{ "id": 94256, "title": "What is the trick for the top teams break 670", "url": "/competitions/imet-2019-fgvc6/discussion/94256", "author_name": "Strideradu", "post_date": "2019-06-03T14:15:44.304000", "votes": 52, "comment_count": 70, "views": 0, "content": "

It seems suddenly there are 4 teams improve their score to 670+, I don't think the pure image classification can achieve that high, so I am wondering if they found some tricks? And also wondering how 4 people can found a similar trick at the same time?

", "messages": [ { "id": 542156, "postDate": "2019-06-03T14:15:44.303Z", "content": "

It seems suddenly there are 4 teams improve their score to 670+, I don't think the pure image classification can achieve that high, so I am wondering if they found some tricks? And also wondering how 4 people can found a similar trick at the same time?

", "rawMarkdown": "It seems suddenly there are 4 teams improve their score to 670+, I don't think the pure image classification can achieve that high, so I am wondering if they found some tricks? And also wondering how 4 people can found a similar trick at the same time?", "votes": 52 }, { "id": 542209, "postDate": "2019-06-03T15:11:13.117Z", "content": "

Yeah, these people look like this. They appeared suddenly and ran ahead :)\n\"\"

", "rawMarkdown": "Yeah, these people look like this. They appeared suddenly and ran ahead :)\n![](https://www.metmuseum.org/toah/images/h4/h4_14.130.12.jpg)", "votes": 40 }, { "id": 542256, "postDate": "2019-06-03T16:10:45.757Z", "content": "

TOP3 are colleagues in the same company :D\nThey worked together in a previous competition\nThey are ods members\nThey suddenly jumped to top3 in only 24h !\nAnd I see people telling us \"their inspiration happened at the same time\" DAMN !!

", "rawMarkdown": "TOP3 are colleagues in the same company :D\nThey worked together in a previous competition\nThey are ods members\nThey suddenly jumped to top3 in only 24h !\nAnd I see people telling us \"their inspiration happened at the same time\" DAMN !!", "votes": 21, "replies": [ { "id": 542261, "postDate": "2019-06-03T16:14:39.737Z", "content": "
\n

TOP3 are colleagues in the same company :D

\n
\n\n

How do you know that? Not from their Kaggle profiles for sure.

", "rawMarkdown": "> TOP3 are colleagues in the same company :D\n\nHow do you know that? Not from their Kaggle profiles for sure.", "votes": -2 }, { "id": 542265, "postDate": "2019-06-03T16:26:10.210Z", "content": "

I just took a look at their LinkedIn profiles:

\n\n

https://www.linkedin.com/in/ikibardin/\nhttps://www.linkedin.com/in/n01z3/\nhttps://www.linkedin.com/in/venheads/

", "rawMarkdown": "I just took a look at their LinkedIn profiles:\n\nhttps://www.linkedin.com/in/ikibardin/\nhttps://www.linkedin.com/in/n01z3/\nhttps://www.linkedin.com/in/venheads/\n", "votes": 19 }, { "id": 542269, "postDate": "2019-06-03T16:29:10.047Z", "content": "

Yes they are. You could check LinkedIn.

", "rawMarkdown": "Yes they are. You could check LinkedIn.", "votes": 2 }, { "id": 542273, "postDate": "2019-06-03T16:32:08.173Z", "content": "

I constantly monitor ods.ai and there was no information sharing in ods.ai. It's just a chat.

\n\n

But top-3 guys are colleagues in X5 Group, I guess they have shared some key approaches at very least.

", "rawMarkdown": "I constantly monitor ods.ai and there was no information sharing in ods.ai. It's just a chat. \n\nBut top-3 guys are colleagues in X5 Group, I guess they have shared some key approaches at very least.", "votes": 12 }, { "id": 542312, "postDate": "2019-06-03T17:31:27.813Z", "content": "

Ok, Pavel Tsai (currently top 4) studies together with Ilya Kibardin in MIPT. I don't know if he works in X5 or not.

", "rawMarkdown": "Ok, Pavel Tsai (currently top 4) studies together with Ilya Kibardin in MIPT. I don't know if he works in X5 or not.", "votes": 4 } ] }, { "id": 542353, "postDate": "2019-06-03T18:32:55.447Z", "content": "

Truth or not if everyone think private sharing is allowed, it harms the value of kaggle medals and ranks. That's why I think kaggle team should make it clear.

\n\n

Anyway competition will end soon, the only thing we can do now is to keep trying to improve our own score and hope kaggle team will do right things :)

", "rawMarkdown": "Truth or not if everyone think private sharing is allowed, it harms the value of kaggle medals and ranks. That's why I think kaggle team should make it clear.\n\nAnyway competition will end soon, the only thing we can do now is to keep trying to improve our own score and hope kaggle team will do right things :) ", "votes": 15 }, { "id": 542291, "postDate": "2019-06-03T16:57:52.023Z", "content": "

As I remember there were few Chinese went to the top 5 almost on the same days in the early days in Santander competition, they were suspected to have private shares, even they're not from the same organisation. So, here, it's reasonable to suspect they had a private share, given they're from same company and had copule of collaberations in other competitions. I did not say they must have private share, but from stats and probability view thechance of no private share is very low.

", "rawMarkdown": "As I remember there were few Chinese went to the top 5 almost on the same days in the early days in Santander competition, they were suspected to have private shares, even they're not from the same organisation. So, here, it's reasonable to suspect they had a private share, given they're from same company and had copule of collaberations in other competitions. I did not say they must have private share, but from stats and probability view thechance of no private share is very low.", "votes": 17 }, { "id": 542221, "postDate": "2019-06-03T15:25:00.620Z", "content": "

They have one common feature: ods.ai

\n\n

Does it mean that they shared privately? Not necessarily, but they are exposed to the same information via ods.ai, and maybe some of it is very useful here.

", "rawMarkdown": "They have one common feature: ods.ai\n\nDoes it mean that they shared privately? Not necessarily, but they are exposed to the same information via ods.ai, and maybe some of it is very useful here.", "votes": 17, "replies": [ { "id": 542229, "postDate": "2019-06-03T15:34:55.877Z", "content": "

It should be called private sharing, in my understanding.

", "rawMarkdown": "It should be called private sharing, in my understanding.", "votes": 5 }, { "id": 542231, "postDate": "2019-06-03T15:35:34.417Z", "content": "

Another thing in common is that they all have a 0.04 jump if you look into the submission history. Seems that there is a universal approach which apparently no other teams have noticed.

", "rawMarkdown": "Another thing in common is that they all have a 0.04 jump if you look into the submission history. Seems that there is a universal approach which apparently no other teams have noticed. ", "votes": 3 }, { "id": 542242, "postDate": "2019-06-03T15:54:25.857Z", "content": "
\n

It should be called private sharing

\n
\n\n

It depends. If they shared a webinar or a link to a paper that is publicly available elsewhere then it is not private sharing IMHO.

\n\n

I am a member of ods.ai, but given it is all in Russian I don't go there often. If I was not busy in another competition that ends today as well then I'd go there and see what was shared.

\n\n

All I can say is that I wish there was a French speaking community as strong as ods.ai ;)

", "rawMarkdown": "> It should be called private sharing\n\nIt depends. If they shared a webinar or a link to a paper that is publicly available elsewhere then it is not private sharing IMHO.\n\nI am a member of ods.ai, but given it is all in Russian I don't go there often. If I was not busy in another competition that ends today as well then I'd go there and see what was shared.\n\nAll I can say is that I wish there was a French speaking community as strong as ods.ai ;)", "votes": 9 }, { "id": 542243, "postDate": "2019-06-03T15:54:36.563Z", "content": "

@naivelamb It strongly suggests that they got a leak.

", "rawMarkdown": "@naivelamb It strongly suggests that they got a leak." }, { "id": 542247, "postDate": "2019-06-03T15:59:33.297Z", "content": "

My concern is not something from ods.ai, but all 4 people were teammates in some previous competition. And suddenly they all found the trick

", "rawMarkdown": "My concern is not something from ods.ai, but all 4 people were teammates in some previous competition. And suddenly they all found the trick", "votes": 4 }, { "id": 542248, "postDate": "2019-06-03T15:59:40.733Z", "content": "

First of all, there are more than 20 ods.ai members doing this one and there are no \"leak\" I can see so it's not related. Second point is that many ods.ai participants are very strong computer vision researchers/engineers winning most CV kaggle/not-kaggle competitions. Finally ods.ai is completely an open platform you can join.

", "rawMarkdown": "First of all, there are more than 20 ods.ai members doing this one and there are no \"leak\" I can see so it's not related. Second point is that many ods.ai participants are very strong computer vision researchers/engineers winning most CV kaggle/not-kaggle competitions. Finally ods.ai is completely an open platform you can join. ", "votes": 7 }, { "id": 542249, "postDate": "2019-06-03T16:03:18.797Z", "content": "

Why did they suddenly find the trick together at the end? Why aren’t they found one after another during the competition?

", "rawMarkdown": "Why did they suddenly find the trick together at the end? Why aren’t they found one after another during the competition?", "votes": 1 }, { "id": 542250, "postDate": "2019-06-03T16:05:18.473Z", "content": "

Have no idea and I wonder about this latest tricks too.

", "rawMarkdown": "Have no idea and I wonder about this latest tricks too.", "votes": 1 }, { "id": 542251, "postDate": "2019-06-03T16:05:34.837Z", "content": "

@yaroshevskiy I'm happy to hear that there's no leak sharing. thanks.

", "rawMarkdown": "@yaroshevskiy I'm happy to hear that there's no leak sharing. thanks." }, { "id": 542254, "postDate": "2019-06-03T16:08:17.963Z", "content": "
\n

Finally ods.ai is completely an open platform you can join.

\n
\n\n

If you pass the little math puzzle ;)

", "rawMarkdown": "> Finally ods.ai is completely an open platform you can join. \n\nIf you pass the little math puzzle ;)" }, { "id": 542325, "postDate": "2019-06-03T17:53:39.590Z", "content": "
\n

First of all, there are more than 20 ods.ai members doing this one and there are no \"leak\" I can see so it's not related.

\n
\n\n

+1 was keeping an eye on this competition and didn't see any sharing in public channels.

\n\n

I hope we learn what happend soon, after the competition ends.

", "rawMarkdown": "> First of all, there are more than 20 ods.ai members doing this one and there are no \"leak\" I can see so it's not related.\n\n+1 was keeping an eye on this competition and didn't see any sharing in public channels.\n\nI hope we learn what happend soon, after the competition ends.", "votes": 8 }, { "id": 543912, "postDate": "2019-06-04T23:42:09.817Z", "content": "

Just came here to say I'm all for the French community as strong as ods.ai! \nI'm way, waaaay down the leaderboard to be affected by what's going on at the top!

", "rawMarkdown": "Just came here to say I'm all for the French community as strong as ods.ai! \nI'm way, waaaay down the leaderboard to be affected by what's going on at the top!" } ] }, { "id": 542159, "postDate": "2019-06-03T14:17:48.733Z", "content": "

And they all come from ods.ai ...

", "rawMarkdown": "And they all come from ods.ai ... ", "votes": 14, "replies": [ { "id": 542190, "postDate": "2019-06-03T14:59:48.960Z", "rawMarkdown": "", "votes": -48 }, { "id": 542198, "postDate": "2019-06-03T15:03:35.063Z", "content": "

so what?

", "rawMarkdown": "so what?", "votes": 3 }, { "id": 542202, "postDate": "2019-06-03T15:07:17.083Z", "content": "

Well I'm a master and I still suck at machine learning

\n\n

(sad face)

\n\n

(no offense)

", "rawMarkdown": "Well I'm a master and I still suck at machine learning \n\n(sad face)\n\n(no offense)", "votes": 6 }, { "id": 542213, "postDate": "2019-06-03T15:17:27.150Z", "content": "

Ай-ай-ай-ай-ай, что сейчас произошло!

", "rawMarkdown": "Ай-ай-ай-ай-ай, что сейчас произошло!", "votes": 2 }, { "id": 542216, "postDate": "2019-06-03T15:19:19.310Z", "rawMarkdown": "", "votes": -27 }, { "id": 542219, "postDate": "2019-06-03T15:23:22.660Z", "content": "
\n

earhian wrote

\n \n
\n

Ай-ай-ай-ай-ай, что сейчас произошло!

\n
\n
\n\n

你气不气?

", "rawMarkdown": "> **earhian wrote**\n> \n> > Ай-ай-ай-ай-ай, что сейчас произошло! \n\n\n你气不气?", "votes": 2 }, { "id": 542222, "postDate": "2019-06-03T15:25:24.650Z", "content": "

RUA！

\n\n
\n

KALE wrote

\n \n
\n

你气不气?

\n
\n
", "rawMarkdown": "RUA！\n\n> **KALE wrote**\n> \n> > 你气不气?", "votes": 1 }, { "id": 543441, "postDate": "2019-06-04T14:28:01.143Z", "content": "

I feel shame for the master group to have you guys.

", "rawMarkdown": "I feel shame for the master group to have you guys.", "votes": 1 } ] }, { "id": 542172, "postDate": "2019-06-03T14:33:10.600Z", "content": "

It could be that their inspiration happened at the same time.

", "rawMarkdown": "It could be that their inspiration happened at the same time.", "votes": 13, "replies": [ { "id": 542197, "postDate": "2019-06-03T15:03:14.220Z", "content": "

I want the same inspiration!

", "rawMarkdown": "I want the same inspiration!", "votes": 4 }, { "id": 542569, "postDate": "2019-06-04T01:38:24.673Z", "content": "

I want the same inspiration too!

", "rawMarkdown": "I want the same inspiration too!", "votes": 1 }, { "id": 542806, "postDate": "2019-06-04T05:19:06.547Z", "content": "

it's a matter of synchronization! lol! :-p

", "rawMarkdown": "it's a matter of synchronization! lol! :-p" } ] }, { "id": 542194, "postDate": "2019-06-03T15:01:24.067Z", "content": "

What a coincidence🤪

", "rawMarkdown": "What a coincidence🤪", "votes": 12 }, { "id": 542178, "postDate": "2019-06-03T14:43:27.453Z", "content": "

I guess more ods.ai will get inspired before the deadline.

", "rawMarkdown": "I guess more ods.ai will get inspired before the deadline.", "votes": 11, "replies": [ { "id": 542195, "postDate": "2019-06-03T15:01:41.187Z", "content": "

unfortunately no :D

", "rawMarkdown": "unfortunately no :D " } ] }, { "id": 542279, "postDate": "2019-06-03T16:40:14.133Z", "content": "

It is correct. Top 3 of this competition work in the same company, where they have been granted a full access to DGX-1 and 4 dev machines with 4 GTX 1080TI\nThey also have been freed from all activities for this period of time except for Kaggle\nAfter each competition there is a meeting in X5 where they analyzed the result and approach of each member. Given that - no wonder that there is finally a success, given amount of free time and computational power. I did also participate in a couple off this meetings though I don't work at X5

", "rawMarkdown": "It is correct. Top 3 of this competition work in the same company, where they have been granted a full access to DGX-1 and 4 dev machines with 4 GTX 1080TI\nThey also have been freed from all activities for this period of time except for Kaggle\nAfter each competition there is a meeting in X5 where they analyzed the result and approach of each member. Given that - no wonder that there is finally a success, given amount of free time and computational power. I did also participate in a couple off this meetings though I don't work at X5", "votes": 10, "replies": [ { "id": 542289, "postDate": "2019-06-03T16:55:20.277Z", "content": "

DGX-1 has 8xV100

", "rawMarkdown": "DGX-1 has 8xV100", "votes": 4 }, { "id": 542397, "postDate": "2019-06-03T20:36:35.887Z", "content": "

Then it is simple. They used their machines til last day and now they replicate best result with kaggle kernels.

", "rawMarkdown": "Then it is simple. They used their machines til last day and now they replicate best result with kaggle kernels.", "votes": 4 } ] }, { "id": 542188, "postDate": "2019-06-03T14:56:49.377Z", "content": "

Well, I just wish to achieve one silver medal, but it seems hard... : )

", "rawMarkdown": "Well, I just wish to achieve one silver medal, but it seems hard... : )", "votes": 7, "replies": [ { "id": 542215, "postDate": "2019-06-03T15:18:27.080Z", "content": "

Haha !

", "rawMarkdown": "Haha !" }, { "id": 542572, "postDate": "2019-06-04T01:39:26.780Z", "content": "

Dalao carry me please !

", "rawMarkdown": "Dalao carry me please !", "votes": 1 }, { "id": 543339, "postDate": "2019-06-04T13:34:48.767Z", "content": "

Please don't speak Chinglish（hua ji

", "rawMarkdown": "Please don't speak Chinglish（hua ji", "votes": 1 } ] }, { "id": 542262, "postDate": "2019-06-03T16:15:05.663Z", "content": "

This is competiton github\nhttps://github.com/visipedia/imet-fgvcx

\n\n

This Imet competiton data is from The Metropolitan Museum of Art in New York.\nurl is https://www.metmuseum.org/art/collection.\nI guess they use that data..(I don't check site in detail...)

", "rawMarkdown": "This is competiton github\nhttps://github.com/visipedia/imet-fgvcx\n\nThis Imet competiton data is from The Metropolitan Museum of Art in New York.\nurl is https://www.metmuseum.org/art/collection.\nI guess they use that data..(I don't check site in detail...)", "votes": 8, "replies": [ { "id": 542320, "postDate": "2019-06-03T17:45:19.213Z", "content": "

If this is the case, it violates the rule. Only whitelist data is allowed and clearly no one asked whether the data is on the whitelist in the organizer's thread.

", "rawMarkdown": "If this is the case, it violates the rule. Only whitelist data is allowed and clearly no one asked whether the data is on the whitelist in the organizer's thread.", "votes": 1 }, { "id": 542440, "postDate": "2019-06-03T23:37:02.203Z", "content": "

Thank for information. So maybe they don't use that.

", "rawMarkdown": "Thank for information. So maybe they don't use that.", "votes": 1 } ] }, { "id": 543387, "postDate": "2019-06-04T14:03:05.997Z", "content": "

I found something funny \"brightspot\"

", "rawMarkdown": "I found something funny ![brightspot](https://storage.googleapis.com/kagglesdsdata/datasets/216969/470346/Inked3HKR3H50G%2905893V94L_LI.jpg?GoogleAccessId=web-data@kaggle-161607.iam.gserviceaccount.com&Expires=1559915851&Signature=nGG5PxAPnx5SRNlPrkM%2ByYUEHRPSQ5Vv8umPPpqiOFf34FAX2FmwVRPrC8IKf1F66AoZud6X5Xoh8kRhnhlphCFC3n6yhj6soC%2B1B4Ybxh%2BGZBASmw8NLjzdh2DEDwSYvlKZPkzZ%2B6GkTxe44rjhl4gBUzOx2x7zhF%2B%2BF3hpArkSNQoZl438vpZ9GlhtuymMzicBG5YYbCk%2BjqMWh7du44ODHeGTaRNAoKtq0Q875xK343H5AZpgEucMHkibbKWtSLTq4Ivv9lQjt33Mw1ij0lfAZebpK%2F7Jt%2FbvYcOgn7hHyFTWFQ57v0RTnS3bBgfTgivHcuzCX1vCXPtKB9OJOQ%3D%3D)", "votes": 5 }, { "id": 542677, "postDate": "2019-06-04T03:25:54.317Z", "content": "

hopefully, those top team/s will share their solution approaches and also code (if possible). Well, I do believe that most of the ods.ai community members have a very strong skill set in deep learning (CV, NLP, etc).

", "rawMarkdown": "hopefully, those top team/s will share their solution approaches and also code (if possible). Well, I do believe that most of the ods.ai community members have a very strong skill set in deep learning (CV, NLP, etc).", "votes": 4 }, { "id": 543346, "postDate": "2019-06-04T13:40:20.070Z", "content": "

I'm falling off the silver medal list! It was a big blow to me as a rookie.

", "rawMarkdown": "I'm falling off the silver medal list! It was a big blow to me as a rookie.", "votes": 1, "replies": [ { "id": 543924, "postDate": "2019-06-05T00:11:43.763Z", "content": "

Shake up or shake down always happens in Kaggle, thus don’t worry: )

", "rawMarkdown": "Shake up or shake down always happens in Kaggle, thus don’t worry: )", "votes": 3 }, { "id": 543971, "postDate": "2019-06-05T01:53:17.893Z", "content": "

Better than me, I am nearly out of bronze!

", "rawMarkdown": "Better than me, I am nearly out of bronze!" } ] }, { "id": 542659, "postDate": "2019-06-04T03:01:07.603Z", "content": "

The best LB score comes to 0.724, amazing!

", "rawMarkdown": "The best LB score comes to 0.724, amazing!", "votes": 1, "replies": [ { "id": 542737, "postDate": "2019-06-04T04:16:11.183Z", "content": "

OMG...!

", "rawMarkdown": "OMG...!", "votes": 1 }, { "id": 543037, "postDate": "2019-06-04T09:42:35.030Z", "content": "

Any way, results higher than 0.7 are fantastic. Probably they were able to train very deep networks locally, will be waiting for exact details

", "rawMarkdown": "Any way, results higher than 0.7 are fantastic. Probably they were able to train very deep networks locally, will be waiting for exact details" }, { "id": 543051, "postDate": "2019-06-04T09:52:23.100Z", "content": "

Anyway, let those toppers show their work after the competition reachs the end 🥇 🥈 🥉

", "rawMarkdown": "Anyway, let those toppers show their work after the competition reachs the end 🥇 🥈 🥉 ", "votes": 1 }, { "id": 543216, "postDate": "2019-06-04T11:39:38.330Z", "content": "

They don't need to show their solutions publicly, only need to show to kaggle and organizer.

", "rawMarkdown": "They don't need to show their solutions publicly, only need to show to kaggle and organizer. " }, { "id": 543323, "postDate": "2019-06-04T13:22:55.923Z", "content": "

@demonplus No, I don't think pure image classification can achieve that high, other teams also have experienced alchemist and tens of Nvidia GPUS

", "rawMarkdown": "@demonplus No, I don't think pure image classification can achieve that high, other teams also have experienced alchemist and tens of Nvidia GPUS" }, { "id": 543361, "postDate": "2019-06-04T13:50:41.770Z", "content": "

@strideradu so you think they found some leak and shared it privately? Or just found the leak simultaneously which is less probable but possible?

", "rawMarkdown": "@strideradu so you think they found some leak and shared it privately? Or just found the leak simultaneously which is less probable but possible?" }, { "id": 543370, "postDate": "2019-06-04T13:56:43.640Z", "content": "

See discussion here

", "rawMarkdown": "See discussion [here](https://www.kaggle.com/c/imet-2019-fgvc6/discussion/94430#latest-543359)" }, { "id": 543455, "postDate": "2019-06-04T14:33:20.593Z", "content": "

Interesting, BTW, what does * 'I got this info during conversation with one of the current top-5 at the leaderboard.'* mean?

", "rawMarkdown": "Interesting, BTW, what does * 'I got this info during conversation with one of the current top-5 at the leaderboard.'* mean?" } ] }, { "id": 542608, "postDate": "2019-06-04T02:04:52.067Z", "content": "

Another 690... And suddenly jump to 690

", "rawMarkdown": "Another 690... And suddenly jump to 690", "votes": 1, "replies": [ { "id": 542613, "postDate": "2019-06-04T02:07:21.160Z", "content": "

615 -> 690 in 6 mins. Looks like post processing/leak to me.

", "rawMarkdown": "615 -> 690 in 6 mins. Looks like post processing/leak to me. ", "votes": 2 } ] }, { "id": 544203, "postDate": "2019-06-05T09:05:22.800Z", "content": "

May the unknown dark magic from current top rank

", "rawMarkdown": "May the unknown dark magic from current top rank" }, { "id": 543454, "postDate": "2019-06-04T14:33:18.227Z", "content": "

Up to now, they have four teams in the top ranking。OMG！

", "rawMarkdown": "Up to now, they have four teams in the top ranking。OMG！" }, { "id": 542771, "postDate": "2019-06-04T04:47:05.227Z", "content": "

All of a sudden, good grades?(*Ӧ)σ

", "rawMarkdown": "All of a sudden, good grades?(*Ӧ)σ" }, { "id": 542734, "postDate": "2019-06-04T04:15:14.183Z", "rawMarkdown": "", "votes": 4, "isDeleted": true }, { "id": 542631, "postDate": "2019-06-04T02:28:55.360Z", "rawMarkdown": "", "votes": -2, "isDeleted": true }, { "id": 542173, "postDate": "2019-06-03T14:34:34.017Z", "rawMarkdown": "", "votes": 17, "isDeleted": true } ], "comments": [ { "id": 542209, "author_name": "dhaqui the kaggler", "author_url": "", "post_date": "2019-06-03T15:11:13.117000", "content": "

Yeah, these people look like this. They appeared suddenly and ran ahead :)\n\"\"

", "votes": 40, "replies": [] }, { "id": 542256, "author_name": "Firas Baba", "author_url": "", "post_date": "2019-06-03T16:10:45.757000", "content": "

TOP3 are colleagues in the same company :D\nThey worked together in a previous competition\nThey are ods members\nThey suddenly jumped to top3 in only 24h !\nAnd I see people telling us \"their inspiration happened at the same time\" DAMN !!

", "votes": 21, "replies": [ { "id": 542261, "author_name": "CPMP", "author_url": "", "post_date": "2019-06-03T16:14:39.737000", "content": "
\n

TOP3 are colleagues in the same company :D

\n
\n\n

How do you know that? Not from their Kaggle profiles for sure.

", "votes": -2, "replies": [] }, { "id": 542265, "author_name": "Yiheng Wang", "author_url": "", "post_date": "2019-06-03T16:26:10.210000", "content": "

I just took a look at their LinkedIn profiles:

\n\n

https://www.linkedin.com/in/ikibardin/\nhttps://www.linkedin.com/in/n01z3/\nhttps://www.linkedin.com/in/venheads/

", "votes": 19, "replies": [] }, { "id": 542269, "author_name": "Artyom Palvelev", "author_url": "", "post_date": "2019-06-03T16:29:10.047000", "content": "

Yes they are. You could check LinkedIn.

", "votes": 2, "replies": [] }, { "id": 542273, "author_name": "Artyom Palvelev", "author_url": "", "post_date": "2019-06-03T16:32:08.173000", "content": "

I constantly monitor ods.ai and there was no information sharing in ods.ai. It's just a chat.

\n\n

But top-3 guys are colleagues in X5 Group, I guess they have shared some key approaches at very least.

", "votes": 12, "replies": [] }, { "id": 542312, "author_name": "Artyom Palvelev", "author_url": "", "post_date": "2019-06-03T17:31:27.813000", "content": "

Ok, Pavel Tsai (currently top 4) studies together with Ilya Kibardin in MIPT. I don't know if he works in X5 or not.

", "votes": 4, "replies": [] } ] }, { "id": 542353, "author_name": "Camaro", "author_url": "", "post_date": "2019-06-03T18:32:55.447000", "content": "

Truth or not if everyone think private sharing is allowed, it harms the value of kaggle medals and ranks. That's why I think kaggle team should make it clear.

\n\n

Anyway competition will end soon, the only thing we can do now is to keep trying to improve our own score and hope kaggle team will do right things :)

", "votes": 15, "replies": [] }, { "id": 542291, "author_name": "Jie Wu", "author_url": "", "post_date": "2019-06-03T16:57:52.023000", "content": "

As I remember there were few Chinese went to the top 5 almost on the same days in the early days in Santander competition, they were suspected to have private shares, even they're not from the same organisation. So, here, it's reasonable to suspect they had a private share, given they're from same company and had copule of collaberations in other competitions. I did not say they must have private share, but from stats and probability view thechance of no private share is very low.

", "votes": 17, "replies": [] }, { "id": 542221, "author_name": "CPMP", "author_url": "", "post_date": "2019-06-03T15:25:00.620000", "content": "

They have one common feature: ods.ai

\n\n

Does it mean that they shared privately? Not necessarily, but they are exposed to the same information via ods.ai, and maybe some of it is very useful here.

", "votes": 17, "replies": [ { "id": 542229, "author_name": "OsciiArt", "author_url": "", "post_date": "2019-06-03T15:34:55.877000", "content": "

It should be called private sharing, in my understanding.

", "votes": 5, "replies": [] }, { "id": 542231, "author_name": "Xuan Cao", "author_url": "", "post_date": "2019-06-03T15:35:34.417000", "content": "

Another thing in common is that they all have a 0.04 jump if you look into the submission history. Seems that there is a universal approach which apparently no other teams have noticed.

", "votes": 3, "replies": [] }, { "id": 542242, "author_name": "CPMP", "author_url": "", "post_date": "2019-06-03T15:54:25.857000", "content": "
\n

It should be called private sharing

\n
\n\n

It depends. If they shared a webinar or a link to a paper that is publicly available elsewhere then it is not private sharing IMHO.

\n\n

I am a member of ods.ai, but given it is all in Russian I don't go there often. If I was not busy in another competition that ends today as well then I'd go there and see what was shared.

\n\n

All I can say is that I wish there was a French speaking community as strong as ods.ai ;)

", "votes": 9, "replies": [] }, { "id": 542243, "author_name": "OsciiArt", "author_url": "", "post_date": "2019-06-03T15:54:36.563000", "content": "

@naivelamb It strongly suggests that they got a leak.

", "votes": 0, "replies": [] }, { "id": 542247, "author_name": "Strideradu", "author_url": "", "post_date": "2019-06-03T15:59:33.297000", "content": "

My concern is not something from ods.ai, but all 4 people were teammates in some previous competition. And suddenly they all found the trick

", "votes": 4, "replies": [] }, { "id": 542248, "author_name": "Oleg Yaroshevskiy", "author_url": "", "post_date": "2019-06-03T15:59:40.733000", "content": "

First of all, there are more than 20 ods.ai members doing this one and there are no \"leak\" I can see so it's not related. Second point is that many ods.ai participants are very strong computer vision researchers/engineers winning most CV kaggle/not-kaggle competitions. Finally ods.ai is completely an open platform you can join.

", "votes": 7, "replies": [] }, { "id": 542249, "author_name": "qrfaction", "author_url": "", "post_date": "2019-06-03T16:03:18.797000", "content": "

Why did they suddenly find the trick together at the end? Why aren’t they found one after another during the competition?

", "votes": 1, "replies": [] }, { "id": 542250, "author_name": "Oleg Yaroshevskiy", "author_url": "", "post_date": "2019-06-03T16:05:18.473000", "content": "

Have no idea and I wonder about this latest tricks too.

", "votes": 1, "replies": [] }, { "id": 542251, "author_name": "OsciiArt", "author_url": "", "post_date": "2019-06-03T16:05:34.837000", "content": "

@yaroshevskiy I'm happy to hear that there's no leak sharing. thanks.

", "votes": 0, "replies": [] }, { "id": 542254, "author_name": "CPMP", "author_url": "", "post_date": "2019-06-03T16:08:17.963000", "content": "
\n

Finally ods.ai is completely an open platform you can join.

\n
\n\n

If you pass the little math puzzle ;)

", "votes": 0, "replies": [] }, { "id": 542325, "author_name": "Konstantin Lopukhin", "author_url": "", "post_date": "2019-06-03T17:53:39.590000", "content": "
\n

First of all, there are more than 20 ods.ai members doing this one and there are no \"leak\" I can see so it's not related.

\n
\n\n

+1 was keeping an eye on this competition and didn't see any sharing in public channels.

\n\n

I hope we learn what happend soon, after the competition ends.

", "votes": 8, "replies": [] }, { "id": 543912, "author_name": "Maxime Lenormand", "author_url": "", "post_date": "2019-06-04T23:42:09.817000", "content": "

Just came here to say I'm all for the French community as strong as ods.ai! \nI'm way, waaaay down the leaderboard to be affected by what's going on at the top!

", "votes": 0, "replies": [] } ] }, { "id": 542159, "author_name": "qrfaction", "author_url": "", "post_date": "2019-06-03T14:17:48.733000", "content": "

And they all come from ods.ai ...

", "votes": 14, "replies": [ { "id": 542190, "author_name": "ValeriyBabushkin", "author_url": "", "post_date": "2019-06-03T14:59:48.960000", "content": "", "votes": -48, "replies": [] }, { "id": 542198, "author_name": "qrfaction", "author_url": "", "post_date": "2019-06-03T15:03:35.063000", "content": "

so what?

", "votes": 3, "replies": [] }, { "id": 542202, "author_name": "Peiyuan Liao", "author_url": "", "post_date": "2019-06-03T15:07:17.083000", "content": "

Well I'm a master and I still suck at machine learning

\n\n

(sad face)

\n\n

(no offense)

", "votes": 6, "replies": [] }, { "id": 542213, "author_name": "earhian", "author_url": "", "post_date": "2019-06-03T15:17:27.150000", "content": "

Ай-ай-ай-ай-ай, что сейчас произошло!

", "votes": 2, "replies": [] }, { "id": 542216, "author_name": "ValeriyBabushkin", "author_url": "", "post_date": "2019-06-03T15:19:19.310000", "content": "", "votes": -27, "replies": [] }, { "id": 542219, "author_name": "KALE", "author_url": "", "post_date": "2019-06-03T15:23:22.660000", "content": "
\n

earhian wrote

\n \n
\n

Ай-ай-ай-ай-ай, что сейчас произошло!

\n
\n
\n\n

你气不气?

", "votes": 2, "replies": [] }, { "id": 542222, "author_name": "YK", "author_url": "", "post_date": "2019-06-03T15:25:24.650000", "content": "

RUA！

\n\n
\n

KALE wrote

\n \n
\n

你气不气?

\n
\n
", "votes": 1, "replies": [] }, { "id": 543441, "author_name": "qrfaction", "author_url": "", "post_date": "2019-06-04T14:28:01.143000", "content": "

I feel shame for the master group to have you guys.

", "votes": 1, "replies": [] } ] }, { "id": 542172, "author_name": "Gary", "author_url": "", "post_date": "2019-06-03T14:33:10.600000", "content": "

It could be that their inspiration happened at the same time.

", "votes": 13, "replies": [ { "id": 542197, "author_name": "Strideradu", "author_url": "", "post_date": "2019-06-03T15:03:14.220000", "content": "

I want the same inspiration!

", "votes": 4, "replies": [] }, { "id": 542569, "author_name": "seefun", "author_url": "", "post_date": "2019-06-04T01:38:24.673000", "content": "

I want the same inspiration too!

", "votes": 1, "replies": [] }, { "id": 542806, "author_name": "FGPC", "author_url": "", "post_date": "2019-06-04T05:19:06.547000", "content": "

it's a matter of synchronization! lol! :-p

", "votes": 0, "replies": [] } ] }, { "id": 542194, "author_name": "wh1te", "author_url": "", "post_date": "2019-06-03T15:01:24.067000", "content": "

What a coincidence🤪

", "votes": 12, "replies": [] }, { "id": 542178, "author_name": "YK", "author_url": "", "post_date": "2019-06-03T14:43:27.453000", "content": "

I guess more ods.ai will get inspired before the deadline.

", "votes": 11, "replies": [ { "id": 542195, "author_name": "Oleg Yaroshevskiy", "author_url": "", "post_date": "2019-06-03T15:01:41.187000", "content": "

unfortunately no :D

", "votes": 0, "replies": [] } ] }, { "id": 542279, "author_name": "Nerses Bagiyan", "author_url": "", "post_date": "2019-06-03T16:40:14.133000", "content": "

It is correct. Top 3 of this competition work in the same company, where they have been granted a full access to DGX-1 and 4 dev machines with 4 GTX 1080TI\nThey also have been freed from all activities for this period of time except for Kaggle\nAfter each competition there is a meeting in X5 where they analyzed the result and approach of each member. Given that - no wonder that there is finally a success, given amount of free time and computational power. I did also participate in a couple off this meetings though I don't work at X5

", "votes": 10, "replies": [ { "id": 542289, "author_name": "Artyom Palvelev", "author_url": "", "post_date": "2019-06-03T16:55:20.277000", "content": "

DGX-1 has 8xV100

", "votes": 4, "replies": [] }, { "id": 542397, "author_name": "CPMP", "author_url": "", "post_date": "2019-06-03T20:36:35.887000", "content": "

Then it is simple. They used their machines til last day and now they replicate best result with kaggle kernels.

", "votes": 4, "replies": [] } ] }, { "id": 542188, "author_name": "Yiheng Wang", "author_url": "", "post_date": "2019-06-03T14:56:49.377000", "content": "

Well, I just wish to achieve one silver medal, but it seems hard... : )

", "votes": 7, "replies": [ { "id": 542215, "author_name": "Filemon", "author_url": "", "post_date": "2019-06-03T15:18:27.080000", "content": "

Haha !

", "votes": 0, "replies": [] }, { "id": 542572, "author_name": "seefun", "author_url": "", "post_date": "2019-06-04T01:39:26.780000", "content": "

Dalao carry me please !

", "votes": 1, "replies": [] }, { "id": 543339, "author_name": "", "author_url": "", "post_date": "2019-06-04T13:34:48.767000", "content": "

Please don't speak Chinglish（hua ji

", "votes": 1, "replies": [] } ] }, { "id": 542262, "author_name": "tereka", "author_url": "", "post_date": "2019-06-03T16:15:05.663000", "content": "

This is competiton github\nhttps://github.com/visipedia/imet-fgvcx

\n\n

This Imet competiton data is from The Metropolitan Museum of Art in New York.\nurl is https://www.metmuseum.org/art/collection.\nI guess they use that data..(I don't check site in detail...)

", "votes": 8, "replies": [ { "id": 542320, "author_name": "Xuan Cao", "author_url": "", "post_date": "2019-06-03T17:45:19.213000", "content": "

If this is the case, it violates the rule. Only whitelist data is allowed and clearly no one asked whether the data is on the whitelist in the organizer's thread.

", "votes": 1, "replies": [] }, { "id": 542440, "author_name": "tereka", "author_url": "", "post_date": "2019-06-03T23:37:02.203000", "content": "

Thank for information. So maybe they don't use that.

", "votes": 1, "replies": [] } ] }, { "id": 543387, "author_name": "", "author_url": "", "post_date": "2019-06-04T14:03:05.997000", "content": "

I found something funny \"brightspot\"

", "votes": 5, "replies": [] }, { "id": 542677, "author_name": "FGPC", "author_url": "", "post_date": "2019-06-04T03:25:54.317000", "content": "

hopefully, those top team/s will share their solution approaches and also code (if possible). Well, I do believe that most of the ods.ai community members have a very strong skill set in deep learning (CV, NLP, etc).

", "votes": 4, "replies": [] }, { "id": 543346, "author_name": "", "author_url": "", "post_date": "2019-06-04T13:40:20.070000", "content": "

I'm falling off the silver medal list! It was a big blow to me as a rookie.

", "votes": 1, "replies": [ { "id": 543924, "author_name": "Yiheng Wang", "author_url": "", "post_date": "2019-06-05T00:11:43.763000", "content": "

Shake up or shake down always happens in Kaggle, thus don’t worry: )

", "votes": 3, "replies": [] }, { "id": 543971, "author_name": "Dilapsky Lee", "author_url": "", "post_date": "2019-06-05T01:53:17.893000", "content": "

Better than me, I am nearly out of bronze!

", "votes": 0, "replies": [] } ] }, { "id": 542659, "author_name": "seefun", "author_url": "", "post_date": "2019-06-04T03:01:07.603000", "content": "

The best LB score comes to 0.724, amazing!

", "votes": 1, "replies": [ { "id": 542737, "author_name": "Tawara", "author_url": "", "post_date": "2019-06-04T04:16:11.183000", "content": "

OMG...!

", "votes": 1, "replies": [] }, { "id": 543037, "author_name": "Anna Novikova", "author_url": "", "post_date": "2019-06-04T09:42:35.030000", "content": "

Any way, results higher than 0.7 are fantastic. Probably they were able to train very deep networks locally, will be waiting for exact details

", "votes": 0, "replies": [] }, { "id": 543051, "author_name": "Dilapsky Lee", "author_url": "", "post_date": "2019-06-04T09:52:23.100000", "content": "

Anyway, let those toppers show their work after the competition reachs the end 🥇 🥈 🥉

", "votes": 1, "replies": [] }, { "id": 543216, "author_name": "Xuan Cao", "author_url": "", "post_date": "2019-06-04T11:39:38.330000", "content": "

They don't need to show their solutions publicly, only need to show to kaggle and organizer.

", "votes": 0, "replies": [] }, { "id": 543323, "author_name": "Strideradu", "author_url": "", "post_date": "2019-06-04T13:22:55.923000", "content": "

@demonplus No, I don't think pure image classification can achieve that high, other teams also have experienced alchemist and tens of Nvidia GPUS

", "votes": 0, "replies": [] }, { "id": 543361, "author_name": "Anna Novikova", "author_url": "", "post_date": "2019-06-04T13:50:41.770000", "content": "

@strideradu so you think they found some leak and shared it privately? Or just found the leak simultaneously which is less probable but possible?

", "votes": 0, "replies": [] }, { "id": 543370, "author_name": "Xuan Cao", "author_url": "", "post_date": "2019-06-04T13:56:43.640000", "content": "

See discussion here

", "votes": 0, "replies": [] }, { "id": 543455, "author_name": "Jie Wu", "author_url": "", "post_date": "2019-06-04T14:33:20.593000", "content": "

Interesting, BTW, what does * 'I got this info during conversation with one of the current top-5 at the leaderboard.'* mean?

", "votes": 0, "replies": [] } ] }, { "id": 542608, "author_name": "Strideradu", "author_url": "", "post_date": "2019-06-04T02:04:52.067000", "content": "

Another 690... And suddenly jump to 690

", "votes": 1, "replies": [ { "id": 542613, "author_name": "Xuan Cao", "author_url": "", "post_date": "2019-06-04T02:07:21.160000", "content": "

615 -> 690 in 6 mins. Looks like post processing/leak to me.

", "votes": 2, "replies": [] } ] }, { "id": 544203, "author_name": "good good study", "author_url": "", "post_date": "2019-06-05T09:05:22.800000", "content": "

May the unknown dark magic from current top rank

", "votes": 0, "replies": [] }, { "id": 543454, "author_name": "", "author_url": "", "post_date": "2019-06-04T14:33:18.227000", "content": "

Up to now, they have four teams in the top ranking。OMG！

", "votes": 0, "replies": [] }, { "id": 542771, "author_name": "", "author_url": "", "post_date": "2019-06-04T04:47:05.227000", "content": "

All of a sudden, good grades?(*Ӧ)σ

", "votes": 0, "replies": [] }, { "id": 542734, "author_name": "", "author_url": "", "post_date": "2019-06-04T04:15:14.183000", "content": "", "votes": 4, "replies": [] }, { "id": 542631, "author_name": "", "author_url": "", "post_date": "2019-06-04T02:28:55.360000", "content": "", "votes": -2, "replies": [] }, { "id": 542173, "author_name": "", "author_url": "", "post_date": "2019-06-03T14:34:34.017000", "content": "", "votes": 17, "replies": [] } ], "raw_markdown_by_id": { "542156": "It seems suddenly there are 4 teams improve their score to 670+, I don't think the pure image classification can achieve that high, so I am wondering if they found some tricks? And also wondering how 4 people can found a similar trick at the same time?", "542209": "Yeah, these people look like this. They appeared suddenly and ran ahead :)\n![](https://www.metmuseum.org/toah/images/h4/h4_14.130.12.jpg)", "542256": "TOP3 are colleagues in the same company :D\nThey worked together in a previous competition\nThey are ods members\nThey suddenly jumped to top3 in only 24h !\nAnd I see people telling us \"their inspiration happened at the same time\" DAMN !!", "542353": "Truth or not if everyone think private sharing is allowed, it harms the value of kaggle medals and ranks. That's why I think kaggle team should make it clear.\n\nAnyway competition will end soon, the only thing we can do now is to keep trying to improve our own score and hope kaggle team will do right things :) ", "542291": "As I remember there were few Chinese went to the top 5 almost on the same days in the early days in Santander competition, they were suspected to have private shares, even they're not from the same organisation. So, here, it's reasonable to suspect they had a private share, given they're from same company and had copule of collaberations in other competitions. I did not say they must have private share, but from stats and probability view thechance of no private share is very low.", "542221": "They have one common feature: ods.ai\n\nDoes it mean that they shared privately? Not necessarily, but they are exposed to the same information via ods.ai, and maybe some of it is very useful here.", "542159": "And they all come from ods.ai ... ", "542172": "It could be that their inspiration happened at the same time.", "542194": "What a coincidence🤪", "542178": "I guess more ods.ai will get inspired before the deadline.", "542279": "It is correct. Top 3 of this competition work in the same company, where they have been granted a full access to DGX-1 and 4 dev machines with 4 GTX 1080TI\nThey also have been freed from all activities for this period of time except for Kaggle\nAfter each competition there is a meeting in X5 where they analyzed the result and approach of each member. Given that - no wonder that there is finally a success, given amount of free time and computational power. I did also participate in a couple off this meetings though I don't work at X5", "542188": "Well, I just wish to achieve one silver medal, but it seems hard... : )", "542262": "This is competiton github\nhttps://github.com/visipedia/imet-fgvcx\n\nThis Imet competiton data is from The Metropolitan Museum of Art in New York.\nurl is https://www.metmuseum.org/art/collection.\nI guess they use that data..(I don't check site in detail...)", "543387": "I found something funny ![brightspot](https://storage.googleapis.com/kagglesdsdata/datasets/216969/470346/Inked3HKR3H50G%2905893V94L_LI.jpg?GoogleAccessId=web-data@kaggle-161607.iam.gserviceaccount.com&Expires=1559915851&Signature=nGG5PxAPnx5SRNlPrkM%2ByYUEHRPSQ5Vv8umPPpqiOFf34FAX2FmwVRPrC8IKf1F66AoZud6X5Xoh8kRhnhlphCFC3n6yhj6soC%2B1B4Ybxh%2BGZBASmw8NLjzdh2DEDwSYvlKZPkzZ%2B6GkTxe44rjhl4gBUzOx2x7zhF%2B%2BF3hpArkSNQoZl438vpZ9GlhtuymMzicBG5YYbCk%2BjqMWh7du44ODHeGTaRNAoKtq0Q875xK343H5AZpgEucMHkibbKWtSLTq4Ivv9lQjt33Mw1ij0lfAZebpK%2F7Jt%2FbvYcOgn7hHyFTWFQ57v0RTnS3bBgfTgivHcuzCX1vCXPtKB9OJOQ%3D%3D)", "542677": "hopefully, those top team/s will share their solution approaches and also code (if possible). Well, I do believe that most of the ods.ai community members have a very strong skill set in deep learning (CV, NLP, etc).", "543346": "I'm falling off the silver medal list! It was a big blow to me as a rookie.", "542659": "The best LB score comes to 0.724, amazing!", "542608": "Another 690... And suddenly jump to 690", "544203": "May the unknown dark magic from current top rank", "543454": "Up to now, they have four teams in the top ranking。OMG！", "542771": "All of a sudden, good grades?(*Ӧ)σ", "542734": "", "542631": "", "542173": "" } }