{ "id": 187225, "title": "Fake accounts in Silver medals? (zero submissions, registered <14 days ago)", "url": "/competitions/landmark-recognition-2020/discussion/187225", "author_name": "Chan Kha Vu", "post_date": "2020-09-28T04:35:52.467000", "votes": 18, "comment_count": 4, "views": 0, "content": "

Allow me to be an a** here and bring this topic up, regardless of the possible downvotes that I'll get (discussion medals and scores are not important to me anyways) πŸ™„ But I think some participants might share my concerns.

\n
\n

So, in the last few days I've been stalking (aka fanboy-ing) the accounts of the top teams in this competition (because they're amazing and I enjoyed their write-ups for the previous competitions that they've won; some of them have just graduated High School and already GMs - just WOW).

\n

However, as I go further down the LB (lower than the gold medals), I've started to notice some strange accounts even in the top 30:

\n\n

I appreciate that many people just wants to remain anonymous. In other cases, those with zero Kaggle contributions might be contributing awesome ideas for the team (this competition relies quite heavily on new ideas). But in some cases, it's just too… suspicious. The same concern has been shared and confirmed by other people in past competitions:

\n\n

Also, the following looks funny (although irrelevant for this competition, it might be a part of making these accounts seem more \"real\" for future competitions)
\n\"\"

\n

There are some in the Retrieval Track as well, but that's already a thing in the past so just let it be 😊

\n
\n

On a brighter note, regardless of the account boosting stuffs expressed above, this has been an amazing competition and a learning experience ^^ Can't wait to read the solutions of top teams (some of them are using GANs, SuperPoint, SuperGlue, and other fancy stuffs :D :D :D)

", "messages": [ { "id": 1029715, "postDate": "2020-09-28T04:35:52.467Z", "content": "

Allow me to be an a** here and bring this topic up, regardless of the possible downvotes that I'll get (discussion medals and scores are not important to me anyways) πŸ™„ But I think some participants might share my concerns.

\n
\n

So, in the last few days I've been stalking (aka fanboy-ing) the accounts of the top teams in this competition (because they're amazing and I enjoyed their write-ups for the previous competitions that they've won; some of them have just graduated High School and already GMs - just WOW).

\n

However, as I go further down the LB (lower than the gold medals), I've started to notice some strange accounts even in the top 30:

\n\n

I appreciate that many people just wants to remain anonymous. In other cases, those with zero Kaggle contributions might be contributing awesome ideas for the team (this competition relies quite heavily on new ideas). But in some cases, it's just too… suspicious. The same concern has been shared and confirmed by other people in past competitions:

\n\n

Also, the following looks funny (although irrelevant for this competition, it might be a part of making these accounts seem more \"real\" for future competitions)
\n\"\"

\n

There are some in the Retrieval Track as well, but that's already a thing in the past so just let it be 😊

\n
\n

On a brighter note, regardless of the account boosting stuffs expressed above, this has been an amazing competition and a learning experience ^^ Can't wait to read the solutions of top teams (some of them are using GANs, SuperPoint, SuperGlue, and other fancy stuffs :D :D :D)

", "rawMarkdown": "Allow me to be an a** here and bring this topic up, regardless of the possible downvotes that I'll get (discussion medals and scores are not important to me anyways) πŸ™„ But I think some participants might share my concerns.\n\n----------------------------------------------------------------------------------------\n\nSo, in the last few days I've been stalking (aka fanboy-ing) the accounts of the top teams in this competition (because they're amazing and I enjoyed their write-ups for the previous competitions that they've won; some of them have just graduated High School and already GMs - just WOW).\n\nHowever, as I go further down the LB (lower than the gold medals), I've started to notice some strange accounts even in the **top 30**:\n\n- Teams of \"novices\" that joined less than a month ago, for this specific competition. Some **just joined 14-15 days ago**, which is ridiculous - 2 weeks is barely enough time to write training pipeline and train a model for 0.56 (including re-ranking improvements and landmark/non-landmark cleaning), assuming that you know exactly what you do. Even @bestfitting ([the **tourist**](https://en.wikipedia.org/wiki/Gennady_Korotkevich) of Kaggle, if you know what I mean, with immense amount of experience in metric-learning competitions) needed around that amount of time to reach 0.55. Some cropped screenshots:\n ![registered 14 days ago](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F2270821%2Fed101bd3ac3101fe467ac81f8025109f%2Fsuspicious_4.png?generation=1601269034778120&alt=media)\n ![joined 15 days ago](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F2270821%2Ff5a0ba777909dffd3f7d247b9cac8c52%2Fsuspicious_8.png?generation=1601269107125525&alt=media)\n\n- Some members of those teams were **\"last seen 8-15 days ago\"**, which is quite strange. Even if we assume that they are busy with life outside Kaggle (which is good, we should all do less Kaggle and more of real-life stuffs), isn't one supposed to be curious about the LB standing every now and then? Or it's just me being a bit obsessed with the leaderboard? Some cropped screenshots:\n ![last seen 13 days ago](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F2270821%2F7b9d4f1ef61727ec143f69a35961a87f%2Fsuspicious_3.png?generation=1601268830488772&alt=media)\n ![last seen 11 days ago](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F2270821%2F9134808be7514179e1e4922b6d4c0b80%2Fsuspicious_5.png?generation=1601268888576295&alt=media)\n\n- **0 (zero) contributions** from those accounts. And only 1 person in the team performs all of the contributions (submissions). I'm pretty sure that's not how it should be in teams.\n ![Zero contributions in Kaggle](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F2270821%2Ffb7ed489f2e4d824d0dfc5f845bd6589%2FScreenshot%20from%202020-09-27%2021-52-06.png?generation=1601269213475186&alt=media)\n\nI appreciate that many people just wants to remain anonymous. In other cases, those with zero Kaggle contributions might be contributing awesome ideas for the team (this competition relies quite heavily on new ideas). But in some cases, it's just too... suspicious. The same concern has been shared and confirmed by other people in past competitions:\n\n- https://www.kaggle.com/c/siim-isic-melanoma-classification/discussion/175156\n- https://www.kaggle.com/c/siim-isic-melanoma-classification/discussion/174635\n\nAlso, the following looks funny (although irrelevant for this competition, it might be a part of making these accounts seem more \"real\" for future competitions)\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F2270821%2F4ac7bc20ae9d1da211a79732aa2b4e15%2FScreenshot%20from%202020-09-27%2020-56-22.png?generation=1601267148132571&alt=media)\n\nThere are some in the [Retrieval Track](https://www.kaggle.com/c/landmark-retrieval-2020/leaderboard) as well, but that's already a thing in the past so just let it be 😊\n\n-----------------------------------------------------------\n\nOn a brighter note, regardless of the account boosting stuffs expressed above, this has been an amazing competition and a learning experience ^^ Can't wait to read the solutions of top teams (some of them are using GANs, SuperPoint, SuperGlue, and other fancy stuffs :D :D :D)", "votes": 18 }, { "id": 1029936, "postDate": "2020-09-28T09:29:39.257Z", "content": "

@chankhavu - Strange user activity has been highlighted in many competition these days. I have been training a model for last 72 hours and my score has just reached 0.4561..

", "rawMarkdown": "@chankhavu - Strange user activity has been highlighted in many competition these days. I have been training a model for last 72 hours and my score has just reached 0.4561.. ", "votes": 2, "replies": [ { "id": 1029941, "postDate": "2020-09-28T09:36:33.163Z", "content": "

Wow, that's really cool. I have a deep admiration for anyone who managed to train a model (any model, regardless of the score) for this competition - this is freaking hard to do πŸ˜€πŸ˜€πŸ˜€πŸ˜€ Huge props to you for attempting to use something other than the baseline model ^^

", "rawMarkdown": "Wow, that's really cool. I have a deep admiration for anyone who managed to train a model (any model, regardless of the score) for this competition - this is freaking hard to do πŸ˜€πŸ˜€πŸ˜€πŸ˜€ Huge props to you for attempting to use something other than the baseline model ^^", "votes": 4 } ] }, { "id": 1029726, "postDate": "2020-09-28T04:55:44.793Z", "rawMarkdown": "", "isDeleted": true, "replies": [ { "id": 1029728, "postDate": "2020-09-28T05:01:33.387Z", "content": "

Yeah, I've linked that topic at the bottom of this post :D

", "rawMarkdown": "Yeah, I've linked that topic at the bottom of this post :D ", "votes": 1 } ] } ], "comments": [ { "id": 1029936, "author_name": "Manoj Prabhakar", "author_url": "", "post_date": "2020-09-28T09:29:39.257000", "content": "

@chankhavu - Strange user activity has been highlighted in many competition these days. I have been training a model for last 72 hours and my score has just reached 0.4561..

", "votes": 2, "replies": [ { "id": 1029941, "author_name": "Chan Kha Vu", "author_url": "", "post_date": "2020-09-28T09:36:33.163000", "content": "

Wow, that's really cool. I have a deep admiration for anyone who managed to train a model (any model, regardless of the score) for this competition - this is freaking hard to do πŸ˜€πŸ˜€πŸ˜€πŸ˜€ Huge props to you for attempting to use something other than the baseline model ^^

", "votes": 4, "replies": [] } ] }, { "id": 1029726, "author_name": "", "author_url": "", "post_date": "2020-09-28T04:55:44.793000", "content": "", "votes": 0, "replies": [ { "id": 1029728, "author_name": "Chan Kha Vu", "author_url": "", "post_date": "2020-09-28T05:01:33.387000", "content": "

Yeah, I've linked that topic at the bottom of this post :D

", "votes": 1, "replies": [] } ] } ], "raw_markdown_by_id": { "1029715": "Allow me to be an a** here and bring this topic up, regardless of the possible downvotes that I'll get (discussion medals and scores are not important to me anyways) πŸ™„ But I think some participants might share my concerns.\n\n----------------------------------------------------------------------------------------\n\nSo, in the last few days I've been stalking (aka fanboy-ing) the accounts of the top teams in this competition (because they're amazing and I enjoyed their write-ups for the previous competitions that they've won; some of them have just graduated High School and already GMs - just WOW).\n\nHowever, as I go further down the LB (lower than the gold medals), I've started to notice some strange accounts even in the **top 30**:\n\n- Teams of \"novices\" that joined less than a month ago, for this specific competition. Some **just joined 14-15 days ago**, which is ridiculous - 2 weeks is barely enough time to write training pipeline and train a model for 0.56 (including re-ranking improvements and landmark/non-landmark cleaning), assuming that you know exactly what you do. Even @bestfitting ([the **tourist**](https://en.wikipedia.org/wiki/Gennady_Korotkevich) of Kaggle, if you know what I mean, with immense amount of experience in metric-learning competitions) needed around that amount of time to reach 0.55. Some cropped screenshots:\n ![registered 14 days ago](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F2270821%2Fed101bd3ac3101fe467ac81f8025109f%2Fsuspicious_4.png?generation=1601269034778120&alt=media)\n ![joined 15 days ago](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F2270821%2Ff5a0ba777909dffd3f7d247b9cac8c52%2Fsuspicious_8.png?generation=1601269107125525&alt=media)\n\n- Some members of those teams were **\"last seen 8-15 days ago\"**, which is quite strange. Even if we assume that they are busy with life outside Kaggle (which is good, we should all do less Kaggle and more of real-life stuffs), isn't one supposed to be curious about the LB standing every now and then? Or it's just me being a bit obsessed with the leaderboard? Some cropped screenshots:\n ![last seen 13 days ago](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F2270821%2F7b9d4f1ef61727ec143f69a35961a87f%2Fsuspicious_3.png?generation=1601268830488772&alt=media)\n ![last seen 11 days ago](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F2270821%2F9134808be7514179e1e4922b6d4c0b80%2Fsuspicious_5.png?generation=1601268888576295&alt=media)\n\n- **0 (zero) contributions** from those accounts. And only 1 person in the team performs all of the contributions (submissions). I'm pretty sure that's not how it should be in teams.\n ![Zero contributions in Kaggle](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F2270821%2Ffb7ed489f2e4d824d0dfc5f845bd6589%2FScreenshot%20from%202020-09-27%2021-52-06.png?generation=1601269213475186&alt=media)\n\nI appreciate that many people just wants to remain anonymous. In other cases, those with zero Kaggle contributions might be contributing awesome ideas for the team (this competition relies quite heavily on new ideas). But in some cases, it's just too... suspicious. The same concern has been shared and confirmed by other people in past competitions:\n\n- https://www.kaggle.com/c/siim-isic-melanoma-classification/discussion/175156\n- https://www.kaggle.com/c/siim-isic-melanoma-classification/discussion/174635\n\nAlso, the following looks funny (although irrelevant for this competition, it might be a part of making these accounts seem more \"real\" for future competitions)\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F2270821%2F4ac7bc20ae9d1da211a79732aa2b4e15%2FScreenshot%20from%202020-09-27%2020-56-22.png?generation=1601267148132571&alt=media)\n\nThere are some in the [Retrieval Track](https://www.kaggle.com/c/landmark-retrieval-2020/leaderboard) as well, but that's already a thing in the past so just let it be 😊\n\n-----------------------------------------------------------\n\nOn a brighter note, regardless of the account boosting stuffs expressed above, this has been an amazing competition and a learning experience ^^ Can't wait to read the solutions of top teams (some of them are using GANs, SuperPoint, SuperGlue, and other fancy stuffs :D :D :D)", "1029936": "@chankhavu - Strange user activity has been highlighted in many competition these days. I have been training a model for last 72 hours and my score has just reached 0.4561.. ", "1029726": "" } }