![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F644036%2F2b570569b038b6d8a9bb5c0ff23461cf%2FScreenshot%20from%202023-04-26%2019-54-53.png?generation=1682553338447098&alt=media\")
I'd like to address some of the questions related to \"External Data\". External Data needs to allow you to license the winning solution, i.e. being used in commercial purpose.
\nHere below please see the relevant rules for details:
\n\n\n\n
\n- WINNER LICENSE.
\n
\n Under Section 11 (Winners Obligations) of the General Rules below, you hereby grant and will grant Competition Sponsor the following license(s) with respect to your Submission if you are a Competition winner:
\n Open Source: You hereby license and will license your winning Submission and the source code used to generate the Submission under an Apache 2.0 License that in no event limits commercial use of such code or model containing or depending on such code. To the extent your Submission makes use of generally commercially available software not owned by you that you used to generate your submission, but that can be procured by the Competition Sponsor without undue expense, you do not grant the license in the preceding sentence to that software.
\n …
\n C. External Data. You may use data other than the Competition Data (“External Data”) to develop and test your Submissions. However, you will ensure the External Data is publicly available and equally accessible to use by all participants of the Competition for purposes of the competition at no cost to the other participants. The ability to use External Data under this Section 7.C (External Data) does not limit your other obligations under these Competition Rules, including but not limited to Section 11 (Winners Obligations).
\n …- WINNERS OBLIGATIONS.
\n
\n As a condition to being awarded a Prize, a Prize winner must fulfill the following obligations:
\n (a) deliver to the Competition Sponsor the final model's software code as used to generate the winning Submission and associated documentation. The delivered software code should follow these documentation guidelines, must be capable of generating the winning Submission, and contain a description of resources required to build and/or run the executable code successfully. To the extent that the final model’s software code includes generally commercially available software that is not owned by you, but that can be procured by the Competition Sponsor without undue expense, then instead of delivering the code for that software to the Competition Sponsor, you must identify that software, method for procuring it, and any parameters or other information necessary to replicate the winning Submission;
\n (b) grant to the Competition Sponsor the license to the winning Submission stated in the Competition Specific Rules above, and represent that you have the unrestricted right to grant that license
With that, you may use any external data as long as it is public and commercially available. Thank you!
", "messages": [ { "id": 2210907, "postDate": "2023-04-05T17:30:16.680Z", "content": "I'd like to address some of the questions related to \"External Data\". External Data needs to allow you to license the winning solution, i.e. being used in commercial purpose.
\nHere below please see the relevant rules for details:
\n\n\n\n
\n- WINNER LICENSE.
\n
\n Under Section 11 (Winners Obligations) of the General Rules below, you hereby grant and will grant Competition Sponsor the following license(s) with respect to your Submission if you are a Competition winner:
\n Open Source: You hereby license and will license your winning Submission and the source code used to generate the Submission under an Apache 2.0 License that in no event limits commercial use of such code or model containing or depending on such code. To the extent your Submission makes use of generally commercially available software not owned by you that you used to generate your submission, but that can be procured by the Competition Sponsor without undue expense, you do not grant the license in the preceding sentence to that software.
\n …
\n C. External Data. You may use data other than the Competition Data (“External Data”) to develop and test your Submissions. However, you will ensure the External Data is publicly available and equally accessible to use by all participants of the Competition for purposes of the competition at no cost to the other participants. The ability to use External Data under this Section 7.C (External Data) does not limit your other obligations under these Competition Rules, including but not limited to Section 11 (Winners Obligations).
\n …- WINNERS OBLIGATIONS.
\n
\n As a condition to being awarded a Prize, a Prize winner must fulfill the following obligations:
\n (a) deliver to the Competition Sponsor the final model's software code as used to generate the winning Submission and associated documentation. The delivered software code should follow these documentation guidelines, must be capable of generating the winning Submission, and contain a description of resources required to build and/or run the executable code successfully. To the extent that the final model’s software code includes generally commercially available software that is not owned by you, but that can be procured by the Competition Sponsor without undue expense, then instead of delivering the code for that software to the Competition Sponsor, you must identify that software, method for procuring it, and any parameters or other information necessary to replicate the winning Submission;
\n (b) grant to the Competition Sponsor the license to the winning Submission stated in the Competition Specific Rules above, and represent that you have the unrestricted right to grant that license
With that, you may use any external data as long as it is public and commercially available. Thank you!
", "rawMarkdown": "I'd like to address some of the questions related to \"External Data\". External Data needs to allow you to license the winning solution, i.e. being used in commercial purpose. \n\nHere below please see the relevant rules for details:\n>1. WINNER LICENSE.\nUnder Section 11 (Winners Obligations) of the General Rules below, you hereby grant and will grant Competition Sponsor the following license(s) with respect to your Submission if you are a Competition winner:\nOpen Source: You hereby license and will license your winning Submission and the source code used to generate the Submission under an Apache 2.0 License that in no event limits commercial use of such code or model containing or depending on such code. To the extent your Submission makes use of generally commercially available software not owned by you that you used to generate your submission, but that can be procured by the Competition Sponsor without undue expense, you do not grant the license in the preceding sentence to that software.\n...\n>C. External Data. You may use data other than the Competition Data (“External Data”) to develop and test your Submissions. However, you will ensure the External Data is publicly available and equally accessible to use by all participants of the Competition for purposes of the competition at no cost to the other participants. The ability to use External Data under this Section 7.C (External Data) does not limit your other obligations under these Competition Rules, including but not limited to Section 11 (Winners Obligations). \n...\n>11. WINNERS OBLIGATIONS.\nAs a condition to being awarded a Prize, a Prize winner must fulfill the following obligations:\n(a) deliver to the Competition Sponsor the final model's software code as used to generate the winning Submission and associated documentation. The delivered software code should follow these documentation guidelines, must be capable of generating the winning Submission, and contain a description of resources required to build and/or run the executable code successfully. To the extent that the final model’s software code includes generally commercially available software that is not owned by you, but that can be procured by the Competition Sponsor without undue expense, then instead of delivering the code for that software to the Competition Sponsor, you must identify that software, method for procuring it, and any parameters or other information necessary to replicate the winning Submission;\n(b) grant to the Competition Sponsor the license to the winning Submission stated in the Competition Specific Rules above, and represent that you have the unrestricted right to grant that license\n\nWith that, you may use any external data as long as it is public and commercially available. Thank you!", "votes": 2 }, { "id": 2236479, "postDate": "2023-04-26T22:34:53.843Z", "content": "Is synthetic data allowed and if so, what are the rules with regard to public access when creating the synthetic data requires code and involves solving part of the challenge (e.g. there could be models use to create the synthetic data)?
", "rawMarkdown": "Is synthetic data allowed and if so, what are the rules with regard to public access when creating the synthetic data requires code and involves solving part of the challenge (e.g. there could be models use to create the synthetic data)?" }, { "id": 2235793, "postDate": "2023-04-26T10:54:52.343Z", "content": "Is data under Computational Use of Data Agreement (C-UDA) license falls into the category of allowed for commercial use?
", "rawMarkdown": "Is data under Computational Use of Data Agreement ([C-UDA](https://spdx.org/licenses/C-UDA-1.0.html)) license falls into the category of allowed for commercial use?", "replies": [ { "id": 2236477, "postDate": "2023-04-26T22:33:09.063Z", "content": "When it comes to the result, yes. The C-UDA license doesn't allow the data itself to be used for commercial use but allows the results (models) trained against the data to be used for commercial. Here's the annotated version https://github.com/microsoft/Computational-Use-of-Data-Agreement/blob/master/C-UDA-1.0_annotated.md.
", "rawMarkdown": "When it comes to the result, yes. The C-UDA license doesn't allow the data itself to be used for commercial use but allows the results (models) trained against the data to be used for commercial. Here's the annotated version https://github.com/microsoft/Computational-Use-of-Data-Agreement/blob/master/C-UDA-1.0_annotated.md.", "replies": [ { "id": 2236491, "postDate": "2023-04-26T22:52:11Z", "content": "From https://github.com/dxli94/WLASL :
\n\n\nAll the WLASL data is intended for academic and computational use only. No commercial usage is allowed.
\n
WLASL which is under C-UDA does not allow for commercial use and they're pretty straight-forward about it.
", "rawMarkdown": "From https://github.com/dxli94/WLASL :\n> All the WLASL data is intended for academic and computational use only. No commercial usage is allowed. \n\nWLASL which is under C-UDA does not allow for commercial use and they're pretty straight-forward about it. ", "votes": 1 }, { "id": 2236510, "postDate": "2023-04-26T23:40:29.750Z", "content": "C-UDA doesn’t allow the data for commercial use but does allow the results for commercial use.
\n“””
\n5.5. “Result” means anything that you develop or improve from your use of Data that does not include more than a de minimis portion of the Data on which the use is based. Results may include de minimis portions of the Data necessary to report on or explain use that has been conducted with the Data, such as figures in scientific papers, but do not include more. Artificial intelligence models trained on Data (and which do not include more than a de minimis portion of Data) are Results.
The results are allowed for commercial usage. The license protects distribution of the data not the results.
\nIt is very straightforward that it is allowed.
", "rawMarkdown": "C-UDA doesn’t allow the data for commercial use but does allow the results for commercial use.\n\n“””\n5.5. “Result” means anything that you develop or improve from your use of Data that does not include more than a de minimis portion of the Data on which the use is based. Results may include de minimis portions of the Data necessary to report on or explain use that has been conducted with the Data, such as figures in scientific papers, but do not include more. **Artificial intelligence models trained on Data (and which do not include more than a de minimis portion of Data) are Results.**\n\nThe results are allowed for commercial usage. The license protects distribution of the data not the results.\n\n**It is very straightforward that it is allowed.**", "votes": -1 }, { "id": 2236511, "postDate": "2023-04-26T23:41:58.173Z", "content": "“
\n2.2 The C-UDA does not impose any restriction with respect to the use, modification, or distribution of Results.
\n“
Unless I'm misunderstanding, WLASL is just a collection of publically available videos and their associated sign/labels. Does it really matter what licence they use? What you really need is approval from the original data source.
\nJust look at the WLASL script: https://github.com/dxli94/WLASL/blob/master/start_kit/video_downloader.py
\nEach video is listed in the json here: https://raw.githubusercontent.com/dxli94/WLASL/master/start_kit/WLASL_v0.3.json
There are many sources: aslpro.com, youtube, handspeak.com
\nYou would need to prove all these sources provide their videos free for commercial use.
\nJust taking www.handspeak.com as an example, their website explicitly says \"NO AI\" here: https://www.handspeak.com/info/index.php?info=terms
\n
I'm not using it.
", "rawMarkdown": "Unless I'm misunderstanding, WLASL is just a collection of publically available videos and their associated sign/labels. Does it really matter what licence they use? What you really need is approval from the original data source.\n\nJust look at the WLASL script: https://github.com/dxli94/WLASL/blob/master/start_kit/video_downloader.py\nEach video is listed in the json here: https://raw.githubusercontent.com/dxli94/WLASL/master/start_kit/WLASL_v0.3.json\n\nThere are many sources: aslpro.com, youtube, handspeak.com\n\nYou would need to prove all these sources provide their videos free for commercial use.\n\nJust taking www.handspeak.com as an example, their website explicitly says \"NO AI\" here: https://www.handspeak.com/info/index.php?info=terms\n\n\n\nI'm not using it." }, { "id": 2236518, "postDate": "2023-04-27T00:04:02.273Z", "content": "There is a different between commercial data use and public data access.
\nThe requirement of the challenge for data is “publicly available and equally accessible” data “at no cost”.
\nThe license of WSASL data is governed under C-UDA which allows “results”, models created from this data to be utilized for commercial purposes but does not allow the data itself to be used for commercial purposes (eg I can’t sell someone else YouTube video for sign language courses).
\nCurators of WSASL cannot license the “data” commercially because they don’t own it but its not necessary because we aren’t distributing the data here but using it for computation use which is its intent.
", "rawMarkdown": "There is a different between commercial data use and public data access.\n\nThe requirement of the challenge for data is “publicly available and equally accessible” data “at no cost”.\n\nThe license of WSASL data is governed under C-UDA which allows “results”, models created from this data to be utilized for commercial purposes but does not allow the data itself to be used for commercial purposes (eg I can’t sell someone else YouTube video for sign language courses).\n\nCurators of WSASL cannot license the “data” commercially because they don’t own it but its not necessary because we aren’t distributing the data here but using it for computation use which is its intent. " }, { "id": 2236529, "postDate": "2023-04-27T00:21:17.453Z", "content": "I love this logic. The ultimate loophole:
\n….now it's allowed?
", "rawMarkdown": "I love this logic. The ultimate loophole:\n\n1. Find a source that explicitly forbids downloading their data to train ML models.\n2. Create a script that scrapes that source.\n3. Release the script under a C-UDA licence.\n\n....now it's allowed?" }, { "id": 2236538, "postDate": "2023-04-27T00:31:35.333Z", "content": "I tried adding WLASL to my training set for fun, but did not increase CV/LB.
\nWhat about you guys?
Not sure what source you are referring to that “explicitly forbids downloading their data to train ML models”, but it’s not the WSASL source. All content they have scraped is licensed under C-UDA. They also have a means for content creators to request removing the data if a content creator doesn’t want it included. But not sure why folks in deaf community creating this content would want to not help the deaf community. At any rate, the terms of the challenge allows any dataset to be used that is “publicly accessible and equally available” at “no cost”.
", "rawMarkdown": "Not sure what source you are referring to that “explicitly forbids downloading their data to train ML models”, but it’s not the WSASL source. All content they have scraped is licensed under C-UDA. They also have a means for content creators to request removing the data if a content creator doesn’t want it included. But not sure why folks in deaf community creating this content would want to not help the deaf community. At any rate, the terms of the challenge allows any dataset to be used that is “publicly accessible and equally available” at “no cost”." }, { "id": 2236540, "postDate": "2023-04-27T00:34:08.310Z", "content": "I haven’t added it yet. It has very low samples per sign for the WLASL-2k set but it’s on the roadmap.
", "rawMarkdown": "I haven’t added it yet. It has very low samples per sign for the WLASL-2k set but it’s on the roadmap." }, { "id": 2236543, "postDate": "2023-04-27T00:38:28.763Z", "content": "Unfortunately we could not provide any legal guidance or interpretation here. However, think of it as baking a cake (winning solutions) using different ingredients (data, model training, etc.), if those ingredients used could not allow you to license the winning solutions to \"WINNER LICENSE TYPE: Apache 2.0\" as stated in Rules page, the ingredients should/ could not be used in this case. Hope it helps. Thanks for your participation!
", "rawMarkdown": "Unfortunately we could not provide any legal guidance or interpretation here. However, think of it as baking a cake (winning solutions) using different ingredients (data, model training, etc.), if those ingredients used could not allow you to license the winning solutions to \"WINNER LICENSE TYPE: Apache 2.0\" as stated in Rules page, the ingredients should/ could not be used in this case. Hope it helps. Thanks for your participation!", "votes": 1 }, { "id": 2236545, "postDate": "2023-04-27T00:45:38.810Z", "content": "It increased my CV when I used it locally.
", "rawMarkdown": "It increased my CV when I used it locally.", "votes": 1 }, { "id": 2236664, "postDate": "2023-04-27T03:20:20.543Z", "content": "As Joseph mentioned, the legal interpretation of \"commercial use\" can be sometimes tricky, this is why I want to have a definitive statement from hosts if something can be used for the competition or not
\nIt wouldn't be the case if external data was prohibited, however it's allowed and most of the datasets are released under non-commercial licenses
\nSince I think that the status of the dataset is defined by it's license only, I would like to have an answer to the original question
", "rawMarkdown": "As Joseph mentioned, the legal interpretation of \"commercial use\" can be sometimes tricky, this is why I want to have a definitive statement from hosts if something can be used for the competition or not\n\nIt wouldn't be the case if external data was prohibited, however it's allowed and most of the datasets are released under non-commercial licenses\n\nSince I think that the status of the dataset is defined by it's license only, I would like to have an answer to the original question" }, { "id": 2237245, "postDate": "2023-04-27T13:32:40.033Z", "content": "Thanks @ashleychow - I really do love cake analogies, and am now feeling a bit hungry 😊.
\nIn this case I think we've identified a specific ingredient that people are unsure if we are able to use. I understand you can't give legal guidance, but do you know of anyone who could clarify if this specific ingredient meets the requirements? Totally understand if you can't answer that but figured I'd ask so we can hopefully avoid a similar situation as what happened in the deepfake competition.
\nI apprecaite the work you all do to host competitions like this.
", "rawMarkdown": "Thanks @ashleychow - I really do love cake analogies, and am now feeling a bit hungry 😊.\n\nIn this case I think we've identified a specific ingredient that people are unsure if we are able to use. I understand you can't give legal guidance, but do you know of anyone who could clarify if this specific ingredient meets the requirements? Totally understand if you can't answer that but figured I'd ask so we can hopefully avoid a similar situation as what happened in the deepfake competition.\n\nI apprecaite the work you all do to host competitions like this.", "votes": 1 }, { "id": 2239835, "postDate": "2023-04-29T23:46:46.337Z", "content": "@ashleychow How will the competition winner be determined if they used external data outside of the C-UDA license? If Kaggle administrators cannot determine whether the data is currently available or not, does that mean the competition winner will not be confirmed until the license check is complete? In my opinion, such matters should be decided before the competition ends.
", "rawMarkdown": "@ashleychow How will the competition winner be determined if they used external data outside of the C-UDA license? If Kaggle administrators cannot determine whether the data is currently available or not, does that mean the competition winner will not be confirmed until the license check is complete? In my opinion, such matters should be decided before the competition ends." } ] } ] }, { "id": 2235101, "postDate": "2023-04-25T18:18:57.847Z", "content": "Can we use youtube videos as external data? If no, why, which ones can be used and which cannot?
", "rawMarkdown": "Can we use youtube videos as external data? If no, why, which ones can be used and which cannot?", "replies": [ { "id": 2236497, "postDate": "2023-04-26T22:57:03.510Z", "content": "I think you need to get the agreement from the personal signing in the video (source: the Deepfake competition)
\nI don't think top teams are using external data. We are not, plus @kolyaforrat and @hoyso48 shared a while ago they were not gonna use any. I might be mistaken though.
\nI'd be disappointed if the competition is won scrapping youtube videos :)
", "rawMarkdown": "I think you need to get the agreement from the personal signing in the video (source: the Deepfake competition)\n\nI don't think top teams are using external data. We are not, plus @kolyaforrat and @hoyso48 shared a while ago they were not gonna use any. I might be mistaken though.\n\nI'd be disappointed if the competition is won scrapping youtube videos :)", "votes": 2, "replies": [ { "id": 2236657, "postDate": "2023-04-27T03:12:41.763Z", "content": "We didn't use any external data either yet, and I think that 0.79 it reachable with just competition data, however since it's relatively easy obtainable we have no choice but to consider the opportunity of using it
", "rawMarkdown": "We didn't use any external data either yet, and I think that 0.79 it reachable with just competition data, however since it's relatively easy obtainable we have no choice but to consider the opportunity of using it" } ] } ] }, { "id": 2234182, "postDate": "2023-04-25T01:41:43.687Z", "content": "Hi Staff. If I train the model locally and package it into tflite, and then submit it later, do I need to eventually commit the source code to ensure that no non-compliant external datasets are used?
", "rawMarkdown": "Hi Staff. If I train the model locally and package it into tflite, and then submit it later, do I need to eventually commit the source code to ensure that no non-compliant external datasets are used?" }, { "id": 2211445, "postDate": "2023-04-06T03:49:59.833Z", "content": "In previous competitions, participants must public the external datasets used by them. Otherwise, they are disqualified.
\nSo is it possible to keep our external datasets secret? Or do we need to publish them to the community somewhere?
\n", "rawMarkdown": ">C. External Data. You may use data other than the Competition Data (“External Data”) to develop and test your Submissions. However, you will ensure the External Data is publicly available and equally accessible to use by all participants of the Competition for purposes of the competition at no cost to the other participants. ", "votes": 1, "replies": [ { "id": 2211958, "postDate": "2023-04-06T11:49:42.957Z", "content": "C. External Data. You may use data other than the Competition Data (“External Data”) to develop and test your Submissions. However, you will ensure the External Data is publicly available and equally accessible to use by all participants of the Competition for purposes of the competition at no cost to the other participants.
\n
I treat this as that you are not required to create a discussion with a dataset link you are using. It is rather a restriction of using datasets that are only accessible to PhD students or smth like that. It would be perfect if someone from kaggle stuff clarify this.
\nAnother interesting question - using your own dataset. If I collect my own ASL dataset - is it treated as an external dataset? Do you need to publish it somewhere in order to satisfy ensure the External Data is publicly available?
Thanks @meowmeowmeowmeowmeow! Yes, data other than the Competition Data is considered as “External Data”, and thus it would need to be \"publicly available and equally accessible to use by all participants of the Competition\". Thanks!
", "rawMarkdown": "Thanks @meowmeowmeowmeowmeow! Yes, data other than the Competition Data is considered as “External Data”, and thus it would need to be \"publicly available and equally accessible to use by all participants of the Competition\". Thanks!" }, { "id": 2212021, "postDate": "2023-04-06T13:00:38.370Z", "content": "I am just curious, would it be considered a rule violation if I create (or buy someone's) dataset and publish it somewhere on the internet on some noname forum? Technically, it would be publicly accessible, but there is no chance to find it by other competitors. It is definitely foul play. How does the competition's rules protect us from this?
", "rawMarkdown": "I am just curious, would it be considered a rule violation if I create (or buy someone's) dataset and publish it somewhere on the internet on some noname forum? Technically, it would be publicly accessible, but there is no chance to find it by other competitors. It is definitely foul play. How does the competition's rules protect us from this?" }, { "id": 2212095, "postDate": "2023-04-06T14:20:55.233Z", "content": "Also you can post this dataset only in the last day of competition.
\nBut I believe Kaggle is a community where people respect each other and fair play
", "rawMarkdown": "Also you can post this dataset only in the last day of competition.\n\nBut I believe Kaggle is a community where people respect each other and fair play", "votes": 2 }, { "id": 2212098, "postDate": "2023-04-06T14:25:54.140Z", "content": "Definitely, we have a great community. But once the community grows there is no way to avoid any dishonest person as well as avoid outliers in the large-scale dataset…
", "rawMarkdown": "Definitely, we have a great community. But once the community grows there is no way to avoid any dishonest person as well as avoid outliers in the large-scale dataset... ", "votes": 1 } ] }, { "id": 2211988, "postDate": "2023-04-06T12:22:40.637Z", "content": "Thanks @yuanzhezhou! Yes, you'd need to make sure the \"External Data is publicly available and equally accessible to use by all participants of the Competition for purposes of the competition at no cost to the other participants\". Thanks!
", "rawMarkdown": "Thanks @yuanzhezhou! Yes, you'd need to make sure the \"External Data is publicly available and equally accessible to use by all participants of the Competition for purposes of the competition at no cost to the other participants\". Thanks!" } ] }, { "id": 2247004, "postDate": "2023-05-05T15:56:11.957Z", "rawMarkdown": "", "isDeleted": true }, { "id": 2239303, "postDate": "2023-04-29T10:45:28.230Z", "rawMarkdown": "", "isDeleted": true }, { "id": 2238367, "postDate": "2023-04-28T13:11:47.717Z", "rawMarkdown": "", "isDeleted": true }, { "id": 2235042, "postDate": "2023-04-25T16:55:11.803Z", "rawMarkdown": "", "isDeleted": true }, { "id": 2231608, "postDate": "2023-04-23T13:20:10.463Z", "content": "Thank you so much ❤️
", "rawMarkdown": "Thank you so much ❤️" } ], "comments": [ { "id": 2236479, "author_name": "Joseph Hillhouse", "author_url": "", "post_date": "2023-04-26T22:34:53.843000", "content": "Is synthetic data allowed and if so, what are the rules with regard to public access when creating the synthetic data requires code and involves solving part of the challenge (e.g. there could be models use to create the synthetic data)?
", "votes": 0, "replies": [] }, { "id": 2235793, "author_name": "slime", "author_url": "", "post_date": "2023-04-26T10:54:52.343000", "content": "Is data under Computational Use of Data Agreement (C-UDA) license falls into the category of allowed for commercial use?
", "votes": 0, "replies": [ { "id": 2236477, "author_name": "Joseph Hillhouse", "author_url": "", "post_date": "2023-04-26T22:33:09.063000", "content": "When it comes to the result, yes. The C-UDA license doesn't allow the data itself to be used for commercial use but allows the results (models) trained against the data to be used for commercial. Here's the annotated version https://github.com/microsoft/Computational-Use-of-Data-Agreement/blob/master/C-UDA-1.0_annotated.md.
", "votes": 0, "replies": [ { "id": 2236491, "author_name": "Theo Viel", "author_url": "", "post_date": "2023-04-26T22:52:11", "content": "From https://github.com/dxli94/WLASL :
\n\n\nAll the WLASL data is intended for academic and computational use only. No commercial usage is allowed.
\n
WLASL which is under C-UDA does not allow for commercial use and they're pretty straight-forward about it.
", "votes": 1, "replies": [] }, { "id": 2236510, "author_name": "Joseph Hillhouse", "author_url": "", "post_date": "2023-04-26T23:40:29.750000", "content": "C-UDA doesn’t allow the data for commercial use but does allow the results for commercial use.
\n“””
\n5.5. “Result” means anything that you develop or improve from your use of Data that does not include more than a de minimis portion of the Data on which the use is based. Results may include de minimis portions of the Data necessary to report on or explain use that has been conducted with the Data, such as figures in scientific papers, but do not include more. Artificial intelligence models trained on Data (and which do not include more than a de minimis portion of Data) are Results.
The results are allowed for commercial usage. The license protects distribution of the data not the results.
\nIt is very straightforward that it is allowed.
", "votes": -1, "replies": [] }, { "id": 2236511, "author_name": "Joseph Hillhouse", "author_url": "", "post_date": "2023-04-26T23:41:58.173000", "content": "“
\n2.2 The C-UDA does not impose any restriction with respect to the use, modification, or distribution of Results.
\n“
Unless I'm misunderstanding, WLASL is just a collection of publically available videos and their associated sign/labels. Does it really matter what licence they use? What you really need is approval from the original data source.
\nJust look at the WLASL script: https://github.com/dxli94/WLASL/blob/master/start_kit/video_downloader.py
\nEach video is listed in the json here: https://raw.githubusercontent.com/dxli94/WLASL/master/start_kit/WLASL_v0.3.json
There are many sources: aslpro.com, youtube, handspeak.com
\nYou would need to prove all these sources provide their videos free for commercial use.
\nJust taking www.handspeak.com as an example, their website explicitly says \"NO AI\" here: https://www.handspeak.com/info/index.php?info=terms
\n
I'm not using it.
", "votes": 0, "replies": [] }, { "id": 2236518, "author_name": "Joseph Hillhouse", "author_url": "", "post_date": "2023-04-27T00:04:02.273000", "content": "There is a different between commercial data use and public data access.
\nThe requirement of the challenge for data is “publicly available and equally accessible” data “at no cost”.
\nThe license of WSASL data is governed under C-UDA which allows “results”, models created from this data to be utilized for commercial purposes but does not allow the data itself to be used for commercial purposes (eg I can’t sell someone else YouTube video for sign language courses).
\nCurators of WSASL cannot license the “data” commercially because they don’t own it but its not necessary because we aren’t distributing the data here but using it for computation use which is its intent.
", "votes": 0, "replies": [] }, { "id": 2236529, "author_name": "Rob Mulla", "author_url": "", "post_date": "2023-04-27T00:21:17.453000", "content": "I love this logic. The ultimate loophole:
\n….now it's allowed?
", "votes": 0, "replies": [] }, { "id": 2236538, "author_name": "arutema47", "author_url": "", "post_date": "2023-04-27T00:31:35.333000", "content": "I tried adding WLASL to my training set for fun, but did not increase CV/LB.
\nWhat about you guys?
Not sure what source you are referring to that “explicitly forbids downloading their data to train ML models”, but it’s not the WSASL source. All content they have scraped is licensed under C-UDA. They also have a means for content creators to request removing the data if a content creator doesn’t want it included. But not sure why folks in deaf community creating this content would want to not help the deaf community. At any rate, the terms of the challenge allows any dataset to be used that is “publicly accessible and equally available” at “no cost”.
", "votes": 0, "replies": [] }, { "id": 2236540, "author_name": "Joseph Hillhouse", "author_url": "", "post_date": "2023-04-27T00:34:08.310000", "content": "I haven’t added it yet. It has very low samples per sign for the WLASL-2k set but it’s on the roadmap.
", "votes": 0, "replies": [] }, { "id": 2236543, "author_name": "Ashley Chow", "author_url": "", "post_date": "2023-04-27T00:38:28.763000", "content": "Unfortunately we could not provide any legal guidance or interpretation here. However, think of it as baking a cake (winning solutions) using different ingredients (data, model training, etc.), if those ingredients used could not allow you to license the winning solutions to \"WINNER LICENSE TYPE: Apache 2.0\" as stated in Rules page, the ingredients should/ could not be used in this case. Hope it helps. Thanks for your participation!
", "votes": 1, "replies": [] }, { "id": 2236545, "author_name": "Rob Mulla", "author_url": "", "post_date": "2023-04-27T00:45:38.810000", "content": "It increased my CV when I used it locally.
", "votes": 1, "replies": [] }, { "id": 2236664, "author_name": "slime", "author_url": "", "post_date": "2023-04-27T03:20:20.543000", "content": "As Joseph mentioned, the legal interpretation of \"commercial use\" can be sometimes tricky, this is why I want to have a definitive statement from hosts if something can be used for the competition or not
\nIt wouldn't be the case if external data was prohibited, however it's allowed and most of the datasets are released under non-commercial licenses
\nSince I think that the status of the dataset is defined by it's license only, I would like to have an answer to the original question
", "votes": 0, "replies": [] }, { "id": 2237245, "author_name": "Rob Mulla", "author_url": "", "post_date": "2023-04-27T13:32:40.033000", "content": "Thanks @ashleychow - I really do love cake analogies, and am now feeling a bit hungry 😊.
\nIn this case I think we've identified a specific ingredient that people are unsure if we are able to use. I understand you can't give legal guidance, but do you know of anyone who could clarify if this specific ingredient meets the requirements? Totally understand if you can't answer that but figured I'd ask so we can hopefully avoid a similar situation as what happened in the deepfake competition.
\nI apprecaite the work you all do to host competitions like this.
", "votes": 1, "replies": [] }, { "id": 2239835, "author_name": "Bilzard", "author_url": "", "post_date": "2023-04-29T23:46:46.337000", "content": "@ashleychow How will the competition winner be determined if they used external data outside of the C-UDA license? If Kaggle administrators cannot determine whether the data is currently available or not, does that mean the competition winner will not be confirmed until the license check is complete? In my opinion, such matters should be decided before the competition ends.
", "votes": 0, "replies": [] } ] } ] }, { "id": 2235101, "author_name": "slime", "author_url": "", "post_date": "2023-04-25T18:18:57.847000", "content": "Can we use youtube videos as external data? If no, why, which ones can be used and which cannot?
", "votes": 0, "replies": [ { "id": 2236497, "author_name": "Theo Viel", "author_url": "", "post_date": "2023-04-26T22:57:03.510000", "content": "I think you need to get the agreement from the personal signing in the video (source: the Deepfake competition)
\nI don't think top teams are using external data. We are not, plus @kolyaforrat and @hoyso48 shared a while ago they were not gonna use any. I might be mistaken though.
\nI'd be disappointed if the competition is won scrapping youtube videos :)
", "votes": 2, "replies": [ { "id": 2236657, "author_name": "slime", "author_url": "", "post_date": "2023-04-27T03:12:41.763000", "content": "We didn't use any external data either yet, and I think that 0.79 it reachable with just competition data, however since it's relatively easy obtainable we have no choice but to consider the opportunity of using it
", "votes": 0, "replies": [] } ] } ] }, { "id": 2234182, "author_name": "Jackson You", "author_url": "", "post_date": "2023-04-25T01:41:43.687000", "content": "Hi Staff. If I train the model locally and package it into tflite, and then submit it later, do I need to eventually commit the source code to ensure that no non-compliant external datasets are used?
", "votes": 0, "replies": [] }, { "id": 2211445, "author_name": "yuanzhe zhou", "author_url": "", "post_date": "2023-04-06T03:49:59.833000", "content": "In previous competitions, participants must public the external datasets used by them. Otherwise, they are disqualified.
\nSo is it possible to keep our external datasets secret? Or do we need to publish them to the community somewhere?
\n", "votes": 1, "replies": [ { "id": 2211958, "author_name": "Mykola", "author_url": "", "post_date": "2023-04-06T11:49:42.957000", "content": "C. External Data. You may use data other than the Competition Data (“External Data”) to develop and test your Submissions. However, you will ensure the External Data is publicly available and equally accessible to use by all participants of the Competition for purposes of the competition at no cost to the other participants.
\n
I treat this as that you are not required to create a discussion with a dataset link you are using. It is rather a restriction of using datasets that are only accessible to PhD students or smth like that. It would be perfect if someone from kaggle stuff clarify this.
\nAnother interesting question - using your own dataset. If I collect my own ASL dataset - is it treated as an external dataset? Do you need to publish it somewhere in order to satisfy ensure the External Data is publicly available?
Thanks @meowmeowmeowmeowmeow! Yes, data other than the Competition Data is considered as “External Data”, and thus it would need to be \"publicly available and equally accessible to use by all participants of the Competition\". Thanks!
", "votes": 0, "replies": [] }, { "id": 2212021, "author_name": "Mykola", "author_url": "", "post_date": "2023-04-06T13:00:38.370000", "content": "I am just curious, would it be considered a rule violation if I create (or buy someone's) dataset and publish it somewhere on the internet on some noname forum? Technically, it would be publicly accessible, but there is no chance to find it by other competitors. It is definitely foul play. How does the competition's rules protect us from this?
", "votes": 0, "replies": [] }, { "id": 2212095, "author_name": "Kolya Forrat", "author_url": "", "post_date": "2023-04-06T14:20:55.233000", "content": "Also you can post this dataset only in the last day of competition.
\nBut I believe Kaggle is a community where people respect each other and fair play
", "votes": 2, "replies": [] }, { "id": 2212098, "author_name": "Mykola", "author_url": "", "post_date": "2023-04-06T14:25:54.140000", "content": "Definitely, we have a great community. But once the community grows there is no way to avoid any dishonest person as well as avoid outliers in the large-scale dataset…
", "votes": 1, "replies": [] } ] }, { "id": 2211988, "author_name": "Ashley Chow", "author_url": "", "post_date": "2023-04-06T12:22:40.637000", "content": "Thanks @yuanzhezhou! Yes, you'd need to make sure the \"External Data is publicly available and equally accessible to use by all participants of the Competition for purposes of the competition at no cost to the other participants\". Thanks!
", "votes": 0, "replies": [] } ] }, { "id": 2247004, "author_name": "", "author_url": "", "post_date": "2023-05-05T15:56:11.957000", "content": "", "votes": 0, "replies": [] }, { "id": 2239303, "author_name": "", "author_url": "", "post_date": "2023-04-29T10:45:28.230000", "content": "", "votes": 0, "replies": [] }, { "id": 2238367, "author_name": "", "author_url": "", "post_date": "2023-04-28T13:11:47.717000", "content": "", "votes": 0, "replies": [] }, { "id": 2235042, "author_name": "", "author_url": "", "post_date": "2023-04-25T16:55:11.803000", "content": "", "votes": 0, "replies": [] }, { "id": 2231608, "author_name": "Miles Morales", "author_url": "", "post_date": "2023-04-23T13:20:10.463000", "content": "Thank you so much ❤️
", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "2210907": "I'd like to address some of the questions related to \"External Data\". External Data needs to allow you to license the winning solution, i.e. being used in commercial purpose. \n\nHere below please see the relevant rules for details:\n>1. WINNER LICENSE.\nUnder Section 11 (Winners Obligations) of the General Rules below, you hereby grant and will grant Competition Sponsor the following license(s) with respect to your Submission if you are a Competition winner:\nOpen Source: You hereby license and will license your winning Submission and the source code used to generate the Submission under an Apache 2.0 License that in no event limits commercial use of such code or model containing or depending on such code. To the extent your Submission makes use of generally commercially available software not owned by you that you used to generate your submission, but that can be procured by the Competition Sponsor without undue expense, you do not grant the license in the preceding sentence to that software.\n...\n>C. External Data. You may use data other than the Competition Data (“External Data”) to develop and test your Submissions. However, you will ensure the External Data is publicly available and equally accessible to use by all participants of the Competition for purposes of the competition at no cost to the other participants. The ability to use External Data under this Section 7.C (External Data) does not limit your other obligations under these Competition Rules, including but not limited to Section 11 (Winners Obligations). \n...\n>11. WINNERS OBLIGATIONS.\nAs a condition to being awarded a Prize, a Prize winner must fulfill the following obligations:\n(a) deliver to the Competition Sponsor the final model's software code as used to generate the winning Submission and associated documentation. The delivered software code should follow these documentation guidelines, must be capable of generating the winning Submission, and contain a description of resources required to build and/or run the executable code successfully. To the extent that the final model’s software code includes generally commercially available software that is not owned by you, but that can be procured by the Competition Sponsor without undue expense, then instead of delivering the code for that software to the Competition Sponsor, you must identify that software, method for procuring it, and any parameters or other information necessary to replicate the winning Submission;\n(b) grant to the Competition Sponsor the license to the winning Submission stated in the Competition Specific Rules above, and represent that you have the unrestricted right to grant that license\n\nWith that, you may use any external data as long as it is public and commercially available. Thank you!", "2236479": "Is synthetic data allowed and if so, what are the rules with regard to public access when creating the synthetic data requires code and involves solving part of the challenge (e.g. there could be models use to create the synthetic data)?", "2235793": "Is data under Computational Use of Data Agreement ([C-UDA](https://spdx.org/licenses/C-UDA-1.0.html)) license falls into the category of allowed for commercial use?", "2235101": "Can we use youtube videos as external data? If no, why, which ones can be used and which cannot?", "2234182": "Hi Staff. If I train the model locally and package it into tflite, and then submit it later, do I need to eventually commit the source code to ensure that no non-compliant external datasets are used?", "2211445": "In previous competitions, participants must public the external datasets used by them. Otherwise, they are disqualified. \nSo is it possible to keep our external datasets secret? Or do we need to publish them to the community somewhere?", "2247004": "", "2239303": "", "2238367": "", "2235042": "", "2231608": "Thank you so much ❤️" } }