author_url": "", "post_date": "2024-06-04T01:01:22.023000", "content": "

@cody11null Many, many, many contributors 0.17

", "votes": 9, "replies": [ { "id": 2853861, "author_name": "Vladislav Ostankovich", "author_url": "", "post_date": "2024-06-04T02:27:38.920000", "content": "

And majority of them are new accs with getting high scores in the last days of competition… I feel so sad for the fair teams who are losing their deserved medals because of this \"phenomenon\" =(

", "votes": 8, "replies": [ { "id": 2853934, "author_name": "Cody_Null", "author_url": "", "post_date": "2024-06-04T03:49:30.847000", "content": "

Very true! I am not sure how “catchable” these potential cheaters would be but I am hoping that the Kaggle staff would find a handful of them but we will see!

", "votes": 2, "replies": [] }, { "id": 2853969, "author_name": "", "author_url": "", "post_date": "2024-06-04T04:13:02.063000", "content": "

@clearwaterkzk @ayushkhaire @cody11null @zhonghaoyi @wolfmedal @vipin20 @vostankovich Two three contributors 0.17 May be a coincidence, it is unlikely that such a large scale 0.17, is there so many solutions for 0.17, I suggest that the kaggle team check from their submitted notes, I think what useful information can be found, such as whether the submitted code is the same open source code copy

", "votes": 6, "replies": [] } ] } ] }, { "id": 2852892, "author_name": "Vladislav Ostankovich", "author_url": "", "post_date": "2024-06-03T14:03:28.493000", "content": "

this might be related to medals selling business again, but it's just an assumption

", "votes": 10, "replies": [ { "id": 2852901, "author_name": "Cody_Null", "author_url": "", "post_date": "2024-06-03T14:07:21.823000", "content": "

I have no clue but that would be a shame. I would think that it would be remarkably easy to catch no? Theoretically if it was true many would have very similar solutions? A little disappointed myself because this is the first time I have seen something like this on Kaggle. Of course it is only the public LB but 2 weeks ago I was in the fight for gold and now I am looking at near bronze hahaha

", "votes": 4, "replies": [ { "id": 2852926, "author_name": "Vladislav Ostankovich", "author_url": "", "post_date": "2024-06-03T14:17:35.723000", "content": "

Unfortunately I've seen it in almost every comp. I am not sure, but I guess there is no automatic code checking for plagiarism etc. Kaggle staff have to manually check thousands of subs, and most probably they check only a small portion of subs, or those that are reported to them directly. I don't know the process =(

", "votes": 4, "replies": [] }, { "id": 2852950, "author_name": "Cody_Null", "author_url": "", "post_date": "2024-06-03T14:28:30.707000", "content": "

Fair enough, either way it’s an unfortunate situation but I know the Kaggle staff will do their best! Hopefully there is some way to look at submitted subs only

", "votes": 3, "replies": [] } ] } ] }, { "id": 2854432, "author_name": "", "author_url": "", "post_date": "2024-06-04T10:13:56.153000", "content": "

@oldufo @addisonhoward I want the kaggle team to check the open source traces of the code notebook copy they submitted, it is too coincidentally, so many 0.17 in two days, and then click on it to see, the activity has not been more than ten days, can ten days be on the list, I do not believe that there are so many 0.17 solutions, may need manual inspection

", "votes": 7, "replies": [ { "id": 2854612, "author_name": "", "author_url": "", "post_date": "2024-06-04T11:24:59.613000", "content": "

", "votes": 5, "replies": [ { "id": 2855748, "author_name": "Roy Wei", "author_url": "", "post_date": "2024-06-05T01:50:00.203000", "content": "

@wolfmedal There are accounts that have almost zero submission throughout the comp and yet are members of the some top-scoring teams. Very suspicious

", "votes": 3, "replies": [] }, { "id": 2855749, "author_name": "", "author_url": "", "post_date": "2024-06-05T01:54:08.390000", "content": "

I only found one with 19 submissions

", "votes": 1, "replies": [] } ] } ] }, { "id": 2853802, "author_name": "HAOYI ZHONG", "author_url": "", "post_date": "2024-06-04T01:24:55.540000", "content": "

I think the kaggle staff should focus on checking codes where the team size is 4 or 5 and the score is 0.16 and 0.17. Since I know people who have purchased medals for 2023 IMC, this year's setup should be similar.

", "votes": 7, "replies": [ { "id": 2853813, "author_name": "Cody_Null", "author_url": "", "post_date": "2024-06-04T01:35:53.153000", "content": "

I think that is reasonable as well but really just the medal range in general. A difficult task but I have faith in the wonderful Kaggle staff to do everything in their power to make the right decisions!

", "votes": 3, "replies": [] }, { "id": 2854438, "author_name": "", "author_url": "", "post_date": "2024-06-04T10:17:18.440000", "content": "

@zhonghaoyi Are there so many solutions for 0.17? It's my problem after all, I didn't find a solution for 0.17😂

", "votes": 4, "replies": [] } ] }, { "id": 2854073, "author_name": "Bhavesh Jain", "author_url": "", "post_date": "2024-06-04T05:56:23.060000", "content": "

At this point, i too am losing the motivation to participate in kaggle challenges.While it remains an awesome place to test my knowledge and try out various problems, but as a competitive platform, Kaggle is losing its value in my opinion.
\nI just hope they genuinely fix the leaderboard

", "votes": 8, "replies": [] }, { "id": 2857074, "author_name": "Gunes Evitan", "author_url": "", "post_date": "2024-06-05T17:25:31.083000", "content": "

Last years competition was hard and there were like 500 teams. This one is way harder but somehow there are 1000 teams lol. I don't buy it.

", "votes": 6, "replies": [] }, { "id": 2852955, "author_name": "Aurora_blue", "author_url": "", "post_date": "2024-06-03T14:33:38.127000", "content": "

Same feeling as me.
\nA weeks ago, the threshold between bronze and silver were ~0.15 but now over 0.16.

\n

This is just the information, but it happened last year and my colleague in China is observing that some participants in other competitions are selling solutions.
\nhttps://www.kaggle.com/competitions/image-matching-challenge-2023/discussion/416647

\n

Maybe Kaggle should host \"Detect Cheaters Challenge\", in which participants have timeseries user log in kaggle and predict cheaters probabilities…

", "votes": 6, "replies": [ { "id": 2852986, "author_name": "Cody_Null", "author_url": "", "post_date": "2024-06-03T14:42:28.880000", "content": "

That with submission code similarity would probably preform pretty well!

", "votes": 5, "replies": [] } ] }, { "id": 2860265, "author_name": "mh", "author_url": "", "post_date": "2024-06-07T14:05:14.240000", "content": "

The distribution is certainly unnatural…

\n

", "votes": 5, "replies": [ { "id": 2873012, "author_name": "mh", "author_url": "", "post_date": "2024-06-15T08:51:03.110000", "content": "

It seems that this competition is still ongoing…
\n2024/06/04 --- 81/1003
\n2024/06/10 --- 54/927
\n2024/06/15 --- 55/929

", "votes": 0, "replies": [ { "id": 2874426, "author_name": "", "author_url": "", "post_date": "2024-06-16T08:52:37.417000", "content": "

@mitsuyasuhoshino https://www.kaggle.com/competitions/rsna-2024-lumbar-spine-degenerative-classification/discussion/511759
\nThis one's more interesting. Three newbie numbers registered on the same day

", "votes": 1, "replies": [] } ] } ] }, { "id": 2854466, "author_name": "Roy Wei", "author_url": "", "post_date": "2024-06-04T10:35:33.050000", "content": "

Kaggle really needs to take a closer look into potential cheaters's solutions - manually if possible rather than relying on some pretrained detector. Medals sellers already get away with medals for too many times

", "votes": 3, "replies": [] }, { "id": 2854036, "author_name": "mh", "author_url": "", "post_date": "2024-06-04T05:19:34.157000", "content": "

I posted some questionable LB results for a recently concluded competition. Even though there were obvious visual issues (not only public->private ranking changes, but also two teams with the exact same private LB scores), Kaggle did not remove these participants until I reported them. Therefore, I am questioning the validity of the LB cleanup that will be done in the next few days.

", "votes": 4, "replies": [ { "id": 2854474, "author_name": "Roy Wei", "author_url": "", "post_date": "2024-06-04T10:39:16.983000", "content": "

Colmap reconstruction is plagued with randomness and thus it's unlikely for two teams to have same scores in this comp, but you are absolutely right: cleanup needs to be more transparent and trustworthy

", "votes": 4, "replies": [ { "id": 2854502, "author_name": "", "author_url": "", "post_date": "2024-06-04T10:47:34.323000", "content": "

@renyiwei That's a good point, but I don't think there are so many solutions. A careful manual inspection will yield some results. If that doesn't work, you can set up a cheating query competition.🤣

", "votes": 5, "replies": [] }, { "id": 2854544, "author_name": "Roy Wei", "author_url": "", "post_date": "2024-06-04T10:58:55.223000", "content": "

You are right. I think most team has used SP+LG and hardnet affnet etc, so it will be a little tricky to detect. It feels really bad to drop suddenly from Top 20 to Bronze medal zone, especially when it might be attributed to some cheaters.

", "votes": 0, "replies": [] }, { "id": 2854553, "author_name": "", "author_url": "", "post_date": "2024-06-04T11:01:32.510000", "content": "

@renyiwei This is not very important. This is my first time to participate in 3D reconstruction. I am looking forward to the solutions and open source codes of the big guys in front.

", "votes": 0, "replies": [] }, { "id": 2854620, "author_name": "", "author_url": "", "post_date": "2024-06-04T11:28:56.663000", "content": "", "votes": 0, "replies": [] }, { "id": 2856084, "author_name": "Chainey", "author_url": "", "post_date": "2024-06-05T06:25:24.087000", "content": "

We dropped from the 12th to the 72nd position. The reason for our initial success, placing us at 12th, was our effective handling of the transparent categories. However, a mistake in the parameter settings for our method dealing with non-transparent categories led to our subsequent drop to the 72nd position.

", "votes": 2, "replies": [] }, { "id": 2856097, "author_name": "Roy Wei", "author_url": "", "post_date": "2024-06-05T06:32:33.843000", "content": "

@chenxupeng this is unfortunate, I am sorry for this as I understand that it must be painful. Have you tried to late submit after fixing the bug?

", "votes": 0, "replies": [] } ] } ] }, { "id": 2853084, "author_name": "Vipin Kumar", "author_url": "", "post_date": "2024-06-03T15:34:33.153000", "content": "

Same feeling !
\nfew hour ago i have 135th rank now i am on 147 so something is going on or maybe they have computation power to experiment locally.

", "votes": 2, "replies": [] }, { "id": 2852987, "author_name": "Ayush Khaire", "author_url": "", "post_date": "2024-06-03T14:42:41.723000", "content": "

@cody11null

\n

Sir I am not in this competition, but I wish to see topper solution code . Can you provide? Specifically 0.28 that make a difference.

\n

The situation you are saying I think can be a private sharing or a copy paste to improve code . Because scores are almost same .

", "votes": 2, "replies": [] }, { "id": 2854013, "author_name": "", "author_url": "", "post_date": "2024-06-04T04:51:18.333000", "content": "

It happens every reward competition,eapecially last day.

", "votes": 1, "replies": [] }, { "id": 2863532, "author_name": "Muhammad Azeem", "author_url": "", "post_date": "2024-06-09T13:13:45.477000", "content": "

The Kaggle staff should prioritize reviewing code submissions where teams have four or five members and scores around 0.16 and 0.17. Based on my observations, some participants acquired medals in the 2023 IMC, suggesting a similar situation could be occurring this year.

", "votes": -1, "replies": [] }, { "id": 2863123, "author_name": "", "author_url": "", "post_date": "2024-06-09T08:32:43.933000", "content": "", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "2852873": "I have faith in the Kaggle staff to prevent any wrong doing that may or may not be going on here but I do see something strange I wanted to point out. In the last few weeks we always expect a pickup in competition but this has been special. Over the last 3 weeks we went from 12 people having a score of 0.16 to now approaching 100. And the number of 0.17 has gone from 10 to 43 in a week as well. What I hadn't noticed until recently is they are a large percentage of people in this group who have created their account during the competition. This competition is very challenging so I certainly did not expect so many to be able to make such a push in the endgame. I do hope that the incredible staff at Kaggle does not have their hands too full with this in their no doubt already busy day. ", "2854343": "Unfortunately, medals were being sold in this competition. \nI hope the organizers do something about it. ![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F3350264%2F32d51d9c38be0998964400a3f1bd6b74%2Fimg_0658.jpg?generation=1717491904290772&alt=media)", "2854179": "@oldufo @addisonhoward @ashleychow @wcukierski Maybe you can give the top 100 codes an extra check this time, many analyzes and reports from different users in this topic. It would help all contributors from doubts, reporters as well as top rankers, and also maintain the competition spirit and energy for coming competitions.", "2853791": "@cody11null Many, many, many contributors 0.17", "2852892": "this might be related to medals selling business again, but it's just an assumption", "2854432": "@oldufo @addisonhoward I want the kaggle team to check the open source traces of the code notebook copy they submitted, it is too coincidentally, so many 0.17 in two days, and then click on it to see, the activity has not been more than ten days, can ten days be on the list, I do not believe that there are so many 0.17 solutions, may need manual inspection", "2853802": "I think the kaggle staff should focus on checking codes where the team size is 4 or 5 and the score is 0.16 and 0.17. Since I know people who have purchased medals for 2023 IMC, this year's setup should be similar.", "2854073": "At this point, i too am losing the motivation to participate in kaggle challenges.While it remains an awesome place to test my knowledge and try out various problems, but as a competitive platform, Kaggle is losing its value in my opinion.\nI just hope they genuinely fix the leaderboard", "2857074": "Last years competition was hard and there were like 500 teams. This one is way harder but somehow there are 1000 teams lol. I don't buy it.", "2852955": "Same feeling as me.\nA weeks ago, the threshold between bronze and silver were ~0.15 but now over 0.16.\n\nThis is just the information, but it happened last year and my colleague in China is observing that some participants in other competitions are selling solutions. \nhttps://www.kaggle.com/competitions/image-matching-challenge-2023/discussion/416647\n\nMaybe Kaggle should host \"Detect Cheaters Challenge\", in which participants have timeseries user log in kaggle and predict cheaters probabilities...", "2860265": "The distribution is certainly unnatural...\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F14688007%2F75e0baebd9a365ab2e9a968e39faf966%2FIMC2024_PLBhist.bmp?generation=1717805398550456&alt=media)", "2854466": "Kaggle really needs to take a closer look into potential cheaters's solutions - manually if possible rather than relying on some pretrained detector. Medals sellers already get away with medals for too many times", "2854036": "I [posted](https://www.kaggle.com/competitions/home-credit-credit-risk-model-stability/discussion/509402) some questionable LB results for a recently concluded competition. Even though there were obvious visual issues (not only public->private ranking changes, but also two teams with the exact same private LB scores), Kaggle did not remove these participants until I reported them. Therefore, I am questioning the validity of the LB cleanup that will be done in the next few days.", "2853084": "Same feeling !\nfew hour ago i have 135th rank now i am on 147 so something is going on or maybe they have computation power to experiment locally.", "2852987": "@cody11null \n\nSir I am not in this competition, but I wish to see topper solution code . Can you provide? Specifically 0.28 that make a difference. \n\nThe situation you are saying I think can be a private sharing or a copy paste to improve code . Because scores are almost same .", "2854013": "It happens every reward competition,eapecially last day.", "2863532": "The Kaggle staff should prioritize reviewing code submissions where teams have four or five members and scores around 0.16 and 0.17. Based on my observations, some participants acquired medals in the 2023 IMC, suggesting a similar situation could be occurring this year.", "2863123": "" } }