![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F113660%2F5261dadd78d259ed20a96a7bf0e2a1bf%2FSelection_214.png?generation=1594172033051804&alt=media\")
Hello All
\n\nI have just joined this competition and have no prior experience in Yolo . Please forgive me if this question is already deliberated .\nI have noticed that the licensing term of Yolov5 and Yolo V4/V3 are little different . V5 license is GPL3.0 I guess which is bit more restrictive than the other opensource licenses . Are competitors allowed to use it in final solution ? (Especially the ones going to be in prize zone )
\n\nTagging hosts : @philculliton @bendvd @maggiemd @stavness
", "messages": [ { "id": 911829, "postDate": "2020-07-02T03:54:18.013Z", "content": "Hello All
\n\nI have just joined this competition and have no prior experience in Yolo . Please forgive me if this question is already deliberated .\nI have noticed that the licensing term of Yolov5 and Yolo V4/V3 are little different . V5 license is GPL3.0 I guess which is bit more restrictive than the other opensource licenses . Are competitors allowed to use it in final solution ? (Especially the ones going to be in prize zone )
\n\nTagging hosts : @philculliton @bendvd @maggiemd @stavness
", "rawMarkdown": "Hello All \n\nI have just joined this competition and have no prior experience in Yolo . Please forgive me if this question is already deliberated .\nI have noticed that the licensing term of Yolov5 and Yolo V4/V3 are little different . V5 license is GPL3.0 I guess which is bit more restrictive than the other opensource licenses . Are competitors allowed to use it in final solution ? (Especially the ones going to be in prize zone )\n\nTagging hosts : @philculliton @bendvd @maggiemd @stavness", "votes": 52 }, { "id": 920645, "postDate": "2020-07-08T18:19:42.740Z", "content": "All,
\n\nThis Global Wheat Detection competition, and this thread in particular were recently brought to my attention here https://github.com/ultralytics/yolov5/issues/317.
\n\nI'm very happy that the community is finding YOLOv5 useful!! I personally have not participated in any Kaggle competitions, but I did participate in a National Geospatial-Intelligence Agency (NGA) competition that got me interested in YOLO and propelled me down my ML path a couple summers ago. I created a repo at the time to document my efforts:\nhttps://github.com/ultralytics/xview-yolov3
\n\nIt's a nice irony now to see the situation reversed, with other young people learning and using YOLOv5 to start their own research and competition efforts. I'm very humbled and hopeful that smart minds might use what we have developed as the basis for further advancements of their own in the future :)
\n\nRegarding the licensing issues, both of our YOLO repositories, https://github.com/ultralytics/yolov3 and https://github.com/ultralytics/yolov5 are licensed as GPL 3.0. The intention of the license is allow full use for any purpose (academic, research or commercial) of the original or modified code, provided that derivative works are open sourced and licensed as the same. The greater good we have in mind here is the capability of all people to benefit from advancements made with future open-sourced works that begin from or leverage YOLOv5 as a starting point.
\n\nUltralytics was founded to work on particle physics applications, we sleep with one eye open towards expanding knowledge and advancing science. We are always open to new ideas, but we believe we have the right license in place for our work. Isaac Newton once said \"If I have seen further it is by standing on the shoulders of Giants.\" I believe that if we allow future works to hide from the world as proprietary code, which other licenses provide provisions for, we are reducing the possible giants that future Isaac Newtons may stand on, and diminishing all of our own futures slightly so.
", "rawMarkdown": "All, \n\nThis Global Wheat Detection competition, and this thread in particular were recently brought to my attention here https://github.com/ultralytics/yolov5/issues/317.\n\nI'm very happy that the community is finding YOLOv5 useful!! I personally have not participated in any Kaggle competitions, but I did participate in a National Geospatial-Intelligence Agency (NGA) competition that got me interested in YOLO and propelled me down my ML path a couple summers ago. I created a repo at the time to document my efforts:\nhttps://github.com/ultralytics/xview-yolov3\n\nIt's a nice irony now to see the situation reversed, with other young people learning and using YOLOv5 to start their own research and competition efforts. I'm very humbled and hopeful that smart minds might use what we have developed as the basis for further advancements of their own in the future :)\n\nRegarding the licensing issues, both of our YOLO repositories, https://github.com/ultralytics/yolov3 and https://github.com/ultralytics/yolov5 are licensed as GPL 3.0. The intention of the license is allow full use for any purpose (academic, research or commercial) of the original or modified code, provided that derivative works are open sourced and licensed as the same. The greater good we have in mind here is the capability of all people to benefit from advancements made with future open-sourced works that begin from or leverage YOLOv5 as a starting point.\n\nUltralytics was founded to work on particle physics applications, we sleep with one eye open towards expanding knowledge and advancing science. We are always open to new ideas, but we believe we have the right license in place for our work. Isaac Newton once said **\"If I have seen further it is by standing on the shoulders of Giants.\"** I believe that if we allow future works to hide from the world as proprietary code, which other licenses provide provisions for, we are reducing the possible giants that future Isaac Newtons may stand on, and diminishing all of our own futures slightly so.\n\n", "votes": 33 }, { "id": 919568, "postDate": "2020-07-08T01:34:09.293Z", "content": "kaggle or someone needs to to make a checking web page like this:
\n\n
welcome board, man. π€
", "rawMarkdown": "welcome board, man. π€ ", "votes": -2 }, { "id": 919573, "postDate": "2020-07-08T01:39:17.310Z", "content": "Great to see you here, @hengck23! And I do like that solution, although I don't expect it'll actually get implemented. Just curious, did you make this \"mock-up\" yourself or is it actually a form somewhere?
", "rawMarkdown": "Great to see you here, @hengck23! And I do like that solution, although I don't expect it'll actually get implemented. Just curious, did you make this \"mock-up\" yourself or is it actually a form somewhere?", "votes": -2 }, { "id": 919580, "postDate": "2020-07-08T01:47:20.567Z", "content": "it is just a mock up in PPT. someone can start web service and charged e.g. $1.00 per check or have it free.
\n\nActually in R&D company i worked for, it is a common practice that in our solution (software prototype, model, dateset) we gave to client, we need to include licensing for any external 3rd-party, open-source solution we used.
\n\nWhen there is money involved, we need to be cautious in company.
\n\nWith more and more people and big company are involved in kaggle competition, kagglers needs to be more aware of licensing. It would be troublesome and sometimes kill the joy of kaggling ... but it is unavoidable when kaggle grows large.
", "rawMarkdown": "it is just a mock up in PPT. someone can start web service and charged e.g. $1.00 per check or have it free.\n\nActually in R&D company i worked for, it is a common practice that in our solution (software prototype, model, dateset) we gave to client, we need to include licensing for any external 3rd-party, open-source solution we used.\n\nWhen there is money involved, we need to be cautious in company.\n\n---\n\nWith more and more people and big company are involved in kaggle competition, kagglers needs to be more aware of licensing. It would be troublesome and sometimes kill the joy of kaggling ... but it is unavoidable when kaggle grows large. ", "votes": 8 }, { "id": 919586, "postDate": "2020-07-08T01:52:50.007Z", "content": "Thank you for your input here @hengck23, your opinion is extremely valuable π
", "rawMarkdown": "Thank you for your input here @hengck23, your opinion is extremely valuable π ", "votes": 1 }, { "id": 919603, "postDate": "2020-07-08T02:21:31.910Z", "content": "I agree , our code goes through a rigorous license verification and open source scan process . A three days coding takes sometimes two weeks or more to validate licensing and vulnerabilities in the libraries we use . This is a pain, but this is how it works in corporate world . My only problem is, most of us are programmer here . For some people like us , it's just a hobby . It is little too much to think all this , if the ownership is given to competitors .
\n\nThats why your idea is great for a checker .
", "rawMarkdown": "I agree , our code goes through a rigorous license verification and open source scan process . A three days coding takes sometimes two weeks or more to validate licensing and vulnerabilities in the libraries we use . This is a pain, but this is how it works in corporate world . My only problem is, most of us are programmer here . For some people like us , it's just a hobby . It is little too much to think all this , if the ownership is given to competitors .\n\n Thats why your idea is great for a checker .", "votes": 2 } ] }, { "id": 922311, "postDate": "2020-07-10T02:41:54.163Z", "content": "Hi. I'm confused about the competitions to rule.
\nIs my understanding correct?
Prize/conference: not GPL-3 license \nmedal: GPL-3 and others are ok.
\n\nCan I choose 2 submissions(for Prize and for medal) within rule If both are correct?
", "rawMarkdown": "Hi. I'm confused about the competitions to rule. \nIs my understanding correct? \n\nPrize/conference: not GPL-3 license \nmedal: GPL-3 and others are ok.\n\nCan I choose 2 submissions(for Prize and for medal) within rule If both are correct?", "votes": 14, "replies": [ { "id": 922315, "postDate": "2020-07-10T02:46:44.970Z", "content": "Yeah, this have been really confusing. How are the two solutions going to be ranked? If one submit two submissions, one compliant and the other non-compliant, and the non-compliant one is within the prize range, do they get disqualified, or simply rescored based on the compliant one? @bendvd @philculliton
", "rawMarkdown": "Yeah, this have been really confusing. How are the two solutions going to be ranked? If one submit two submissions, one compliant and the other non-compliant, and the non-compliant one is within the prize range, do they get disqualified, or simply rescored based on the compliant one? @bendvd @philculliton", "votes": 1 }, { "id": 923347, "postDate": "2020-07-10T18:33:59.563Z", "content": "This is always the way Kaggle challenges have worked so this is nothing new. The rules are generally checked for the winning solutions. It is not feasible to check all solutions. That doesn't mean that it is OK to use a non compliant license because you won't be in winning contention just as it wouldn't be okay to use datasets you're not allowed to use or to share code privately or break other rules because you won't be in the winners circle. That being said, I do believe flagrant instances of breaking the rules by non winners have been dealt with.
\n\nThis is a community. Everyone should strive to conduct themselves ethically and keep it healthy and vibrant.
", "rawMarkdown": "This is always the way Kaggle challenges have worked so this is nothing new. The rules are generally checked for the winning solutions. It is not feasible to check all solutions. That doesn't mean that it is OK to use a non compliant license because you won't be in winning contention just as it wouldn't be okay to use datasets you're not allowed to use or to share code privately or break other rules because you won't be in the winners circle. That being said, I do believe flagrant instances of breaking the rules by non winners have been dealt with.\n\nThis is a community. Everyone should strive to conduct themselves ethically and keep it healthy and vibrant.", "votes": 7 }, { "id": 923354, "postDate": "2020-07-10T18:41:08.547Z", "content": "Well said Ross!
", "rawMarkdown": "Well said Ross!", "votes": 1 }, { "id": 923616, "postDate": "2020-07-11T02:58:59.180Z", "content": "I almost agree with @rwightman but host comment is only winning solution required.
\nAlso, the other host said the final solution is not allowed.
\nSo, the host should clearly that rule.
\nIt leads to keep fairness in this competition.
https://www.kaggle.com/c/global-wheat-detection/discussion/163433#919908\nhttps://www.kaggle.com/c/global-wheat-detection/discussion/163433#91907
", "rawMarkdown": "I almost agree with @rwightman but host comment is only winning solution required. \nAlso, the other host said the final solution is not allowed. \nSo, the host should clearly that rule. \nIt leads to keep fairness in this competition.\n\nhttps://www.kaggle.com/c/global-wheat-detection/discussion/163433#919908\nhttps://www.kaggle.com/c/global-wheat-detection/discussion/163433#91907", "votes": 4 }, { "id": 925051, "postDate": "2020-07-11T19:32:17.073Z", "content": "@tereka I do understand the sentiment, the hosts stating that essentially only the winning solutions need to follow the rules isn't great... but in practice that is more or less the case in all challenges as I mentioned. The hosts care about the winning solutions, those are usually the only ones that have their code or data sources checked so from their point of view the statement makes sense.
\n\nAnyone participating that ends up in a winner spot when they weren't expecting and didn't decide to clean up their solution will be pretty upset though. Some shakeups are pretty significant so you never know :)
", "rawMarkdown": "@tereka I do understand the sentiment, the hosts stating that essentially only the winning solutions need to follow the rules isn't great... but in practice that is more or less the case in all challenges as I mentioned. The hosts care about the winning solutions, those are usually the only ones that have their code or data sources checked so from their point of view the statement makes sense. \n\nAnyone participating that ends up in a winner spot when they weren't expecting and didn't decide to clean up their solution will be pretty upset though. Some shakeups are pretty significant so you never know :) " }, { "id": 925290, "postDate": "2020-07-12T01:48:10.340Z", "content": "From the admin post, the two final submissions must be compliant with the rules. The admins have an history of checking all submissions one by one for several days in order to check whether or not the competitors followed the rules.\nI am going to go back to your git repo!
", "rawMarkdown": "From the admin post, the two final submissions must be compliant with the rules. The admins have an history of checking all submissions one by one for several days in order to check whether or not the competitors followed the rules.\nI am going to go back to your git repo!", "votes": 2 } ] }, { "id": 914432, "postDate": "2020-07-03T21:33:49.783Z", "content": "@phoenix9032 If I was interpreting the rules as written
\n\n\nOpen Source: You hereby license and will license your winning Submission and the source code used to generate the Submission under an Open Source Initiative-approved license (see www.opensource.org) that in no event limits commercial use of such code or model containing or depending on such code. For generally commercially available software that you used to generate your submission that is not owned by you, but that can be procured by the Competition Sponsor without undue expense, you do not need to grant the license in the preceding sentence for that software..
Then no, you would not be allowed to use Yolov5 as it is GPL 3 and GPL 3 'limits commercial use of such code'. Yet many have posted use of yolov5 in the external data thread with no comment from the organisers leading to a potential issue at the end (once again). There should be more proactive monitoring of this.
\n\nThere are also mentions of \n* https://github.com/zylo117/Yet-Another-EfficientDet-Pytorch - LGPL, likely not limiting commercial use of such code if you ONLY import, but if you modify the code in any way it would no longer be useable commercially without releasing the code and all changes
EDIT: GPL dependant code is technically releasable commercially, but requires releasing all of the other code and also releasing under GPL, it is viral. In most commerical use wrt to the term 'limit' I'd consider as limiting having done many code audits for companies releasing commercial software and they never want the limitation of releasing all of their code. Either way, 'limit' should be explicitly defined in this context.
", "rawMarkdown": "@phoenix9032 If I was interpreting the rules as written\n\n```\nOpen Source: You hereby license and will license your winning Submission and the source code used to generate the Submission under an Open Source Initiative-approved license (see www.opensource.org) that in no event limits commercial use of such code or model containing or depending on such code. For generally commercially available software that you used to generate your submission that is not owned by you, but that can be procured by the Competition Sponsor without undue expense, you do not need to grant the license in the preceding sentence for that software..```\n\nThen no, you would not be allowed to use Yolov5 as it is GPL 3 and GPL 3 'limits commercial use of such code'. Yet many have posted use of yolov5 in the external data thread with no comment from the organisers leading to a potential issue at the end (once again). There should be more proactive monitoring of this.\n\nThere are also mentions of \n* https://github.com/zylo117/Yet-Another-EfficientDet-Pytorch - LGPL, likely not limiting commercial use of such code if you ONLY `import`, but if you modify the code in any way it would no longer be useable commercially without releasing the code and all changes\n\n* https://pytorch.org/hub/facebookresearch_WSL-Images_resnext/ - explicitly licensed by Facebook as CC-BY-NC 4.0 as with their other SSL/SWSL models\n\nEDIT: GPL dependant code is technically releasable commercially, but requires releasing all of the other code and also releasing under GPL, it is viral. In most commerical use wrt to the term 'limit' I'd consider as limiting having done many code audits for companies releasing commercial software and they never want the limitation of releasing all of their code. Either way, 'limit' should be explicitly defined in this context.", "votes": 11, "replies": [ { "id": 914504, "postDate": "2020-07-04T01:44:30.927Z", "content": "@rwightman Thanks Ross for the insights .. would love to take a chance here to thank you for all your valuable libraries too...
\n\n@philculliton @bendvd @maggiemd @stavness I think the host needs to re-verify this restriction if the use case of this competition is acceptable or not under GPL-3 ... Many top performers now use YoloV5, it would be great to know earlier whether it's OK, so that we all can think/implement of the correct strategy in the remaining time.
", "rawMarkdown": "@rwightman Thanks Ross for the insights .. would love to take a chance here to thank you for all your valuable libraries too...\n\n@philculliton @bendvd @maggiemd @stavness I think the host needs to re-verify this restriction if the use case of this competition is acceptable or not under GPL-3 ... Many top performers now use YoloV5, it would be great to know earlier whether it's OK, so that we all can think/implement of the correct strategy in the remaining time.", "votes": 7 }, { "id": 914563, "postDate": "2020-07-04T03:31:26.840Z", "content": "Thanks so much @rwightman for the insights .
", "rawMarkdown": "Thanks so much @rwightman for the insights .", "votes": 3 } ] }, { "id": 917308, "postDate": "2020-07-06T12:03:24.267Z", "content": "Thanks for your question @phoenix9032 and thanks for your insights @rwightman . It is a very interesting question, as I was more used to MIT licence. I'm not an expert but GPL seems good for me. I will confirm by tomorrow.
", "rawMarkdown": "Thanks for your question @phoenix9032 and thanks for your insights @rwightman . It is a very interesting question, as I was more used to MIT licence. I'm not an expert but GPL seems good for me. I will confirm by tomorrow.", "votes": 9, "replies": [ { "id": 917326, "postDate": "2020-07-06T12:16:42.817Z", "content": "Excellent . Thanks
", "rawMarkdown": "Excellent . Thanks ", "votes": 3 }, { "id": 917855, "postDate": "2020-07-06T19:12:01.413Z", "content": "Great to have a response from the host, thanks for looking into it.
\n\nGPL or LPGL will not disallow a company from using code commercially. It does mean that the code, modifications to the code, and potentially all other code used with it has to be made available under the same license. LGPL is a bit less demanding in that you are allowed to dynamically link without releasing the code that links with the LGPL library, but still have to release any modifications to the original code. What LGPL means in the context of Python is AFAIK not 100% clear, the assumption is import blah is equivalent to 'dynamic linking' with blah but I don't think it's been tested in court.
As the host, if you're learning from this, eassentially treating it as a demo, and using solutions to guide future work, you likely won't be concerned about GPL. If you were planning to throw it straight into a product with proprietary code, you'd likely need to call up the lawyers :) In this case I believe the author of Yolov5 made his code GPL so that commercial entities would get in touch re licensing if they wanted to put it in a proprietary product.
", "rawMarkdown": "Great to have a response from the host, thanks for looking into it.\n\nGPL or LPGL will not disallow a company from using code commercially. It does mean that the code, modifications to the code, and potentially all other code used with it has to be made available under the same license. LGPL is a bit less demanding in that you are allowed to dynamically link without releasing the code that links with the LGPL library, but still have to release any modifications to the original code. What LGPL means in the context of Python is AFAIK not 100% clear, the assumption is `import blah` is equivalent to 'dynamic linking' with blah but I don't think it's been tested in court.\n\nAs the host, if you're learning from this, eassentially treating it as a demo, and using solutions to guide future work, you likely won't be concerned about GPL. If you were planning to throw it straight into a product with proprietary code, you'd likely need to call up the lawyers :) In this case I believe the author of Yolov5 made his code GPL so that commercial entities would get in touch re licensing if they wanted to put it in a proprietary product.", "votes": 3 } ] }, { "id": 922162, "postDate": "2020-07-09T20:42:46.800Z", "content": "Hard solution: carefully select packages that are \"MIT compatible\".\nSimple solution: only submit models that don't win. ;)
", "rawMarkdown": "Hard solution: carefully select packages that are \"MIT compatible\".\nSimple solution: only submit models that don't win. ;) ", "votes": 10 }, { "id": 919074, "postDate": "2020-07-07T17:37:15.737Z", "content": "Thanks for the question, we've reviewed the rules and have clarified that the winning solution needs to fit the MIT license. Any data or code with a GPL license would not be allowed in the final solution.
", "rawMarkdown": "Thanks for the question, we've reviewed the rules and have clarified that the winning solution needs to fit the MIT license. Any data or code with a GPL license would not be allowed in the final solution. ", "votes": 8, "replies": [ { "id": 919080, "postDate": "2020-07-07T17:40:01.147Z", "content": "Thanks for the response Maggie!
", "rawMarkdown": "Thanks for the response Maggie! ", "votes": 2 }, { "id": 919119, "postDate": "2020-07-07T18:05:34.383Z", "content": "That clearly means YoloV5 is not allowed..is that correct?
", "rawMarkdown": "That clearly means YoloV5 is not allowed..is that correct?", "votes": 2 }, { "id": 919133, "postDate": "2020-07-07T18:15:10.027Z", "content": "Thanks so much @maggiemd . So Yolov5 is out of the question . Time to go back to drawing board .
", "rawMarkdown": "Thanks so much @maggiemd . So Yolov5 is out of the question . Time to go back to drawing board .", "votes": 2 }, { "id": 919300, "postDate": "2020-07-07T19:27:58.303Z", "content": "Could you extend your answer a bit please? Why is MIT the only valid open source license? I thought it is about commercial use. The license does allow commercial use (even patent and private use).
\n\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F712087%2Fcad2a9fa34c4b708071c2df298353b22%2FGPL.jpg?generation=1594149872697651&alt=media\")
I guess you are aware that you just invalidated almost all solutions in the top 50. Not to mention the time and hardware that was spent.
", "rawMarkdown": "Could you extend your answer a bit please? Why is MIT the only valid open source license? I thought it is about commercial use. The license does allow commercial use (even patent and private use).\n\n\n\nI guess you are aware that you just invalidated almost all solutions in the top 50. Not to mention the time and hardware that was spent.", "votes": 6 }, { "id": 919324, "postDate": "2020-07-07T19:43:27.097Z", "content": "Is the \"darknet\" license valid? I think that it allows everything: https://github.com/AlexeyAB/darknet/blob/master/LICENSE. One could even train v5 with the v4 repo π€
", "rawMarkdown": "Is the \"darknet\" license valid? I think that it allows everything: https://github.com/AlexeyAB/darknet/blob/master/LICENSE. One could even train v5 with the v4 repo π€ ", "votes": 3 }, { "id": 919365, "postDate": "2020-07-07T20:04:43.790Z", "content": "@maggiemd Only winning solution has to be on MIT or all leaderboard solutions are concerned ?
", "rawMarkdown": "@maggiemd Only winning solution has to be on MIT or all leaderboard solutions are concerned ? ", "votes": 4 }, { "id": 919382, "postDate": "2020-07-07T20:21:58.267Z", "content": "@seesee The issue with the GPL license is that it requires the same license to be used in all derivative works. The organizers are likely against this idea as they eventually want a closed-source version for commercial means. And yes, that license does appear to allow re-licensing under MIT as it is in the public domain.
\n\n@nacirbouazizi I doubt they will invest the time to examine each top LB solution for it's license compatibility. So sure, you could probably continue to use YoloV5 or whatever framework you want to get a nice medal but if you actually win, it would be very disappointing to get disqualified over.
", "rawMarkdown": "@seesee The issue with the GPL license is that it requires the same license to be used in all derivative works. The organizers are likely against this idea as they eventually want a closed-source version for commercial means. And yes, that license does appear to allow re-licensing under MIT as it is in the public domain.\n\n@nacirbouazizi I doubt they will invest the time to examine each top LB solution for it's license compatibility. So sure, you could probably continue to use YoloV5 or whatever framework you want to get a nice medal but if you actually win, it would be very disappointing to get disqualified over.", "votes": 4 }, { "id": 919405, "postDate": "2020-07-07T20:42:22.893Z", "content": "@matthewmasters Thanks for your response. \nIt's actually sad to know this after all the hardwork and 21 days until merger deadline.\nNow I have so start from 0 and try fasterrcnn !
", "rawMarkdown": "@matthewmasters Thanks for your response. \nIt's actually sad to know this after all the hardwork and 21 days until merger deadline.\nNow I have so start from 0 and try fasterrcnn ! ", "votes": 3 }, { "id": 919523, "postDate": "2020-07-08T00:19:39.507Z", "content": "Thanks for your response.
\nPlease announce about the official view of license in Kaggle discussion(not this thread).
\nMany competitors use yolov5(GPL-V3 License). It's a huge impact on this competiton.
Normally when we do license checks in companies, MIT(keras), Apache, BSD(pytorch) are considered as open-source licences because they are commercially available without restrictions.\nThe one you should watch out is GPL (you have to submit the source code when requested).. but this is not work!
Since the hosts are universities and not business, I'm not sure why they are not permitting GPLs. I'm not a big fan of yolov5 but I think the hosts want good and innovative solutions, not license-free solutions.
I have one specific question: can we use Mosaic augmentation? I believe the author of YOLOv3-ultralytics and YOLOv5, Glenn Jocher, came up with this idea originally. Are we allowed to write our own Mosaic augmentation then?
\n\nMosaic is very straightforward and intuitive. Other researchers have also came up with similar ideas (e.g. Sticher) before. I assume that we can play around with the wording by calling Mosaic \"modified/customized Sticher augmentation\", can't we?
", "rawMarkdown": "I have one specific question: can we use Mosaic augmentation? I believe the author of YOLOv3-ultralytics and YOLOv5, Glenn Jocher, came up with this idea originally. Are we allowed to write our own Mosaic augmentation then?\n\nMosaic is very straightforward and intuitive. Other researchers have also came up with similar ideas (e.g. [Sticher](https://arxiv.org/pdf/2004.12432.pdf)) before. I assume that we can play around with the wording by calling Mosaic \"modified/customized Sticher augmentation\", can't we?", "votes": 6 }, { "id": 919571, "postDate": "2020-07-08T01:39:02.063Z", "content": "Hi, thanks for replying. But I should mention that Pytorch is under BSD license, which is a little stricter than MIT license. So we should start from a deep learning framework ???
", "rawMarkdown": "Hi, thanks for replying. But I should mention that Pytorch is under BSD license, which is a little stricter than MIT license. So we should start from a deep learning framework ???", "votes": 4 }, { "id": 919645, "postDate": "2020-07-08T03:08:12.787Z", "content": "@jihangz\nthese licenses become a problem when you use codes directly (copyright).\nIf you implement the idea with your own codes, you have the copyright and of cource are free to use in kaggle competitions.
", "rawMarkdown": "@jihangz\nthese licenses become a problem when you use codes directly (copyright).\nIf you implement the idea with your own codes, you have the copyright and of cource are free to use in kaggle competitions.", "votes": 2 }, { "id": 919814, "postDate": "2020-07-08T05:54:39.827Z", "content": "everyone above 0.76 rn : Years of academy training wasted
", "rawMarkdown": "everyone above 0.76 rn : Years of academy training wasted ", "votes": 6 }, { "id": 919908, "postDate": "2020-07-08T07:22:31.630Z", "content": "The main issue with changing the license now is that we change the rule during competitions. Maybe somebody has developed a solution compliant with the rules and it won't be fair either for him or her. The code of Yolo-V5 by ultralytics is GPL3 but not the idea (however I agree that recoding everything is not what a competitor want to do).
\n\nOnly the winning solutions are required to be MIT-licensed.
", "rawMarkdown": "The main issue with changing the license now is that we change the rule **during** competitions. Maybe somebody has developed a solution compliant with the rules and it won't be fair either for him or her. The code of Yolo-V5 by ultralytics is GPL3 but not the idea (however I agree that recoding everything is not what a competitor want to do).\n\nOnly the winning solutions are required to be MIT-licensed.\n", "votes": 2 }, { "id": 919925, "postDate": "2020-07-08T07:40:47.897Z", "content": "So (hypothetically) if all competitors in the gold zone ended up using yolov5, do they all get disqualified? Or do the first compliant-ranking submission on the leader board gets the prize? And what about the medals and points? A lot of people (including me) compete for medals not prizes.
", "rawMarkdown": "So (hypothetically) if all competitors in the gold zone ended up using yolov5, do they all get disqualified? Or do the first compliant-ranking submission on the leader board gets the prize? And what about the medals and points? A lot of people (including me) compete for medals not prizes. ", "votes": 6 }, { "id": 919942, "postDate": "2020-07-08T07:58:20.897Z", "content": "Can I ask what is \"winning solutions\"? \nIn the money?
", "rawMarkdown": "Can I ask what is \"winning solutions\"? \nIn the money?", "votes": 3 }, { "id": 919943, "postDate": "2020-07-08T07:59:31.110Z", "content": "Yeah I think they might have to go pretty far down the list before finding an eligible solution π
", "rawMarkdown": "Yeah I think they might have to go pretty far down the list before finding an eligible solution π ", "votes": 1 }, { "id": 919945, "postDate": "2020-07-08T08:02:35.530Z", "content": "Fun fact: I'm actually not using Yolov5 at the moment, and I believe neither @tereka is
", "rawMarkdown": "Fun fact: I'm actually not using Yolov5 at the moment, and I believe neither @tereka is" }, { "id": 919947, "postDate": "2020-07-08T08:03:58.137Z", "rawMarkdown": "", "votes": -1, "isDeleted": true }, { "id": 922088, "postDate": "2020-07-09T19:51:21.103Z", "content": "Dear @alexanderliao, will you also enlighten us on what you are using ? ;) DETR? FasterRCNN, EfficientDet ? Just the Model's name would be quite interesting. \nThanks a lot! - and no hard feelings π
", "rawMarkdown": "Dear @alexanderliao, will you also enlighten us on what you are using ? ;) DETR? FasterRCNN, EfficientDet ? Just the Model's name would be quite interesting. \nThanks a lot! - and no hard feelings π ", "votes": -1 }, { "id": 922367, "postDate": "2020-07-10T04:03:03.733Z", "content": "It would be trolling to say that I'm using OverFeat.
", "rawMarkdown": "It would be trolling to say that I'm using [OverFeat](https://arxiv.org/abs/1312.6229). ", "votes": -1 }, { "id": 929393, "postDate": "2020-07-14T16:36:14.647Z", "content": "Hi @maggiemd, the people here still don't get it, and neither am I getting it.\nCan we use YOLOv5 and we will just be excluded from WINING the \"in the money\" places, or will all YOLO submissions completely be removed from the LB? Please make a clear statement. This competition get's worse and worse.
", "rawMarkdown": "Hi @maggiemd, the people here still don't get it, and neither am I getting it.\nCan we use YOLOv5 and we will just be excluded from WINING the \"in the money\" places, or will all YOLO submissions completely be removed from the LB? Please make a clear statement. This competition get's worse and worse.\n" }, { "id": 929513, "postDate": "2020-07-14T18:04:01.037Z", "content": "@chrisden You can't win prizes if you use YOLOv5.
", "rawMarkdown": "@chrisden You can't win prizes if you use YOLOv5." }, { "id": 930013, "postDate": "2020-07-15T06:18:26.193Z", "content": "So yolov5 at top area will get medals normally?
", "rawMarkdown": "So yolov5 at top area will get medals normally?", "votes": 1 } ] }, { "id": 911848, "postDate": "2020-07-02T04:07:37.377Z", "content": "Thatβs a good question. From my understanding, the rules require the solution to be published under an MIT license. And since a GPL license is more restrictive (as it requires derivatives to be open sourced), it may not be allowed for this competition. Though maybe since most people are simply using the framework, not creating derivatives, it may be fine? Iβm certainly no expert so donβt take this information as fact.
", "rawMarkdown": "Thatβs a good question. From my understanding, the rules require the solution to be published under an MIT license. And since a GPL license is more restrictive (as it requires derivatives to be open sourced), it may not be allowed for this competition. Though maybe since most people are simply using the framework, not creating derivatives, it may be fine? Iβm certainly no expert so donβt take this information as fact.", "votes": 5, "replies": [ { "id": 911864, "postDate": "2020-07-02T04:31:17.883Z", "content": "Yes even I had that question. Ofcourse we are finetuning. We also dont know if organizer wants to be restricted to open source their solutions once they adopt the top solution (if it involves yolov5) . I have very less idea about legality and licensing myself . So thought of asking the forum .
", "rawMarkdown": "Yes even I had that question. Ofcourse we are finetuning. We also dont know if organizer wants to be restricted to open source their solutions once they adopt the top solution (if it involves yolov5) . I have very less idea about legality and licensing myself . So thought of asking the forum .", "votes": 2 } ] }, { "id": 912554, "postDate": "2020-07-02T15:05:18.257Z", "content": "Thanks for the good question. Tagging the hosts @bendvd @stavness @maggiemd @philculliton for clarification .
\n\nupdated July 5 I emailed Phil just to make sure that he is notified on this thread, but not yet reply... I will update instantly if I get reply
\n\nupdated July 6 Not yet reply. Maybe I will also email other hosts. @phoenix9032 Nirjhar, to get attention from hosts, could you please try renaming this topic e.g. \"Hosts answer needed : GPL-3 licence on YoloV5\"
\n\nupdated July 7 Thailand early morning We got the kind reply here... Thanks the hosts ... I haven't emailed any other hosts :)
", "rawMarkdown": "Thanks for the good question. Tagging the hosts @bendvd @stavness @maggiemd @philculliton for clarification .\n\n**updated July 5** I emailed Phil just to make sure that he is notified on this thread, but not yet reply... I will update instantly if I get reply\n\n**updated July 6** Not yet reply. Maybe I will also email other hosts. @phoenix9032 Nirjhar, to get attention from hosts, could you please try renaming this topic e.g. \"Hosts answer needed : GPL-3 licence on YoloV5\" \n\n**updated July 7 Thailand early morning** We got the kind reply here... Thanks the hosts ... I haven't emailed any other hosts :)", "votes": 6, "replies": [ { "id": 915854, "postDate": "2020-07-05T06:38:27.433Z", "content": "Thanks @ratthachat
", "rawMarkdown": "Thanks @ratthachat ", "votes": 1 }, { "id": 917035, "postDate": "2020-07-06T07:33:47.890Z", "content": "Done @ratthachat
", "rawMarkdown": "Done @ratthachat ", "votes": 2 }, { "id": 917414, "postDate": "2020-07-06T13:34:53.460Z", "content": "Hi! July 2-5 was a holiday in the US. Also July 5 was a Sunday. It's 9:30 AM where I am right now, so I'm just getting into work. No need to email other people! Maggie and I just haven't been in the office.
\n\nThanks for your questions! We'll discuss ASAP with the host.
", "rawMarkdown": "Hi! July 2-5 was a holiday in the US. Also July 5 was a Sunday. It's 9:30 AM where I am right now, so I'm just getting into work. No need to email other people! Maggie and I just haven't been in the office.\n\nThanks for your questions! We'll discuss ASAP with the host.", "votes": 5 }, { "id": 917447, "postDate": "2020-07-06T14:17:21.197Z", "content": "@philculliton As Richard Marx once said \" We will be right here waiting for you \" :) \nBy the way , hope you had a good time in long weekend and we would be patient for your response and consideration.
", "rawMarkdown": "@philculliton As Richard Marx once said \" We will be right here waiting for you \" :) \nBy the way , hope you had a good time in long weekend and we would be patient for your response and consideration. ", "votes": 1 } ] }, { "id": 923984, "postDate": "2020-07-11T07:59:28.553Z", "content": "A lot of people hem and haw about the licensing issues...\nHowever when you look at it from the host's point of view...\nIt makes sense... \nHost a competition... money goes only to the solution that is MIT licensed...\nGPL license necessarily implies that the host or anyone else must necessarily share improvements made over and above the winning solution... { distribution closed source versions is prohibited }
\n\nThis is going to be a thorny issue going forward, for kaggle.\nThe optics of it simply doesn't look good, whichever way you spin it...
\n\nUpside: increased awareness of licensing... among the community.
", "rawMarkdown": "A lot of people hem and haw about the licensing issues...\nHowever when you look at it from the host's point of view...\nIt makes sense... \nHost a competition... money goes only to the solution that is MIT licensed...\nGPL license necessarily implies that the host or anyone else must necessarily share improvements made over and above the winning solution... { distribution closed source versions is prohibited }\n\nThis is going to be a thorny issue going forward, for kaggle.\nThe optics of it simply doesn't look good, whichever way you spin it...\n\nUpside: increased awareness of licensing... among the community.", "votes": 3 }, { "id": 920379, "postDate": "2020-07-08T14:55:09.343Z", "content": "I think kaggle administrators needs to start looking into licensing issues for different competition.\nAs kaggle platform is very popular now, ultimately participants are going to face more lincesing issues in the future.\nOne of the best example is https://www.kaggle.com/c/deepfake-detection-challenge/discussion/157983 challenge. I feel bad for them.
", "rawMarkdown": "I think kaggle administrators needs to start looking into licensing issues for different competition.\nAs kaggle platform is very popular now, ultimately participants are going to face more lincesing issues in the future.\nOne of the best example is https://www.kaggle.com/c/deepfake-detection-challenge/discussion/157983 challenge. I feel bad for them. ", "votes": 3 }, { "id": 920239, "postDate": "2020-07-08T13:00:35.733Z", "content": "So basically if you are in the golden zone with YOLOv5 its fine but if you win using the solution with YOLOv5 you are screwed :p.\nNow I am feeling bad for the top 10 in the leader board.
", "rawMarkdown": "So basically if you are in the golden zone with YOLOv5 its fine but if you win using the solution with YOLOv5 you are screwed :p.\nNow I am feeling bad for the top 10 in the leader board.", "votes": 3, "replies": [ { "id": 920248, "postDate": "2020-07-08T13:05:50.397Z", "content": "They will face the same unfortunate fate like Mr. Giba faced here
\n\nhttps://www.kaggle.com/c/deepfake-detection-challenge/discussion/157983
", "rawMarkdown": "They will face the same unfortunate fate like Mr. Giba faced here \n\nhttps://www.kaggle.com/c/deepfake-detection-challenge/discussion/157983", "votes": 4 }, { "id": 921552, "postDate": "2020-07-09T11:25:23.313Z", "content": "This gentleman lost 500,000$ just like that. π°\nHow unfortunate!!!
", "rawMarkdown": "This gentleman lost 500,000$ just like that. π°\nHow unfortunate!!! ", "votes": 1 }, { "id": 930032, "postDate": "2020-07-15T06:36:38.683Z", "content": "π π π
", "rawMarkdown": "π π π ", "votes": -1 } ] }, { "id": 919522, "postDate": "2020-07-08T00:19:24.157Z", "content": "Okay I must admit that this is bad news guys. But life has to move on. \nNow we should go back and carefully check licence of YoloV4 / EffDet / ResNest too.\nGood luck everybody.
", "rawMarkdown": "Okay I must admit that this is bad news guys. But life has to move on. \nNow we should go back and carefully check licence of YoloV4 / EffDet / ResNest too.\nGood luck everybody. ", "votes": 4, "replies": [ { "id": 919928, "postDate": "2020-07-08T07:45:19.840Z", "content": "EffDet use Apache-2.0 License, so not allowed neither?
", "rawMarkdown": "EffDet use Apache-2.0 License, so not allowed neither?", "votes": 2 }, { "id": 919933, "postDate": "2020-07-08T07:51:37.777Z", "content": "Apache v2.0 is very open license . That's what the term \"Fit for MIT license \" not \" Only MIT license \" .
", "rawMarkdown": "Apache v2.0 is very open license . That's what the term \"Fit for MIT license \" not \" Only MIT license \" . ", "votes": 2 }, { "id": 922351, "postDate": "2020-07-10T03:36:06.267Z", "content": "My repo changed torchvision's sourcecode, so it is also can't be used?
", "rawMarkdown": "My repo changed torchvision's sourcecode, so it is also can't be used?" } ] }, { "id": 919454, "postDate": "2020-07-07T21:45:56.630Z", "content": "As the LB is populated with a lot of YoloV5 solutions, it will make it harder to situate our results for other algorithms if we cannot make them perform as well as YoloV5... As far as I know, I didnt see top performers with EfficientDet or Faster RCNN.. Maybe we should try other Yolo versions ?
", "rawMarkdown": "As the LB is populated with a lot of YoloV5 solutions, it will make it harder to situate our results for other algorithms if we cannot make them perform as well as YoloV5... As far as I know, I didnt see top performers with EfficientDet or Faster RCNN.. Maybe we should try other Yolo versions ?", "votes": 3, "replies": [ { "id": 919567, "postDate": "2020-07-08T01:25:42.637Z", "content": "yes. now, public board means nothing.
", "rawMarkdown": "yes. now, public board means nothing. ", "votes": 2 } ] }, { "id": 919958, "postDate": "2020-07-08T08:14:47.827Z", "content": "When I came to this competition I was really surprised just cause I've realized that almost everyone uses Yolo. That's a little bit upset cause diversity is reduced to zero. Half of all people just use the following submit baseline: https://www.kaggle.com/nvnnghia/yolov5-pseudo-labeling (I am not an exception but I use my own baseline at the same time). One more interesting fact: almost all the people who placed in the gold now did their submits before this thread has been created and EVEN before the following discussion has been started: https://www.kaggle.com/c/global-wheat-detection/discussion/163291 (this is the source of the future shake up). Probably, the hosts just weren't ready for that. Apparently, they should understand it before a competition has started. But they haven't done it and now they need to complete a huge work for filtering all the solutions.
", "rawMarkdown": "When I came to this competition I was really surprised just cause I've realized that almost everyone uses Yolo. That's a little bit upset cause diversity is reduced to zero. Half of all people just use the following submit baseline: https://www.kaggle.com/nvnnghia/yolov5-pseudo-labeling (I am not an exception but I use my own baseline at the same time). One more interesting fact: almost all the people who placed in the gold now did their submits before this thread has been created and EVEN before the following discussion has been started: https://www.kaggle.com/c/global-wheat-detection/discussion/163291 (this is the source of the future shake up). Probably, the hosts just weren't ready for that. Apparently, they should understand it before a competition has started. But they haven't done it and now they need to complete a huge work for filtering all the solutions.", "votes": 1, "replies": [ { "id": 925308, "postDate": "2020-07-12T02:16:42.920Z", "content": "@vadimtimakin Thanks for pointing this out.. π
", "rawMarkdown": "@vadimtimakin Thanks for pointing this out.. π \n" } ] }, { "id": 919625, "postDate": "2020-07-08T02:46:32.217Z", "content": "I wonder the ones only in medal zone will submit the licenses too? Or the ones in medal zone can use YOLO v5
", "rawMarkdown": "I wonder the ones only in medal zone will submit the licenses too? Or the ones in medal zone can use YOLO v5", "votes": 1, "replies": [ { "id": 919901, "postDate": "2020-07-08T07:18:42.913Z", "content": "Only winning solutions are required to be MIT licensed
", "rawMarkdown": "Only winning solutions are required to be MIT licensed", "votes": 2 }, { "id": 920000, "postDate": "2020-07-08T08:39:29.070Z", "content": "Both Mmdetection and Detectron2 are Apache-2.0 License, so does that exclude them too β¦ ?! This would be an amazing competition without the most commonly used libraries.
", "rawMarkdown": "Both Mmdetection and Detectron2 are Apache-2.0 License, so does that exclude them too β¦ ?! This would be an amazing competition without the most commonly used libraries." }, { "id": 920295, "postDate": "2020-07-08T13:35:57.850Z", "content": "And how can you know upfront if you will win? :) Will you be disqualified if you use YoloV5 on position 3 but not if you use it on position 4?
", "rawMarkdown": "And how can you know upfront if you will win? :) Will you be disqualified if you use YoloV5 on position 3 but not if you use it on position 4?", "votes": 5 }, { "id": 920302, "postDate": "2020-07-08T13:41:57.880Z", "content": "Thanks for reply :D
", "rawMarkdown": "Thanks for reply :D" }, { "id": 920326, "postDate": "2020-07-08T14:00:40.683Z", "rawMarkdown": "", "votes": 9, "isDeleted": true }, { "id": 920486, "postDate": "2020-07-08T16:19:25.843Z", "content": "@sokazaki while that would be nice, that would involve an insane amount of work on the part of the competition organizers and compliance team, itβs simply unrealistic for this competition
", "rawMarkdown": "@sokazaki while that would be nice, that would involve an insane amount of work on the part of the competition organizers and compliance team, itβs simply unrealistic for this competition" } ] }, { "id": 919510, "postDate": "2020-07-07T23:30:38.917Z", "content": "A little disappointed about the answer, but at least we know what is not good to do now..π \nThanks for bringing up this question @phoenix9032
", "rawMarkdown": "A little disappointed about the answer, but at least we know what is not good to do now..π \nThanks for bringing up this question @phoenix9032 ", "votes": 2, "replies": [ { "id": 919589, "postDate": "2020-07-08T01:59:07.410Z", "content": "It was a very harmless question to start with.Honestly I thought it would be just someone answering me \"Its already clarified , it is okay \" . I feel a bit stupid now .
", "rawMarkdown": "It was a very harmless question to start with.Honestly I thought it would be just someone answering me \"Its already clarified , it is okay \" . I feel a bit stupid now . ", "votes": 5 }, { "id": 919594, "postDate": "2020-07-08T02:06:49.077Z", "content": "@phoenix9032 No need to feel stupid, it was a great question!
\n\nIf you hadn't asked, many people would have continued spending their time on an approach that would not be allowed in the end. The way I see it, you saved a lot of people from wasting time and potential disappointment. And if anything, I'm the stupid one for not considering the implications of different licenses. So thank you for bringing this up! π
", "rawMarkdown": "@phoenix9032 No need to feel stupid, it was a great question!\n\nIf you hadn't asked, many people would have continued spending their time on an approach that would not be allowed in the end. The way I see it, you saved a lot of people from wasting time and potential disappointment. And if anything, I'm the stupid one for not considering the implications of different licenses. So thank you for bringing this up! π ", "votes": 3 }, { "id": 921274, "postDate": "2020-07-09T07:13:07.873Z", "content": "It was a great question and show that choosing the right open-source licence for a competition is not as easy as we thought !
", "rawMarkdown": "It was a great question and show that choosing the right open-source licence for a competition is not as easy as we thought !", "votes": 1 } ] }, { "id": 921577, "postDate": "2020-07-09T11:53:39.083Z", "content": "Hi All,
\n\nI think we can use AlexeyAB/darknet/YOLOv4. See it's license here-\nLicense
\n\nFor details please check my comment in below discussion-\ndiscussion
", "rawMarkdown": "Hi All,\n\nI think we can use [AlexeyAB/darknet/YOLOv4](https://github.com/AlexeyAB/darknet). See it's license here-\n[License](https://github.com/AlexeyAB/darknet/blob/master/LICENSE)\n\nFor details please check my comment in below discussion-\n[discussion](https://www.kaggle.com/c/global-wheat-detection/discussion/164845)" }, { "id": 930035, "postDate": "2020-07-15T06:40:48.183Z", "content": "So what's the result? yolov5 disallow? or medals area yolov5 will get medal but can't win prize? @phoenix9032
", "rawMarkdown": "So what's the result? yolov5 disallow? or medals area yolov5 will get medal but can't win prize? @phoenix9032", "replies": [ { "id": 930067, "postDate": "2020-07-15T07:05:52Z", "content": "if you are using YOLOv5 your disqualified. Also, you will not receive any medals. So dont use YOLOv5. Use of YOLOv5 forbidden in Kaggle
", "rawMarkdown": "if you are using YOLOv5 your disqualified. Also, you will not receive any medals. So dont use YOLOv5. Use of YOLOv5 forbidden in Kaggle" }, { "id": 939082, "postDate": "2020-07-22T02:48:57.027Z", "content": "Thank you, disgusted by this gameπ€
", "rawMarkdown": "Thank you, disgusted by this gameπ€ ", "votes": -2 } ] }, { "id": 939271, "postDate": "2020-07-22T06:10:32.777Z", "rawMarkdown": "", "votes": -1, "isDeleted": true, "replies": [ { "id": 939284, "postDate": "2020-07-22T06:23:47.820Z", "content": "Read it again, it states
\n\n\n\n\nCompetitions are open to residents of the United States and worldwide, except that if you are a resident of Crimea, Cuba, Iran, Syria, North Korea, Sudan, or are subject to U.S. export controls or sanctions, you may not enter the Competition.
\n
Only a few countries are excluded and itβs because the US itself has sanctions against these countries, it is not Kaggles doing
", "rawMarkdown": "Read it again, it states\n> Competitions are open to residents of the United States **and worldwide**, except that if you are a resident of Crimea, Cuba, Iran, Syria, North Korea, Sudan, or are subject to U.S. export controls or sanctions, you may not enter the Competition.\n\nOnly a few countries are excluded and itβs because the US itself has sanctions against these countries, it is not Kaggles doing ", "votes": 1 } ] }, { "id": 919999, "postDate": "2020-07-08T08:38:38.017Z", "rawMarkdown": "", "isDeleted": true, "replies": [ { "id": 920031, "postDate": "2020-07-08T08:55:08.133Z", "content": "", "rawMarkdown": "See [here](https://www.kaggle.com/c/global-wheat-detection/discussion/164845#919913) and [here](https://www.kaggle.com/c/global-wheat-detection/discussion/164845#919611)." } ] }, { "id": 919337, "postDate": "2020-07-07T19:47:38.103Z", "rawMarkdown": "", "votes": 2, "isDeleted": true, "replies": [ { "id": 919342, "postDate": "2020-07-07T19:52:27.950Z", "content": "YOLO is not out . YOLOv5 is alone GPLV3
", "rawMarkdown": "YOLO is not out . YOLOv5 is alone GPLV3", "votes": 3 }, { "id": 919899, "postDate": "2020-07-08T07:16:38.617Z", "rawMarkdown": "", "votes": 2, "isDeleted": true } ] } ], "comments": [ { "id": 920645, "author_name": "Glenn Jocher", "author_url": "", "post_date": "2020-07-08T18:19:42.740000", "content": "All,
\n\nThis Global Wheat Detection competition, and this thread in particular were recently brought to my attention here https://github.com/ultralytics/yolov5/issues/317.
\n\nI'm very happy that the community is finding YOLOv5 useful!! I personally have not participated in any Kaggle competitions, but I did participate in a National Geospatial-Intelligence Agency (NGA) competition that got me interested in YOLO and propelled me down my ML path a couple summers ago. I created a repo at the time to document my efforts:\nhttps://github.com/ultralytics/xview-yolov3
\n\nIt's a nice irony now to see the situation reversed, with other young people learning and using YOLOv5 to start their own research and competition efforts. I'm very humbled and hopeful that smart minds might use what we have developed as the basis for further advancements of their own in the future :)
\n\nRegarding the licensing issues, both of our YOLO repositories, https://github.com/ultralytics/yolov3 and https://github.com/ultralytics/yolov5 are licensed as GPL 3.0. The intention of the license is allow full use for any purpose (academic, research or commercial) of the original or modified code, provided that derivative works are open sourced and licensed as the same. The greater good we have in mind here is the capability of all people to benefit from advancements made with future open-sourced works that begin from or leverage YOLOv5 as a starting point.
\n\nUltralytics was founded to work on particle physics applications, we sleep with one eye open towards expanding knowledge and advancing science. We are always open to new ideas, but we believe we have the right license in place for our work. Isaac Newton once said \"If I have seen further it is by standing on the shoulders of Giants.\" I believe that if we allow future works to hide from the world as proprietary code, which other licenses provide provisions for, we are reducing the possible giants that future Isaac Newtons may stand on, and diminishing all of our own futures slightly so.
", "votes": 33, "replies": [] }, { "id": 919568, "author_name": "hengck23", "author_url": "", "post_date": "2020-07-08T01:34:09.293000", "content": "kaggle or someone needs to to make a checking web page like this:
\n\n
welcome board, man. π€
", "votes": -2, "replies": [] }, { "id": 919573, "author_name": "Matt", "author_url": "", "post_date": "2020-07-08T01:39:17.310000", "content": "Great to see you here, @hengck23! And I do like that solution, although I don't expect it'll actually get implemented. Just curious, did you make this \"mock-up\" yourself or is it actually a form somewhere?
", "votes": -2, "replies": [] }, { "id": 919580, "author_name": "hengck23", "author_url": "", "post_date": "2020-07-08T01:47:20.567000", "content": "it is just a mock up in PPT. someone can start web service and charged e.g. $1.00 per check or have it free.
\n\nActually in R&D company i worked for, it is a common practice that in our solution (software prototype, model, dateset) we gave to client, we need to include licensing for any external 3rd-party, open-source solution we used.
\n\nWhen there is money involved, we need to be cautious in company.
\n\nWith more and more people and big company are involved in kaggle competition, kagglers needs to be more aware of licensing. It would be troublesome and sometimes kill the joy of kaggling ... but it is unavoidable when kaggle grows large.
", "votes": 8, "replies": [] }, { "id": 919586, "author_name": "Matt", "author_url": "", "post_date": "2020-07-08T01:52:50.007000", "content": "Thank you for your input here @hengck23, your opinion is extremely valuable π
", "votes": 1, "replies": [] }, { "id": 919603, "author_name": "Nirjhar Roy", "author_url": "", "post_date": "2020-07-08T02:21:31.910000", "content": "I agree , our code goes through a rigorous license verification and open source scan process . A three days coding takes sometimes two weeks or more to validate licensing and vulnerabilities in the libraries we use . This is a pain, but this is how it works in corporate world . My only problem is, most of us are programmer here . For some people like us , it's just a hobby . It is little too much to think all this , if the ownership is given to competitors .
\n\nThats why your idea is great for a checker .
", "votes": 2, "replies": [] } ] }, { "id": 922311, "author_name": "tereka", "author_url": "", "post_date": "2020-07-10T02:41:54.163000", "content": "Hi. I'm confused about the competitions to rule.
\nIs my understanding correct?
Prize/conference: not GPL-3 license \nmedal: GPL-3 and others are ok.
\n\nCan I choose 2 submissions(for Prize and for medal) within rule If both are correct?
", "votes": 14, "replies": [ { "id": 922315, "author_name": "Peiyuan Liao", "author_url": "", "post_date": "2020-07-10T02:46:44.970000", "content": "Yeah, this have been really confusing. How are the two solutions going to be ranked? If one submit two submissions, one compliant and the other non-compliant, and the non-compliant one is within the prize range, do they get disqualified, or simply rescored based on the compliant one? @bendvd @philculliton
", "votes": 1, "replies": [] }, { "id": 923347, "author_name": "RossWightman", "author_url": "", "post_date": "2020-07-10T18:33:59.563000", "content": "This is always the way Kaggle challenges have worked so this is nothing new. The rules are generally checked for the winning solutions. It is not feasible to check all solutions. That doesn't mean that it is OK to use a non compliant license because you won't be in winning contention just as it wouldn't be okay to use datasets you're not allowed to use or to share code privately or break other rules because you won't be in the winners circle. That being said, I do believe flagrant instances of breaking the rules by non winners have been dealt with.
\n\nThis is a community. Everyone should strive to conduct themselves ethically and keep it healthy and vibrant.
", "votes": 7, "replies": [] }, { "id": 923354, "author_name": "Matt", "author_url": "", "post_date": "2020-07-10T18:41:08.547000", "content": "Well said Ross!
", "votes": 1, "replies": [] }, { "id": 923616, "author_name": "tereka", "author_url": "", "post_date": "2020-07-11T02:58:59.180000", "content": "I almost agree with @rwightman but host comment is only winning solution required.
\nAlso, the other host said the final solution is not allowed.
\nSo, the host should clearly that rule.
\nIt leads to keep fairness in this competition.
https://www.kaggle.com/c/global-wheat-detection/discussion/163433#919908\nhttps://www.kaggle.com/c/global-wheat-detection/discussion/163433#91907
", "votes": 4, "replies": [] }, { "id": 925051, "author_name": "RossWightman", "author_url": "", "post_date": "2020-07-11T19:32:17.073000", "content": "@tereka I do understand the sentiment, the hosts stating that essentially only the winning solutions need to follow the rules isn't great... but in practice that is more or less the case in all challenges as I mentioned. The hosts care about the winning solutions, those are usually the only ones that have their code or data sources checked so from their point of view the statement makes sense.
\n\nAnyone participating that ends up in a winner spot when they weren't expecting and didn't decide to clean up their solution will be pretty upset though. Some shakeups are pretty significant so you never know :)
", "votes": 0, "replies": [] }, { "id": 925290, "author_name": "eagle4", "author_url": "", "post_date": "2020-07-12T01:48:10.340000", "content": "From the admin post, the two final submissions must be compliant with the rules. The admins have an history of checking all submissions one by one for several days in order to check whether or not the competitors followed the rules.\nI am going to go back to your git repo!
", "votes": 2, "replies": [] } ] }, { "id": 914432, "author_name": "RossWightman", "author_url": "", "post_date": "2020-07-03T21:33:49.783000", "content": "@phoenix9032 If I was interpreting the rules as written
\n\n\nOpen Source: You hereby license and will license your winning Submission and the source code used to generate the Submission under an Open Source Initiative-approved license (see www.opensource.org) that in no event limits commercial use of such code or model containing or depending on such code. For generally commercially available software that you used to generate your submission that is not owned by you, but that can be procured by the Competition Sponsor without undue expense, you do not need to grant the license in the preceding sentence for that software..
Then no, you would not be allowed to use Yolov5 as it is GPL 3 and GPL 3 'limits commercial use of such code'. Yet many have posted use of yolov5 in the external data thread with no comment from the organisers leading to a potential issue at the end (once again). There should be more proactive monitoring of this.
\n\nThere are also mentions of \n* https://github.com/zylo117/Yet-Another-EfficientDet-Pytorch - LGPL, likely not limiting commercial use of such code if you ONLY import, but if you modify the code in any way it would no longer be useable commercially without releasing the code and all changes
EDIT: GPL dependant code is technically releasable commercially, but requires releasing all of the other code and also releasing under GPL, it is viral. In most commerical use wrt to the term 'limit' I'd consider as limiting having done many code audits for companies releasing commercial software and they never want the limitation of releasing all of their code. Either way, 'limit' should be explicitly defined in this context.
", "votes": 11, "replies": [ { "id": 914504, "author_name": "Neuron Engineer", "author_url": "", "post_date": "2020-07-04T01:44:30.927000", "content": "@rwightman Thanks Ross for the insights .. would love to take a chance here to thank you for all your valuable libraries too...
\n\n@philculliton @bendvd @maggiemd @stavness I think the host needs to re-verify this restriction if the use case of this competition is acceptable or not under GPL-3 ... Many top performers now use YoloV5, it would be great to know earlier whether it's OK, so that we all can think/implement of the correct strategy in the remaining time.
", "votes": 7, "replies": [] }, { "id": 914563, "author_name": "Nirjhar Roy", "author_url": "", "post_date": "2020-07-04T03:31:26.840000", "content": "Thanks so much @rwightman for the insights .
", "votes": 3, "replies": [] } ] }, { "id": 917308, "author_name": "Etienne David", "author_url": "", "post_date": "2020-07-06T12:03:24.267000", "content": "Thanks for your question @phoenix9032 and thanks for your insights @rwightman . It is a very interesting question, as I was more used to MIT licence. I'm not an expert but GPL seems good for me. I will confirm by tomorrow.
", "votes": 9, "replies": [ { "id": 917326, "author_name": "Nirjhar Roy", "author_url": "", "post_date": "2020-07-06T12:16:42.817000", "content": "Excellent . Thanks
", "votes": 3, "replies": [] }, { "id": 917855, "author_name": "RossWightman", "author_url": "", "post_date": "2020-07-06T19:12:01.413000", "content": "Great to have a response from the host, thanks for looking into it.
\n\nGPL or LPGL will not disallow a company from using code commercially. It does mean that the code, modifications to the code, and potentially all other code used with it has to be made available under the same license. LGPL is a bit less demanding in that you are allowed to dynamically link without releasing the code that links with the LGPL library, but still have to release any modifications to the original code. What LGPL means in the context of Python is AFAIK not 100% clear, the assumption is import blah is equivalent to 'dynamic linking' with blah but I don't think it's been tested in court.
As the host, if you're learning from this, eassentially treating it as a demo, and using solutions to guide future work, you likely won't be concerned about GPL. If you were planning to throw it straight into a product with proprietary code, you'd likely need to call up the lawyers :) In this case I believe the author of Yolov5 made his code GPL so that commercial entities would get in touch re licensing if they wanted to put it in a proprietary product.
", "votes": 3, "replies": [] } ] }, { "id": 922162, "author_name": "Yassine Alouini", "author_url": "", "post_date": "2020-07-09T20:42:46.800000", "content": "Hard solution: carefully select packages that are \"MIT compatible\".\nSimple solution: only submit models that don't win. ;)
", "votes": 10, "replies": [] }, { "id": 919074, "author_name": "Maggie", "author_url": "", "post_date": "2020-07-07T17:37:15.737000", "content": "Thanks for the question, we've reviewed the rules and have clarified that the winning solution needs to fit the MIT license. Any data or code with a GPL license would not be allowed in the final solution.
", "votes": 8, "replies": [ { "id": 919080, "author_name": "Matt", "author_url": "", "post_date": "2020-07-07T17:40:01.147000", "content": "Thanks for the response Maggie!
", "votes": 2, "replies": [] }, { "id": 919119, "author_name": "Kaushal Shah", "author_url": "", "post_date": "2020-07-07T18:05:34.383000", "content": "That clearly means YoloV5 is not allowed..is that correct?
", "votes": 2, "replies": [] }, { "id": 919133, "author_name": "Nirjhar Roy", "author_url": "", "post_date": "2020-07-07T18:15:10.027000", "content": "Thanks so much @maggiemd . So Yolov5 is out of the question . Time to go back to drawing board .
", "votes": 2, "replies": [] }, { "id": 919300, "author_name": "See--", "author_url": "", "post_date": "2020-07-07T19:27:58.303000", "content": "Could you extend your answer a bit please? Why is MIT the only valid open source license? I thought it is about commercial use. The license does allow commercial use (even patent and private use).
\n\n
I guess you are aware that you just invalidated almost all solutions in the top 50. Not to mention the time and hardware that was spent.
", "votes": 6, "replies": [] }, { "id": 919324, "author_name": "See--", "author_url": "", "post_date": "2020-07-07T19:43:27.097000", "content": "Is the \"darknet\" license valid? I think that it allows everything: https://github.com/AlexeyAB/darknet/blob/master/LICENSE. One could even train v5 with the v4 repo π€
", "votes": 3, "replies": [] }, { "id": 919365, "author_name": "Nacir Bouazizi", "author_url": "", "post_date": "2020-07-07T20:04:43.790000", "content": "@maggiemd Only winning solution has to be on MIT or all leaderboard solutions are concerned ?
", "votes": 4, "replies": [] }, { "id": 919382, "author_name": "Matt", "author_url": "", "post_date": "2020-07-07T20:21:58.267000", "content": "@seesee The issue with the GPL license is that it requires the same license to be used in all derivative works. The organizers are likely against this idea as they eventually want a closed-source version for commercial means. And yes, that license does appear to allow re-licensing under MIT as it is in the public domain.
\n\n@nacirbouazizi I doubt they will invest the time to examine each top LB solution for it's license compatibility. So sure, you could probably continue to use YoloV5 or whatever framework you want to get a nice medal but if you actually win, it would be very disappointing to get disqualified over.
", "votes": 4, "replies": [] }, { "id": 919405, "author_name": "Nacir Bouazizi", "author_url": "", "post_date": "2020-07-07T20:42:22.893000", "content": "@matthewmasters Thanks for your response. \nIt's actually sad to know this after all the hardwork and 21 days until merger deadline.\nNow I have so start from 0 and try fasterrcnn !
", "votes": 3, "replies": [] }, { "id": 919523, "author_name": "tereka", "author_url": "", "post_date": "2020-07-08T00:19:39.507000", "content": "Thanks for your response.
\nPlease announce about the official view of license in Kaggle discussion(not this thread).
\nMany competitors use yolov5(GPL-V3 License). It's a huge impact on this competiton.
Normally when we do license checks in companies, MIT(keras), Apache, BSD(pytorch) are considered as open-source licences because they are commercially available without restrictions.\nThe one you should watch out is GPL (you have to submit the source code when requested).. but this is not work!
Since the hosts are universities and not business, I'm not sure why they are not permitting GPLs. I'm not a big fan of yolov5 but I think the hosts want good and innovative solutions, not license-free solutions.
I have one specific question: can we use Mosaic augmentation? I believe the author of YOLOv3-ultralytics and YOLOv5, Glenn Jocher, came up with this idea originally. Are we allowed to write our own Mosaic augmentation then?
\n\nMosaic is very straightforward and intuitive. Other researchers have also came up with similar ideas (e.g. Sticher) before. I assume that we can play around with the wording by calling Mosaic \"modified/customized Sticher augmentation\", can't we?
", "votes": 6, "replies": [] }, { "id": 919571, "author_name": "Zikai", "author_url": "", "post_date": "2020-07-08T01:39:02.063000", "content": "Hi, thanks for replying. But I should mention that Pytorch is under BSD license, which is a little stricter than MIT license. So we should start from a deep learning framework ???
", "votes": 4, "replies": [] }, { "id": 919645, "author_name": "arutema47", "author_url": "", "post_date": "2020-07-08T03:08:12.787000", "content": "@jihangz\nthese licenses become a problem when you use codes directly (copyright).\nIf you implement the idea with your own codes, you have the copyright and of cource are free to use in kaggle competitions.
", "votes": 2, "replies": [] }, { "id": 919814, "author_name": "Tanmay Pandey", "author_url": "", "post_date": "2020-07-08T05:54:39.827000", "content": "everyone above 0.76 rn : Years of academy training wasted
", "votes": 6, "replies": [] }, { "id": 919908, "author_name": "Etienne David", "author_url": "", "post_date": "2020-07-08T07:22:31.630000", "content": "The main issue with changing the license now is that we change the rule during competitions. Maybe somebody has developed a solution compliant with the rules and it won't be fair either for him or her. The code of Yolo-V5 by ultralytics is GPL3 but not the idea (however I agree that recoding everything is not what a competitor want to do).
\n\nOnly the winning solutions are required to be MIT-licensed.
", "votes": 2, "replies": [] }, { "id": 919925, "author_name": "Peiyuan Liao", "author_url": "", "post_date": "2020-07-08T07:40:47.897000", "content": "So (hypothetically) if all competitors in the gold zone ended up using yolov5, do they all get disqualified? Or do the first compliant-ranking submission on the leader board gets the prize? And what about the medals and points? A lot of people (including me) compete for medals not prizes.
", "votes": 6, "replies": [] }, { "id": 919942, "author_name": "Tian", "author_url": "", "post_date": "2020-07-08T07:58:20.897000", "content": "Can I ask what is \"winning solutions\"? \nIn the money?
", "votes": 3, "replies": [] }, { "id": 919943, "author_name": "Matt", "author_url": "", "post_date": "2020-07-08T07:59:31.110000", "content": "Yeah I think they might have to go pretty far down the list before finding an eligible solution π
", "votes": 1, "replies": [] }, { "id": 919945, "author_name": "Peiyuan Liao", "author_url": "", "post_date": "2020-07-08T08:02:35.530000", "content": "Fun fact: I'm actually not using Yolov5 at the moment, and I believe neither @tereka is
", "votes": 0, "replies": [] }, { "id": 919947, "author_name": "", "author_url": "", "post_date": "2020-07-08T08:03:58.137000", "content": "", "votes": -1, "replies": [] }, { "id": 922088, "author_name": "from coffee import *", "author_url": "", "post_date": "2020-07-09T19:51:21.103000", "content": "Dear @alexanderliao, will you also enlighten us on what you are using ? ;) DETR? FasterRCNN, EfficientDet ? Just the Model's name would be quite interesting. \nThanks a lot! - and no hard feelings π
", "votes": -1, "replies": [] }, { "id": 922367, "author_name": "Peiyuan Liao", "author_url": "", "post_date": "2020-07-10T04:03:03.733000", "content": "It would be trolling to say that I'm using OverFeat.
", "votes": -1, "replies": [] }, { "id": 929393, "author_name": "from coffee import *", "author_url": "", "post_date": "2020-07-14T16:36:14.647000", "content": "Hi @maggiemd, the people here still don't get it, and neither am I getting it.\nCan we use YOLOv5 and we will just be excluded from WINING the \"in the money\" places, or will all YOLO submissions completely be removed from the LB? Please make a clear statement. This competition get's worse and worse.
", "votes": 0, "replies": [] }, { "id": 929513, "author_name": "Phil Culliton", "author_url": "", "post_date": "2020-07-14T18:04:01.037000", "content": "@chrisden You can't win prizes if you use YOLOv5.
", "votes": 0, "replies": [] }, { "id": 930013, "author_name": "cswwp", "author_url": "", "post_date": "2020-07-15T06:18:26.193000", "content": "So yolov5 at top area will get medals normally?
", "votes": 1, "replies": [] } ] }, { "id": 911848, "author_name": "Matt", "author_url": "", "post_date": "2020-07-02T04:07:37.377000", "content": "Thatβs a good question. From my understanding, the rules require the solution to be published under an MIT license. And since a GPL license is more restrictive (as it requires derivatives to be open sourced), it may not be allowed for this competition. Though maybe since most people are simply using the framework, not creating derivatives, it may be fine? Iβm certainly no expert so donβt take this information as fact.
", "votes": 5, "replies": [ { "id": 911864, "author_name": "Nirjhar Roy", "author_url": "", "post_date": "2020-07-02T04:31:17.883000", "content": "Yes even I had that question. Ofcourse we are finetuning. We also dont know if organizer wants to be restricted to open source their solutions once they adopt the top solution (if it involves yolov5) . I have very less idea about legality and licensing myself . So thought of asking the forum .
", "votes": 2, "replies": [] } ] }, { "id": 912554, "author_name": "Neuron Engineer", "author_url": "", "post_date": "2020-07-02T15:05:18.257000", "content": "Thanks for the good question. Tagging the hosts @bendvd @stavness @maggiemd @philculliton for clarification .
\n\nupdated July 5 I emailed Phil just to make sure that he is notified on this thread, but not yet reply... I will update instantly if I get reply
\n\nupdated July 6 Not yet reply. Maybe I will also email other hosts. @phoenix9032 Nirjhar, to get attention from hosts, could you please try renaming this topic e.g. \"Hosts answer needed : GPL-3 licence on YoloV5\"
\n\nupdated July 7 Thailand early morning We got the kind reply here... Thanks the hosts ... I haven't emailed any other hosts :)
", "votes": 6, "replies": [ { "id": 915854, "author_name": "Nirjhar Roy", "author_url": "", "post_date": "2020-07-05T06:38:27.433000", "content": "Thanks @ratthachat
", "votes": 1, "replies": [] }, { "id": 917035, "author_name": "Nirjhar Roy", "author_url": "", "post_date": "2020-07-06T07:33:47.890000", "content": "Done @ratthachat
", "votes": 2, "replies": [] }, { "id": 917414, "author_name": "Phil Culliton", "author_url": "", "post_date": "2020-07-06T13:34:53.460000", "content": "Hi! July 2-5 was a holiday in the US. Also July 5 was a Sunday. It's 9:30 AM where I am right now, so I'm just getting into work. No need to email other people! Maggie and I just haven't been in the office.
\n\nThanks for your questions! We'll discuss ASAP with the host.
", "votes": 5, "replies": [] }, { "id": 917447, "author_name": "Nirjhar Roy", "author_url": "", "post_date": "2020-07-06T14:17:21.197000", "content": "@philculliton As Richard Marx once said \" We will be right here waiting for you \" :) \nBy the way , hope you had a good time in long weekend and we would be patient for your response and consideration.
", "votes": 1, "replies": [] } ] }, { "id": 923984, "author_name": "Rony K Roy", "author_url": "", "post_date": "2020-07-11T07:59:28.553000", "content": "A lot of people hem and haw about the licensing issues...\nHowever when you look at it from the host's point of view...\nIt makes sense... \nHost a competition... money goes only to the solution that is MIT licensed...\nGPL license necessarily implies that the host or anyone else must necessarily share improvements made over and above the winning solution... { distribution closed source versions is prohibited }
\n\nThis is going to be a thorny issue going forward, for kaggle.\nThe optics of it simply doesn't look good, whichever way you spin it...
\n\nUpside: increased awareness of licensing... among the community.
", "votes": 3, "replies": [] }, { "id": 920379, "author_name": "Ahmmad Rashid", "author_url": "", "post_date": "2020-07-08T14:55:09.343000", "content": "I think kaggle administrators needs to start looking into licensing issues for different competition.\nAs kaggle platform is very popular now, ultimately participants are going to face more lincesing issues in the future.\nOne of the best example is https://www.kaggle.com/c/deepfake-detection-challenge/discussion/157983 challenge. I feel bad for them.
", "votes": 3, "replies": [] }, { "id": 920239, "author_name": "Ahmmad Rashid", "author_url": "", "post_date": "2020-07-08T13:00:35.733000", "content": "So basically if you are in the golden zone with YOLOv5 its fine but if you win using the solution with YOLOv5 you are screwed :p.\nNow I am feeling bad for the top 10 in the leader board.
", "votes": 3, "replies": [ { "id": 920248, "author_name": "Nirjhar Roy", "author_url": "", "post_date": "2020-07-08T13:05:50.397000", "content": "They will face the same unfortunate fate like Mr. Giba faced here
\n\nhttps://www.kaggle.com/c/deepfake-detection-challenge/discussion/157983
", "votes": 4, "replies": [] }, { "id": 921552, "author_name": "Zaber Ibn Abdul Hakim", "author_url": "", "post_date": "2020-07-09T11:25:23.313000", "content": "This gentleman lost 500,000$ just like that. π°\nHow unfortunate!!!
", "votes": 1, "replies": [] }, { "id": 930032, "author_name": "Tian", "author_url": "", "post_date": "2020-07-15T06:36:38.683000", "content": "π π π
", "votes": -1, "replies": [] } ] }, { "id": 919522, "author_name": "Neuron Engineer", "author_url": "", "post_date": "2020-07-08T00:19:24.157000", "content": "Okay I must admit that this is bad news guys. But life has to move on. \nNow we should go back and carefully check licence of YoloV4 / EffDet / ResNest too.\nGood luck everybody.
", "votes": 4, "replies": [ { "id": 919928, "author_name": "Tian", "author_url": "", "post_date": "2020-07-08T07:45:19.840000", "content": "EffDet use Apache-2.0 License, so not allowed neither?
", "votes": 2, "replies": [] }, { "id": 919933, "author_name": "Nirjhar Roy", "author_url": "", "post_date": "2020-07-08T07:51:37.777000", "content": "Apache v2.0 is very open license . That's what the term \"Fit for MIT license \" not \" Only MIT license \" .
", "votes": 2, "replies": [] }, { "id": 922351, "author_name": "27Wilson", "author_url": "", "post_date": "2020-07-10T03:36:06.267000", "content": "My repo changed torchvision's sourcecode, so it is also can't be used?
", "votes": 0, "replies": [] } ] }, { "id": 919454, "author_name": "Thomas Di Martino", "author_url": "", "post_date": "2020-07-07T21:45:56.630000", "content": "As the LB is populated with a lot of YoloV5 solutions, it will make it harder to situate our results for other algorithms if we cannot make them perform as well as YoloV5... As far as I know, I didnt see top performers with EfficientDet or Faster RCNN.. Maybe we should try other Yolo versions ?
", "votes": 3, "replies": [ { "id": 919567, "author_name": "Tian", "author_url": "", "post_date": "2020-07-08T01:25:42.637000", "content": "yes. now, public board means nothing.
", "votes": 2, "replies": [] } ] }, { "id": 919958, "author_name": "Vadim Timakin", "author_url": "", "post_date": "2020-07-08T08:14:47.827000", "content": "When I came to this competition I was really surprised just cause I've realized that almost everyone uses Yolo. That's a little bit upset cause diversity is reduced to zero. Half of all people just use the following submit baseline: https://www.kaggle.com/nvnnghia/yolov5-pseudo-labeling (I am not an exception but I use my own baseline at the same time). One more interesting fact: almost all the people who placed in the gold now did their submits before this thread has been created and EVEN before the following discussion has been started: https://www.kaggle.com/c/global-wheat-detection/discussion/163291 (this is the source of the future shake up). Probably, the hosts just weren't ready for that. Apparently, they should understand it before a competition has started. But they haven't done it and now they need to complete a huge work for filtering all the solutions.
", "votes": 1, "replies": [ { "id": 925308, "author_name": "Redwan Sony", "author_url": "", "post_date": "2020-07-12T02:16:42.920000", "content": "@vadimtimakin Thanks for pointing this out.. π
", "votes": 0, "replies": [] } ] }, { "id": 919625, "author_name": "Benlei Cui", "author_url": "", "post_date": "2020-07-08T02:46:32.217000", "content": "I wonder the ones only in medal zone will submit the licenses too? Or the ones in medal zone can use YOLO v5
", "votes": 1, "replies": [ { "id": 919901, "author_name": "Etienne David", "author_url": "", "post_date": "2020-07-08T07:18:42.913000", "content": "Only winning solutions are required to be MIT licensed
", "votes": 2, "replies": [] }, { "id": 920000, "author_name": "Pak Lau9", "author_url": "", "post_date": "2020-07-08T08:39:29.070000", "content": "Both Mmdetection and Detectron2 are Apache-2.0 License, so does that exclude them too β¦ ?! This would be an amazing competition without the most commonly used libraries.
", "votes": 0, "replies": [] }, { "id": 920295, "author_name": "Psi", "author_url": "", "post_date": "2020-07-08T13:35:57.850000", "content": "And how can you know upfront if you will win? :) Will you be disqualified if you use YoloV5 on position 3 but not if you use it on position 4?
", "votes": 5, "replies": [] }, { "id": 920302, "author_name": "Benlei Cui", "author_url": "", "post_date": "2020-07-08T13:41:57.880000", "content": "Thanks for reply :D
", "votes": 0, "replies": [] }, { "id": 920326, "author_name": "", "author_url": "", "post_date": "2020-07-08T14:00:40.683000", "content": "", "votes": 9, "replies": [] }, { "id": 920486, "author_name": "Matt", "author_url": "", "post_date": "2020-07-08T16:19:25.843000", "content": "@sokazaki while that would be nice, that would involve an insane amount of work on the part of the competition organizers and compliance team, itβs simply unrealistic for this competition
", "votes": 0, "replies": [] } ] }, { "id": 919510, "author_name": "cooleel", "author_url": "", "post_date": "2020-07-07T23:30:38.917000", "content": "A little disappointed about the answer, but at least we know what is not good to do now..π \nThanks for bringing up this question @phoenix9032
", "votes": 2, "replies": [ { "id": 919589, "author_name": "Nirjhar Roy", "author_url": "", "post_date": "2020-07-08T01:59:07.410000", "content": "It was a very harmless question to start with.Honestly I thought it would be just someone answering me \"Its already clarified , it is okay \" . I feel a bit stupid now .
", "votes": 5, "replies": [] }, { "id": 919594, "author_name": "Matt", "author_url": "", "post_date": "2020-07-08T02:06:49.077000", "content": "@phoenix9032 No need to feel stupid, it was a great question!
\n\nIf you hadn't asked, many people would have continued spending their time on an approach that would not be allowed in the end. The way I see it, you saved a lot of people from wasting time and potential disappointment. And if anything, I'm the stupid one for not considering the implications of different licenses. So thank you for bringing this up! π
", "votes": 3, "replies": [] }, { "id": 921274, "author_name": "Etienne David", "author_url": "", "post_date": "2020-07-09T07:13:07.873000", "content": "It was a great question and show that choosing the right open-source licence for a competition is not as easy as we thought !
", "votes": 1, "replies": [] } ] }, { "id": 921577, "author_name": "Prateek Gupta", "author_url": "", "post_date": "2020-07-09T11:53:39.083000", "content": "Hi All,
\n\nI think we can use AlexeyAB/darknet/YOLOv4. See it's license here-\nLicense
\n\nFor details please check my comment in below discussion-\ndiscussion
", "votes": 0, "replies": [] }, { "id": 930035, "author_name": "cswwp", "author_url": "", "post_date": "2020-07-15T06:40:48.183000", "content": "So what's the result? yolov5 disallow? or medals area yolov5 will get medal but can't win prize? @phoenix9032
", "votes": 0, "replies": [ { "id": 930067, "author_name": "Rifat Rashid", "author_url": "", "post_date": "2020-07-15T07:05:52", "content": "if you are using YOLOv5 your disqualified. Also, you will not receive any medals. So dont use YOLOv5. Use of YOLOv5 forbidden in Kaggle
", "votes": 0, "replies": [] }, { "id": 939082, "author_name": "cswwp", "author_url": "", "post_date": "2020-07-22T02:48:57.027000", "content": "Thank you, disgusted by this gameπ€
", "votes": -2, "replies": [] } ] }, { "id": 939271, "author_name": "", "author_url": "", "post_date": "2020-07-22T06:10:32.777000", "content": "", "votes": -1, "replies": [ { "id": 939284, "author_name": "Matt", "author_url": "", "post_date": "2020-07-22T06:23:47.820000", "content": "Read it again, it states
\n\n\n\n\nCompetitions are open to residents of the United States and worldwide, except that if you are a resident of Crimea, Cuba, Iran, Syria, North Korea, Sudan, or are subject to U.S. export controls or sanctions, you may not enter the Competition.
\n
Only a few countries are excluded and itβs because the US itself has sanctions against these countries, it is not Kaggles doing
", "votes": 1, "replies": [] } ] }, { "id": 919999, "author_name": "", "author_url": "", "post_date": "2020-07-08T08:38:38.017000", "content": "", "votes": 0, "replies": [ { "id": 920031, "author_name": "Matt", "author_url": "", "post_date": "2020-07-08T08:55:08.133000", "content": "", "votes": 0, "replies": [] } ] }, { "id": 919337, "author_name": "", "author_url": "", "post_date": "2020-07-07T19:47:38.103000", "content": "", "votes": 2, "replies": [ { "id": 919342, "author_name": "Nirjhar Roy", "author_url": "", "post_date": "2020-07-07T19:52:27.950000", "content": "YOLO is not out . YOLOv5 is alone GPLV3
", "votes": 3, "replies": [] }, { "id": 919899, "author_name": "", "author_url": "", "post_date": "2020-07-08T07:16:38.617000", "content": "", "votes": 2, "replies": [] } ] } ], "raw_markdown_by_id": { "911829": "Hello All \n\nI have just joined this competition and have no prior experience in Yolo . Please forgive me if this question is already deliberated .\nI have noticed that the licensing term of Yolov5 and Yolo V4/V3 are little different . V5 license is GPL3.0 I guess which is bit more restrictive than the other opensource licenses . Are competitors allowed to use it in final solution ? (Especially the ones going to be in prize zone )\n\nTagging hosts : @philculliton @bendvd @maggiemd @stavness", "920645": "All, \n\nThis Global Wheat Detection competition, and this thread in particular were recently brought to my attention here https://github.com/ultralytics/yolov5/issues/317.\n\nI'm very happy that the community is finding YOLOv5 useful!! I personally have not participated in any Kaggle competitions, but I did participate in a National Geospatial-Intelligence Agency (NGA) competition that got me interested in YOLO and propelled me down my ML path a couple summers ago. I created a repo at the time to document my efforts:\nhttps://github.com/ultralytics/xview-yolov3\n\nIt's a nice irony now to see the situation reversed, with other young people learning and using YOLOv5 to start their own research and competition efforts. I'm very humbled and hopeful that smart minds might use what we have developed as the basis for further advancements of their own in the future :)\n\nRegarding the licensing issues, both of our YOLO repositories, https://github.com/ultralytics/yolov3 and https://github.com/ultralytics/yolov5 are licensed as GPL 3.0. The intention of the license is allow full use for any purpose (academic, research or commercial) of the original or modified code, provided that derivative works are open sourced and licensed as the same. The greater good we have in mind here is the capability of all people to benefit from advancements made with future open-sourced works that begin from or leverage YOLOv5 as a starting point.\n\nUltralytics was founded to work on particle physics applications, we sleep with one eye open towards expanding knowledge and advancing science. We are always open to new ideas, but we believe we have the right license in place for our work. Isaac Newton once said **\"If I have seen further it is by standing on the shoulders of Giants.\"** I believe that if we allow future works to hide from the world as proprietary code, which other licenses provide provisions for, we are reducing the possible giants that future Isaac Newtons may stand on, and diminishing all of our own futures slightly so.\n\n", "919568": "kaggle or someone needs to to make a checking web page like this:\n\n\n", "922311": "Hi. I'm confused about the competitions to rule. \nIs my understanding correct? \n\nPrize/conference: not GPL-3 license \nmedal: GPL-3 and others are ok.\n\nCan I choose 2 submissions(for Prize and for medal) within rule If both are correct?", "914432": "@phoenix9032 If I was interpreting the rules as written\n\n```\nOpen Source: You hereby license and will license your winning Submission and the source code used to generate the Submission under an Open Source Initiative-approved license (see www.opensource.org) that in no event limits commercial use of such code or model containing or depending on such code. For generally commercially available software that you used to generate your submission that is not owned by you, but that can be procured by the Competition Sponsor without undue expense, you do not need to grant the license in the preceding sentence for that software..```\n\nThen no, you would not be allowed to use Yolov5 as it is GPL 3 and GPL 3 'limits commercial use of such code'. Yet many have posted use of yolov5 in the external data thread with no comment from the organisers leading to a potential issue at the end (once again). There should be more proactive monitoring of this.\n\nThere are also mentions of \n* https://github.com/zylo117/Yet-Another-EfficientDet-Pytorch - LGPL, likely not limiting commercial use of such code if you ONLY `import`, but if you modify the code in any way it would no longer be useable commercially without releasing the code and all changes\n\n* https://pytorch.org/hub/facebookresearch_WSL-Images_resnext/ - explicitly licensed by Facebook as CC-BY-NC 4.0 as with their other SSL/SWSL models\n\nEDIT: GPL dependant code is technically releasable commercially, but requires releasing all of the other code and also releasing under GPL, it is viral. In most commerical use wrt to the term 'limit' I'd consider as limiting having done many code audits for companies releasing commercial software and they never want the limitation of releasing all of their code. Either way, 'limit' should be explicitly defined in this context.", "917308": "Thanks for your question @phoenix9032 and thanks for your insights @rwightman . It is a very interesting question, as I was more used to MIT licence. I'm not an expert but GPL seems good for me. I will confirm by tomorrow.", "922162": "Hard solution: carefully select packages that are \"MIT compatible\".\nSimple solution: only submit models that don't win. ;) ", "919074": "Thanks for the question, we've reviewed the rules and have clarified that the winning solution needs to fit the MIT license. Any data or code with a GPL license would not be allowed in the final solution. ", "911848": "Thatβs a good question. From my understanding, the rules require the solution to be published under an MIT license. And since a GPL license is more restrictive (as it requires derivatives to be open sourced), it may not be allowed for this competition. Though maybe since most people are simply using the framework, not creating derivatives, it may be fine? Iβm certainly no expert so donβt take this information as fact.", "912554": "Thanks for the good question. Tagging the hosts @bendvd @stavness @maggiemd @philculliton for clarification .\n\n**updated July 5** I emailed Phil just to make sure that he is notified on this thread, but not yet reply... I will update instantly if I get reply\n\n**updated July 6** Not yet reply. Maybe I will also email other hosts. @phoenix9032 Nirjhar, to get attention from hosts, could you please try renaming this topic e.g. \"Hosts answer needed : GPL-3 licence on YoloV5\" \n\n**updated July 7 Thailand early morning** We got the kind reply here... Thanks the hosts ... I haven't emailed any other hosts :)", "923984": "A lot of people hem and haw about the licensing issues...\nHowever when you look at it from the host's point of view...\nIt makes sense... \nHost a competition... money goes only to the solution that is MIT licensed...\nGPL license necessarily implies that the host or anyone else must necessarily share improvements made over and above the winning solution... { distribution closed source versions is prohibited }\n\nThis is going to be a thorny issue going forward, for kaggle.\nThe optics of it simply doesn't look good, whichever way you spin it...\n\nUpside: increased awareness of licensing... among the community.", "920379": "I think kaggle administrators needs to start looking into licensing issues for different competition.\nAs kaggle platform is very popular now, ultimately participants are going to face more lincesing issues in the future.\nOne of the best example is https://www.kaggle.com/c/deepfake-detection-challenge/discussion/157983 challenge. I feel bad for them. ", "920239": "So basically if you are in the golden zone with YOLOv5 its fine but if you win using the solution with YOLOv5 you are screwed :p.\nNow I am feeling bad for the top 10 in the leader board.", "919522": "Okay I must admit that this is bad news guys. But life has to move on. \nNow we should go back and carefully check licence of YoloV4 / EffDet / ResNest too.\nGood luck everybody. ", "919454": "As the LB is populated with a lot of YoloV5 solutions, it will make it harder to situate our results for other algorithms if we cannot make them perform as well as YoloV5... As far as I know, I didnt see top performers with EfficientDet or Faster RCNN.. Maybe we should try other Yolo versions ?", "919958": "When I came to this competition I was really surprised just cause I've realized that almost everyone uses Yolo. That's a little bit upset cause diversity is reduced to zero. Half of all people just use the following submit baseline: https://www.kaggle.com/nvnnghia/yolov5-pseudo-labeling (I am not an exception but I use my own baseline at the same time). One more interesting fact: almost all the people who placed in the gold now did their submits before this thread has been created and EVEN before the following discussion has been started: https://www.kaggle.com/c/global-wheat-detection/discussion/163291 (this is the source of the future shake up). Probably, the hosts just weren't ready for that. Apparently, they should understand it before a competition has started. But they haven't done it and now they need to complete a huge work for filtering all the solutions.", "919625": "I wonder the ones only in medal zone will submit the licenses too? Or the ones in medal zone can use YOLO v5", "919510": "A little disappointed about the answer, but at least we know what is not good to do now..π \nThanks for bringing up this question @phoenix9032 ", "921577": "Hi All,\n\nI think we can use [AlexeyAB/darknet/YOLOv4](https://github.com/AlexeyAB/darknet). See it's license here-\n[License](https://github.com/AlexeyAB/darknet/blob/master/LICENSE)\n\nFor details please check my comment in below discussion-\n[discussion](https://www.kaggle.com/c/global-wheat-detection/discussion/164845)", "930035": "So what's the result? yolov5 disallow? or medals area yolov5 will get medal but can't win prize? @phoenix9032", "939271": "", "919999": "", "919337": "" } }