Per the Competition Rules, you may use any external data you wish, as long as you post it to this forum thread with its provenance clearly stated. But only submissions using external data that is publicly and freely available for use that includes research or academic purposes will be eligible for prizes. External data which is publicly available should include a link to access it, when posting here. All external data must be posted to this forum thread no later than the Entry Deadline (one week before competition close).
\n\nOnce someone posts an external dataset to this thread, you do not need to re-post it if you are using the same one.
\n\nYou only need to declare the original dataset used; you do not need to declare re-labeled or augmented or otherwise processed versions of datasets. Pre-trained models can be declared, as well; however, models resulting from your own original work, that you have trained yourself offline do not need to be shared/declared.
\n\nLastly, ensure you do not hand label the test set.
", "messages": [ { "id": 886037, "postDate": "2020-06-14T17:00:36.730Z", "content": "@juliaelliott \nGiven the situation what happened in deepfake challenge, could we get a clarification about this competition? From my personal experience it is practically impossible to get the data with the consent of the patients unless you were to have consent forms with a paragraph that your (as a patient) data can be used for research.
\n\nDoes this mean that all external data here would be considered against the rules as well (tbh have not read it thoroughly yet, but assume similar to deepfake one)
\n\nIf external data was allowed, it is going to look very bad if double standards would be applied to different competitions involving personal data...
", "rawMarkdown": "@juliaelliott \nGiven the situation what happened in deepfake challenge, could we get a clarification about this competition? From my personal experience it is practically impossible to get the data with the consent of the patients unless you were to have consent forms with a paragraph that your (as a patient) data can be used for research. \n\nDoes this mean that all external data here would be considered against the rules as well (tbh have not read it thoroughly yet, but assume similar to deepfake one)\n\nIf external data was allowed, it is going to look very bad if double standards would be applied to different competitions involving personal data...", "votes": 27, "replies": [ { "id": 887024, "postDate": "2020-06-15T12:40:15.150Z", "content": "What I can say is to use external data at your own risk, even if its allowed in competition rules. My own experience.
", "rawMarkdown": "What I can say is to use external data at your own risk, even if its allowed in competition rules. My own experience. ", "votes": 16 }, { "id": 887949, "postDate": "2020-06-16T02:56:35.190Z", "content": "@raddar A quick reply to let you know your question is not being ignored; the host is evaluating their response and will post a clarification as soon as possible.
", "rawMarkdown": "@raddar A quick reply to let you know your question is not being ignored; the host is evaluating their response and will post a clarification as soon as possible.", "votes": 5 }, { "id": 920461, "postDate": "2020-07-08T15:59:06.990Z", "content": "We apologize for the delay on this - we are still discussing internally how to proceed. Thank you!
", "rawMarkdown": "We apologize for the delay on this - we are still discussing internally how to proceed. Thank you! ", "votes": 3 }, { "id": 963580, "postDate": "2020-08-09T06:34:24.573Z", "content": "Is there any decision about that yet? @veronicarotemberg @juliaelliott
", "rawMarkdown": "Is there any decision about that yet? @veronicarotemberg @juliaelliott " } ] }, { "id": 864316, "postDate": "2020-05-28T00:15:53.433Z", "content": "Per the Competition Rules, you may use any external data you wish, as long as you post it to this forum thread with its provenance clearly stated. But only submissions using external data that is publicly and freely available for use that includes research or academic purposes will be eligible for prizes. External data which is publicly available should include a link to access it, when posting here. All external data must be posted to this forum thread no later than the Entry Deadline (one week before competition close).
\n\nOnce someone posts an external dataset to this thread, you do not need to re-post it if you are using the same one.
\n\nYou only need to declare the original dataset used; you do not need to declare re-labeled or augmented or otherwise processed versions of datasets. Pre-trained models can be declared, as well; however, models resulting from your own original work, that you have trained yourself offline do not need to be shared/declared.
\n\nLastly, ensure you do not hand label the test set.
", "rawMarkdown": "Per the [Competition Rules](https://www.kaggle.com/c/siim-isic-melanoma-classification/rules), you may use any external data you wish, as long as you post it to this forum thread with its provenance clearly stated. **But** only submissions using external data that is publicly and freely available for use that includes research or academic purposes will be eligible for prizes. External data which is publicly available should include a link to access it, when posting here. All external data must be posted to this forum thread no later than the Entry Deadline (one week before competition close).\n\nOnce someone posts an external dataset to this thread, you do not need to re-post it if you are using the same one.\n\nYou only need to declare the original dataset used; you do not need to declare re-labeled or augmented or otherwise processed versions of datasets. Pre-trained models can be declared, as well; however, models resulting from your own original work, that you have trained yourself offline do not need to be shared/declared.\n\nLastly, ensure you do not hand label the test set.", "votes": 27 }, { "id": 877677, "postDate": "2020-06-07T20:56:04.337Z", "content": "I added raw datasets from last years ISIC:\n- ISIC2016: https://www.kaggle.com/shonenkov/isic2016\n- ISIC2017: https://www.kaggle.com/shonenkov/isic2017\n- ISIC2018: https://www.kaggle.com/shonenkov/isic2018
", "rawMarkdown": "I added raw datasets from last years ISIC:\n- [ISIC2016](https://challenge.kitware.com/#phase/5667455bcad3a56fac786791): https://www.kaggle.com/shonenkov/isic2016\n- [ISIC2017](https://challenge.kitware.com/#phase/5840f53ccad3a51cc66c8dab): https://www.kaggle.com/shonenkov/isic2017\n- [ISIC2018](https://challenge.kitware.com/#phase/5abcbc6f56357d0139260e66): https://www.kaggle.com/shonenkov/isic2018", "votes": 14, "replies": [ { "id": 886409, "postDate": "2020-06-15T03:05:01.183Z", "content": "Thanks, this is a great place to start using external data!
\n\nThanks for uploading the datasets to Kaggle, really helps.
\n\nUnfortunately, I am unable to access the website, but any chance you have a copy of 2018 Lesion Semgentation dataset please?
@shonenkov did you find duplicates between those 3 datasets using DBSCAN approach?
", "rawMarkdown": "@shonenkov did you find duplicates between those 3 datasets using DBSCAN approach?", "votes": 2 }, { "id": 887407, "postDate": "2020-06-15T16:55:25.203Z", "content": "@optimo I didn't try. isic16 + isic17 + isic18 gives isic19, maybe it helps you
", "rawMarkdown": "@optimo I didn't try. isic16 + isic17 + isic18 gives isic19, maybe it helps you", "votes": 3 }, { "id": 887627, "postDate": "2020-06-15T19:20:41.717Z", "content": "Well I read that isic19 contained previous competition data too but @hengck23 in this comment https://www.kaggle.com/c/siim-isic-melanoma-classification/discussion/155201#880993 seems to be using isic2019 + 16 + archive, so I was wondering whether 2019 = 2017 + 2018 only or 2019 = 2018 + 2017 + 2016. Maybe it's easier to ask him directly.
\n\n@hengck23 would you mind explaining what isic2019 + 16 + archive stands for in your experiments? 😃
", "rawMarkdown": "Well I read that isic19 contained previous competition data too but @hengck23 in this comment https://www.kaggle.com/c/siim-isic-melanoma-classification/discussion/155201#880993 seems to be using isic2019 + 16 + archive, so I was wondering whether 2019 = 2017 + 2018 only or 2019 = 2018 + 2017 + 2016. Maybe it's easier to ask him directly.\n\n@hengck23 would you mind explaining what isic2019 + 16 + archive stands for in your experiments? 😃 " } ] }, { "id": 886521, "postDate": "2020-06-15T05:22:59.040Z", "content": "I'm confused about this statement in the rules:\n> Participants who do not use Public External Data are not eligible to receive a Prize, but may retain their non-monetary winning features in accordance with Section 10 of the General Rules
\n\nSo does that mean to be eligible for a prize, the use of external data is mandatory?
\n\nAs raddar mentioned, given what happened in the Deepfake competition, this could create issues for us
", "rawMarkdown": "I'm confused about this statement in the rules:\n> Participants who do not use Public External Data are not eligible to receive a Prize, but may retain their non-monetary winning features in accordance with Section 10 of the General Rules\n\nSo does that mean to be eligible for a prize, the use of external data is mandatory? \n\nAs raddar mentioned, given what happened in the Deepfake competition, this could create issues for us", "votes": 11, "replies": [ { "id": 886740, "postDate": "2020-06-15T08:45:30.390Z", "content": "@anjum48 Beautiful rule! 😅🤣
", "rawMarkdown": "@anjum48 Beautiful rule! 😅🤣", "votes": 2 }, { "id": 886775, "postDate": "2020-06-15T09:13:08.870Z", "content": "@juliaelliott the same paragraph includes.\nHowever, you will post any External Data used and the provenance of the data to the official competition forum prior to the Entry Deadline, and if the data is publicly available, include the link to such access.
So the external data DOES NOT need to be public.
", "rawMarkdown": "@juliaelliott the same paragraph includes.\n`However, you will post any External Data used and the provenance of the data to the official competition forum prior to the Entry Deadline, and if the data is publicly available, include the link to such access. `\n\nSo **the external data DOES NOT need to be public**. ", "votes": 2 }, { "id": 886787, "postDate": "2020-06-15T09:19:16.200Z", "content": "I think I understand now. You can use private data, but you will be ineligible for the prize.
\n\nI guess this will require very clear guidance around what is public and what is not.
", "rawMarkdown": "I think I understand now. You can use private data, but you will be ineligible for the prize.\n\nI guess this will require very clear guidance around what is public and what is not." }, { "id": 887498, "postDate": "2020-06-15T17:42:09.320Z", "content": "It is not required to use external data. If you do use external data and wish to be eligible for prizes, the dataset must be publicly and freely available for use that includes academic/research use with provenance of the data clearly posted to this thread. If you do use external data and it does not meet these requirements to be considered \"Public External Data\" then it will not be prize-eligible.
", "rawMarkdown": "It is **not required** to use external data. **If** you do use external data and wish to be eligible for prizes, the dataset must be publicly and freely available for use that includes academic/research use with provenance of the data clearly posted to this thread. If you do use external data and it does not meet these requirements to be considered \"Public External Data\" then it will not be prize-eligible.", "votes": 3 } ] }, { "id": 958676, "postDate": "2020-08-05T05:34:31.420Z", "content": "128x128 JPEGs with CSV target, meta, sample submission (240MB)
\n\n1024x1024 TFRecords with targets, meta, and sample submission (8.9GB)
128x128 JPEGs with CSV target and meta (150MB)
\n\nHello,
\n\nHere's some external(ish) datasets:\n- ISIC 2019;\n- ISIC 2018;\n- ISIC 2017
\n\n(I'm aware that the 2019 versions contains both datasets from 2018 and 2017 as well)
\n\nHope it helps\n😄
", "rawMarkdown": "Hello,\n\nHere's some external(ish) datasets:\n- [ISIC 2019](https://www.kaggle.com/andrewmvd/isic-2019);\n- [ISIC 2018](https://www.kaggle.com/kmader/skin-cancer-mnist-ham10000);\n- [ISIC 2017](https://www.kaggle.com/wanderdust/skin-lesion-analysis-toward-melanoma-detection)\n\n(I'm aware that the 2019 versions contains both datasets from 2018 and 2017 as well)\n\nHope it helps\n😄 ", "votes": 8, "replies": [ { "id": 864906, "postDate": "2020-05-28T09:02:10.913Z", "content": "I think all of these are used in competition train set, right?
", "rawMarkdown": "I think all of these are used in competition train set, right?" }, { "id": 864989, "postDate": "2020-05-28T10:17:08.773Z", "content": "External(ish)....Figured the best place to discover would be this very thread. With only 584 positive cases, this will be crucial.
\n\nSo far, official information doesn't clarify this:
\n\n\n\n\nThe dataset was generated by the International Skin Imaging Collaboration (ISIC) and images are from the following sources: Hospital Clínic de Barcelona, Medical University of Vienna, Memorial Sloan Kettering Cancer Center, Melanoma Institute Australia, The University of Queensland, and the University of Athens Medical School.
\n
It could be the aggregate training data from 2016 to 2019, It could also be train + test of 2019 (which sums to 33k images) and on top of that there's also their online archive with some 24k images to it.
\n\nIf its the first option, that might mean that the testing set is an aggregate from those years as well...and thus teams who participated in the past have probed what will be our private testing set.
\n\nI haven't participated in the past, but I would like to know anyway.
\n\ntl;dr;\nI don't know, I'm asking by suggesting.
", "rawMarkdown": "External(ish)....Figured the best place to discover would be this very thread. With only 584 positive cases, this will be crucial.\n\nSo far, [official information](https://challenge2020.isic-archive.com/) doesn't clarify this:\n> The dataset was generated by the International Skin Imaging Collaboration (ISIC) and images are from the following sources: Hospital Clínic de Barcelona, Medical University of Vienna, Memorial Sloan Kettering Cancer Center, Melanoma Institute Australia, The University of Queensland, and the University of Athens Medical School.\n\nIt could be the aggregate training data from 2016 to 2019, It could also be train + test of 2019 (which sums to 33k images) and on top of that there's also their [online archive](https://www.isic-archive.com/#!/topWithHeader/onlyHeaderTop/gallery) with some 24k images to it.\n\nIf its the first option, that might mean that the testing set is an aggregate from those years as well...and thus teams who participated in the past have probed what will be our private testing set.\n\nI haven't participated in the past, but I would like to know anyway.\n\ntl;dr;\nI don't know, I'm asking by suggesting.\n", "votes": 2 }, { "id": 865350, "postDate": "2020-05-28T15:23:53.890Z", "content": "Hi all - great questions! We've done the best we can to make the train and test datasets here unique. While there may be a few overlapping lesions between the training set here and prior ISIC training datasets and images available at https://isic-archive.com/ or via the ISIC API due to the retrospective nature of our clinical database queries, we expect that they will have been taken with different photographic equipment and be very few overall.
\n\nWe did this because prior datasets don't contain the patient-contextual information (aka other images from the same patient) which we hypothesize might improve performance.
", "rawMarkdown": "Hi all - great questions! We've done the best we can to make the train and test datasets here unique. While there may be a few overlapping lesions between the training set here and prior ISIC training datasets and images available at https://isic-archive.com/ or via the ISIC API due to the retrospective nature of our clinical database queries, we expect that they will have been taken with different photographic equipment and be very few overall. \n\nWe did this because prior datasets don't contain the patient-contextual information (aka other images from the same patient) which we hypothesize might improve performance. ", "votes": 6 }, { "id": 866234, "postDate": "2020-05-29T08:09:07.233Z", "content": "thanks for setting up this challenge @veronicarotemberg. Here are my two cents.
\n\nIt would be super super interesting if other metadata such as location is available. Maybe something to think about in the future.
\n\nYou could use a large radius (i.e. 5km) to avoid privacy issues. Already this could be sufficient to link it to many other sources of information for which evidence exists to have an effect on skin cancer. To name a few with location you get access to a number of sunny days and UV radiation, altitude, air pollution etc.
", "rawMarkdown": "thanks for setting up this challenge @veronicarotemberg. Here are my two cents.\n \nIt would be super super interesting if other metadata such as location is available. Maybe something to think about in the future.\n\nYou could use a large radius (i.e. 5km) to avoid privacy issues. Already this could be sufficient to link it to many other sources of information for which evidence exists to have an effect on skin cancer. To name a few with location you get access to a number of sunny days and UV radiation, altitude, air pollution etc.", "votes": 13 } ] }, { "id": 866123, "postDate": "2020-05-29T06:06:19.723Z", "content": "Other skin lesion datasets that may help (excluding the ISIC archive):\nDermofit\nAtlas of Dermotoscopy\nPH2\nDerm7pt
", "rawMarkdown": "Other skin lesion datasets that may help (excluding the ISIC archive):\n[Dermofit](https://licensing.edinburgh-innovations.ed.ac.uk/i/software/dermofit-image-library.html)\n[Atlas of Dermotoscopy](http://www.dermoscopy.org/atlas/cd_review.asp)\n[PH2](https://www.fc.up.pt/addi/ph2%20database.html)\n[Derm7pt](http://derm.cs.sfu.ca/Welcome.html)", "votes": 6, "replies": [ { "id": 910204, "postDate": "2020-07-01T03:51:55.270Z", "content": "@ipateam Dermofit library seems not to be public, can we use this then?
", "rawMarkdown": "@ipateam Dermofit library seems not to be public, can we use this then?", "votes": 1 }, { "id": 910613, "postDate": "2020-07-01T08:51:22.267Z", "content": "Yes, you are right. I was not aware of it at the beginning.
", "rawMarkdown": "Yes, you are right. I was not aware of it at the beginning. ", "votes": 1 }, { "id": 918773, "postDate": "2020-07-07T13:39:14.120Z", "content": "In PH2 website:
\n The data included in the PH² database can be used for research and educational purposes. It is important to note that redistribution and commercial use is not allowed.
Using PH2 in kaggle belongs to the condition of 'commercial use' ?
", "rawMarkdown": "In PH2 website: \n The data included in the PH² database can be used for research and educational purposes. It is important to note that redistribution and commercial use is not allowed.\n\n Using PH2 in kaggle belongs to the condition of 'commercial use' ?" }, { "id": 919017, "postDate": "2020-07-07T16:45:06.857Z", "content": "@bubble94 In this competition, external data that is licensed \"for use that includes research or academic purposes\" still qualifies as permitted for prize eligibility.\n@ipateam & @ankitsajwan Correct, any dataset that comes at any cost is not permitted for use as \"publicly and freely available\" if you wish to be prize eligible.
", "rawMarkdown": "@bubble94 In this competition, external data that is licensed \"for use that includes research or academic purposes\" still qualifies as permitted for prize eligibility.\n@ipateam & @ankitsajwan Correct, any dataset that comes at any cost is not permitted for use as \"publicly and freely available\" if you wish to be prize eligible.", "votes": 1 } ] }, { "id": 942760, "postDate": "2020-07-24T01:36:17.710Z", "content": "Following datasets :\nJPEG image dataset :\n2019 and 2020 dataset merged:\n- https://www.kaggle.com/shonenkov/melanoma-merged-external-data-512x512-jpeg
\n\nTFrecords\n- https://www.kaggle.com/cdeotte/isic2019-768x768 \n- https://www.kaggle.com/cdeotte/melanoma-768x768
\n- https://www.kaggle.com/cdeotte/melanoma-512x512\n- https://www.kaggle.com/cdeotte/melanoma-384x384
@juliaelliott: could you please revisit the competition rules and make some correction. Competition rules made confusion again to the participants. Kaggle Team \"MUST\" learn from their mistakes in DeepFake Competition. You must address it with the highest priority. Please respect competitors' free time and hard work they dedicated to this competition @anjum48 thanks for the information regarding conflicting rules.
", "rawMarkdown": "@juliaelliott: could you please revisit the competition rules and make some correction. Competition rules made confusion again to the participants. Kaggle Team \"MUST\" learn from their mistakes in DeepFake Competition. You must address it with the highest priority. Please respect competitors' free time and hard work they dedicated to this competition @anjum48 thanks for the information regarding conflicting rules.", "votes": 3, "replies": [ { "id": 887947, "postDate": "2020-06-16T02:55:34.543Z", "content": "@projdev The rules in this competition are distinct from other competitions. Please specify where you are confused, and we will work with the hosts to help with clarification.
", "rawMarkdown": "@projdev The rules in this competition are distinct from other competitions. Please specify where you are confused, and we will work with the hosts to help with clarification." } ] }, { "id": 964082, "postDate": "2020-08-09T15:17:31.843Z", "content": "https://challenge2019.isic-archive.com/data.html\nhttps://github.com/lukemelas/EfficientNet-PyTorch\nhttps://github.com/Cadene/pretrained-models.pytorch\nhttps://github.com/facebookresearch/semi-supervised-ImageNet1K-models\nhttps://github.com/rwightman/pytorch-image-models
", "rawMarkdown": "https://challenge2019.isic-archive.com/data.html\nhttps://github.com/lukemelas/EfficientNet-PyTorch\nhttps://github.com/Cadene/pretrained-models.pytorch\nhttps://github.com/facebookresearch/semi-supervised-ImageNet1K-models\nhttps://github.com/rwightman/pytorch-image-models", "votes": 1 }, { "id": 951755, "postDate": "2020-07-30T11:45:08.577Z", "content": "Wow, just fumbled across this very nice repo: https://github.com/CoinCheung/pytorch-loss
", "rawMarkdown": "Wow, just fumbled across this very nice repo: https://github.com/CoinCheung/pytorch-loss", "votes": 1 }, { "id": 950325, "postDate": "2020-07-29T10:38:53.400Z", "content": "@juliaelliott @veronicarotemberg Is GAN based melanoma generation allowed in this competition?
\n\nI have also created a separate discussion for this here.
", "rawMarkdown": "@juliaelliott @veronicarotemberg Is GAN based melanoma generation allowed in this competition?\n\nI have also created a separate discussion for this [here](https://www.kaggle.com/c/siim-isic-melanoma-classification/discussion/170846).\n", "votes": 1, "replies": [ { "id": 950862, "postDate": "2020-07-29T17:17:18.567Z", "content": "See my response here: https://www.kaggle.com/c/siim-isic-melanoma-classification/discussion/170846#950860 -- You may use generated images, as long as they are not generated from the test set, as that effectively constitutes hand-labeling.
", "rawMarkdown": "See my response here: https://www.kaggle.com/c/siim-isic-melanoma-classification/discussion/170846#950860 -- You may use generated images, as long as they are not generated from the test set, as that effectively constitutes hand-labeling." }, { "id": 950943, "postDate": "2020-07-29T18:34:39.977Z", "content": "@juliaelliott Can we use GANs to generate images without target labels? Is that still considered \"hand-labeling\"? (We can then use these images without labels in unsupervised learning)
", "rawMarkdown": "@juliaelliott Can we use GANs to generate images **without** target labels? Is that still considered \"hand-labeling\"? (We can then use these images without labels in unsupervised learning)", "votes": 1 } ] }, { "id": 940387, "postDate": "2020-07-22T22:49:57.687Z", "content": "https://github.com/osmr/imgclsmob/tree/master/pytorch
", "rawMarkdown": "https://github.com/osmr/imgclsmob/tree/master/pytorch", "votes": 1 }, { "id": 934882, "postDate": "2020-07-18T21:07:51.480Z", "content": "GhostNet:\nhttps://www.kaggle.com/ipythonx/ghostnetpretrained
", "rawMarkdown": "GhostNet:\nhttps://www.kaggle.com/ipythonx/ghostnetpretrained", "votes": 1 }, { "id": 913008, "postDate": "2020-07-02T22:07:10.737Z", "content": "@juliaelliott One of the fellow participant ( @cdeotte , Huge Thanks) resized the images and combined the meta features into a TFRecords dataset. Does this count as external data?\nLink:https://www.kaggle.com/cdeotte/melanoma-256x256
", "rawMarkdown": "@juliaelliott One of the fellow participant ( @cdeotte , Huge Thanks) resized the images and combined the meta features into a TFRecords dataset. Does this count as external data?\nLink:https://www.kaggle.com/cdeotte/melanoma-256x256", "votes": 1, "replies": [ { "id": 913043, "postDate": "2020-07-02T23:06:19.503Z", "content": "@spideysloth Yes.
", "rawMarkdown": "@spideysloth Yes.", "votes": 2 }, { "id": 915040, "postDate": "2020-07-04T12:43:12.667Z", "content": "Isn't this just preprocessing? How is it external data?
", "rawMarkdown": "Isn't this just preprocessing? How is it external data?", "votes": 2 }, { "id": 919012, "postDate": "2020-07-07T16:41:21.093Z", "content": "Apologies for not being clear in my prior response, as I don't think I understood the question. What you describe is external data by virtue of being loaded for use in your modeling. However, the dataset itself does not need to be declared separately as external data, as it already constitutes the dataset provided, as @jonathanrayner suggests.
", "rawMarkdown": "Apologies for not being clear in my prior response, as I don't think I understood the question. What you describe is external data by virtue of being loaded for use in your modeling. However, the dataset itself does not need to be declared separately as external data, as it already constitutes the dataset provided, as @jonathanrayner suggests." } ] }, { "id": 888476, "postDate": "2020-06-16T11:40:56.913Z", "content": "Here is a dataset aggregating (some) of the public submissions : https://www.kaggle.com/louise2001/melanoma
", "rawMarkdown": "Here is a dataset aggregating (some) of the public submissions : https://www.kaggle.com/louise2001/melanoma", "votes": 1 }, { "id": 878823, "postDate": "2020-06-08T22:09:54.283Z", "content": "", "rawMarkdown": "https://www.isic-archive.com/", "votes": 1 }, { "id": 867882, "postDate": "2020-05-30T17:47:04.767Z", "content": "https://skincancer.net/images/types/
", "rawMarkdown": "https://skincancer.net/images/types/", "votes": 1, "replies": [ { "id": 868003, "postDate": "2020-05-30T20:09:28.043Z", "content": "Is it OK?
\n\n\n", "rawMarkdown": "Is it OK?\n\n> The skin cancer images gallery is not all-encompassing and should not be used for diagnostic purposes.", "votes": 1 }, { "id": 878994, "postDate": "2020-06-09T05:22:49.860Z", "content": "The skin cancer images gallery is not all-encompassing and should not be used for diagnostic purposes.
\n
I too have the same question...
", "rawMarkdown": "I too have the same question..." }, { "id": 879986, "postDate": "2020-06-09T22:18:11.957Z", "content": "Hi all, good question. Upon closer inspection, we suspect this message is a disclaimer. We haven't identified any licensing restrictions, and so the host has confirmed skincancer.net is permitted for use from a rules perspective. A side note that we weren't able to determine whether these are dermoscopic images, which could impact how useful they are.
", "rawMarkdown": "Hi all, good question. Upon closer inspection, we suspect this message is a disclaimer. We haven't identified any licensing restrictions, and so the host has confirmed skincancer.net is permitted for use from a rules perspective. A side note that we weren't able to determine whether these are dermoscopic images, which could impact how useful they are." } ] }, { "id": 964563, "postDate": "2020-08-10T02:25:36.903Z", "content": "https://github.com/rwightman/pytorch-image-models
\nhttps://github.com/rwightman/gen-efficientnet-pytorch
\nhttps://github.com/lukemelas/EfficientNet-PyTorch
\nhttps://github.com/clovaai/rexnet
\nhttps://github.com/Cadene/pretrained-models.pytorch
@juliaelliott @veronicarotemberg The Rules section A.1. EXTERNAL DATA defines \"Public External Data\" as External Data available to use by all participants of the competition for purposes of the competition, for use that includes research or academic purposes, at no cost to the other participants.\nNOTE: \"Public External Data\" does not mention dataset \"commercial use\" anywhere!\nHowever, the A.4. WINNER LICENSE section requires that the source code (not dataset) should have a license allowing \"commercial use\"
\n\nIn a new forum post by Chris More Kaggle Datasets - 4000 Malignant Images!, I see participants referring to PH2 and 7-point datasets.
\n\nThese datasets are released as \"CC BY-NC-SA 4.0\", which allows research and educational purposes, but not redistribution and commercial use.
\n\nCould you please clarify whether we can use these Ph2 and 7-point datasets because they meet the \"research, academic, no cost\" requirements of the \"Public External Data\", even though the datasets do not allow commercial use?
", "rawMarkdown": "@juliaelliott @veronicarotemberg The [Rules](https://www.kaggle.com/c/siim-isic-melanoma-classification/rules) section A.1. EXTERNAL DATA defines \"Public External Data\" as External Data available to use by all participants of the competition for purposes of the competition, for use that includes **research or academic purposes**, at no cost to the other participants.\nNOTE: \"Public External Data\" does not mention dataset \"commercial use\" anywhere!\nHowever, the A.4. WINNER LICENSE section requires that the **source code** (not dataset) should have a license allowing \"commercial use\"\n\nIn a new forum post by Chris **More Kaggle Datasets - 4000 Malignant Images!**, I see participants referring to [PH2 and 7-point datasets](https://www.kaggle.com/c/siim-isic-melanoma-classification/discussion/169139#940621).\n\nThese datasets are released as \"CC BY-NC-SA 4.0\", which allows research and educational purposes, but not redistribution and commercial use.\n\nCould you please clarify whether we can use these Ph2 and 7-point datasets because they meet the \"research, academic, no cost\" requirements of the \"Public External Data\", even though the datasets do not allow commercial use?\n", "replies": [ { "id": 942379, "postDate": "2020-07-23T18:04:24.060Z", "content": "@sirishks Thanks for raising this question. As you have observed, the host is permitting use of external data that is public as eligible for prizes, as long as it has been made available, minimally, for research/academic use. However, this is also provided that the resulting solution can also be licensed per the WINNERS LICENSE requirements specified in the competition's rules, which does include commercial use. Therefore, the \"CC BY-NC-SA 4.0\" license (which, incidentally, is similar to how the provided competition dataset is also licensed) would fall under eligible, conditional on the final model ultimately being licensed under one of the eligible licenses in the rules. I'll also add that it is not a requirement to use external data at all.
", "rawMarkdown": "@sirishks Thanks for raising this question. As you have observed, the host is permitting use of external data that is public as eligible for prizes, as long as it has been made available, minimally, for research/academic use. However, this is also provided that the resulting solution can also be licensed per the WINNERS LICENSE requirements specified in the competition's rules, which does include commercial use. Therefore, the \"CC BY-NC-SA 4.0\" license (which, incidentally, is similar to how the provided competition dataset is also licensed) would fall under eligible, conditional on the final model ultimately being licensed under one of the eligible licenses in the [rules](https://www.kaggle.com/c/siim-isic-melanoma-classification/rules). I'll also add that it is not a requirement to use external data at all.", "replies": [ { "id": 942428, "postDate": "2020-07-23T18:27:51.960Z", "content": "@juliaelliott Thank you for your reply.
\n\nJust to avoid a DFDC situation, I quickly checked the license details and found that one of the licensing terms means that \"If you remix, transform, or build upon the material, you must distribute your contributions under the same license as the original\" i.e. the model built using a dataset licensed like this can not be used for commercial purposes.
\n\nCould you please confirm this conclusion?
\n\nPS: I am also confused because last year winners DAISYLab wrote in their report section 2.1 that they have used \"dermoscopic images from the 7-point dataset\" (which does not allow commercial use).
", "rawMarkdown": "@juliaelliott Thank you for your reply.\n\nJust to avoid a [DFDC situation](https://www.kaggle.com/c/deepfake-detection-challenge/discussion/157983), I quickly checked the [license details](https://creativecommons.org/licenses/by-nc-sa/4.0/) and found that one of the licensing terms means that \"If you remix, transform, or build upon the material, you must distribute your contributions under the same license as the original\" i.e. the model built using a dataset licensed like this **can not** be used for commercial purposes.\n\nCould you please confirm this conclusion?\n\nPS: I am also confused because [last year winners](https://challenge2019.isic-archive.com/leaderboard.html) [**DAISYLab**](https://isic-challenge-stade.s3.amazonaws.com/99bdfa5c-4b6b-4c3c-94c0-f614e6a05bc4/method_description.pdf?AWSAccessKeyId=AKIA2FPBP3II4S6KTWEU&Signature=bgrKzgxC5ZCKgQ%2B5F%2FUt21zGy94%3D&Expires=1595554666) wrote in their report section 2.1 that they have used \"dermoscopic images from the 7-point dataset\" (which **does not** allow commercial use)." } ] }, { "id": 944995, "postDate": "2020-07-25T13:45:26.520Z", "content": "\n\nDo I understand correctly that *we are eligible to use those data * for the final model training and using it wouldn't disqualify us from the competition?
", "rawMarkdown": "@juliaelliott \n\nDo I understand correctly that **we are eligible to use those data ** for the final model training and using it wouldn't disqualify us from the competition? ", "votes": 1 }, { "id": 948437, "postDate": "2020-07-27T23:59:52.813Z", "content": "@janidziak Yes, datasets that are licensed as available for research/academic use can be used and remain eligible to receive the competition's prizes, as long as you will license your final solution and code as open source, including for commercial use, per the winners' licensing requirements in the rules.
", "rawMarkdown": "@janidziak Yes, datasets that are licensed as available for research/academic use **can be used** and remain eligible to receive the competition's prizes, as long as you will license your final solution and code as open source, including for commercial use, per the winners' licensing requirements in the rules.", "votes": 3 } ] }, { "id": 888596, "postDate": "2020-06-16T13:00:26.150Z", "content": "I have created a dataset which uses original data from this competition with an addition of only malignan cases from 2019 ISIC cometition.\nDataset: https://www.kaggle.com/nroman/melanoma-external-malignant-256\nKernel using it: https://www.kaggle.com/nroman/melanoma-pytorch-starter-efficientnet/comments\nA post: https://www.kaggle.com/c/siim-isic-melanoma-classification/discussion/159101
", "rawMarkdown": "I have created a dataset which uses original data from this competition with an addition of only malignan cases from 2019 ISIC cometition.\nDataset: https://www.kaggle.com/nroman/melanoma-external-malignant-256\nKernel using it: https://www.kaggle.com/nroman/melanoma-pytorch-starter-efficientnet/comments\nA post: https://www.kaggle.com/c/siim-isic-melanoma-classification/discussion/159101", "votes": 2, "replies": [ { "id": 942199, "postDate": "2020-07-23T16:26:05.143Z", "content": "Hey can i use your data ? also any update regarding it. I am a novice
", "rawMarkdown": "Hey can i use your data ? also any update regarding it. I am a novice" } ] }, { "id": 872033, "postDate": "2020-06-02T20:45:07.377Z", "content": "Am using Keras model weights: https://github.com/fchollet/deep-learning-models/releases/download/v0.7/inception_resnet_v2_weights_tf_dim_ordering_tf_kernels_notop.h5
", "rawMarkdown": "Am using Keras model weights: https://github.com/fchollet/deep-learning-models/releases/download/v0.7/inception_resnet_v2_weights_tf_dim_ordering_tf_kernels_notop.h5" }, { "id": 969925, "postDate": "2020-08-14T04:16:13.453Z", "content": "pie torch efficient
\n2009 cancer data
https://www.isic-archive.com/#!/onlyHeaderTop/gallery
\nISIC melanoma pictures (might be redundant though, see post by Chris Deotte)
", "rawMarkdown": "https://www.isic-archive.com/#!/onlyHeaderTop/gallery\n\nISIC melanoma pictures (might be redundant though, see post by Chris Deotte)", "votes": -1 }, { "id": 966878, "postDate": "2020-08-11T18:00:51.513Z", "content": "https://github.com/anindox8/Ensemble-of-Multi-Scale-CNN-for-Dermatoscopy-Classification.git
", "rawMarkdown": "https://github.com/anindox8/Ensemble-of-Multi-Scale-CNN-for-Dermatoscopy-Classification.git", "votes": -1 }, { "id": 965753, "postDate": "2020-08-10T20:20:56.693Z", "content": "https://github.com/pytorch/vision/tree/master/torchvision/models
", "rawMarkdown": "[https://github.com/pytorch/vision/tree/master/torchvision/models](https://github.com/pytorch/vision/tree/master/torchvision/models)", "votes": -1 }, { "id": 868032, "postDate": "2020-05-30T20:27:53.790Z", "content": "oki-toki 😏 \nhttps://www.kaggle.com/kmader/skin-cancer-mnist-ham10000
", "rawMarkdown": "oki-toki 😏 \nhttps://www.kaggle.com/kmader/skin-cancer-mnist-ham10000", "votes": -1 }, { "id": 974292, "postDate": "2020-08-17T21:13:38.990Z", "content": "I am using efficient-net, Chris notebooks and external data
\nhttps://www.kaggle.com/cdeotte/isic2019-128x128
\nhttps://www.kaggle.com/cdeotte/isic2019-192x192
\nhttps://www.kaggle.com/cdeotte/isic2019-192x192
\nhttps://www.kaggle.com/cdeotte/melanoma-768x768
\nhttps://www.kaggle.com/cdeotte/melanoma-512x512
\nhttps://www.kaggle.com/cdeotte/melanoma-384x384
\nhttps://www.kaggle.com/cdeotte/melanoma-256x256
\nhttps://www.kaggle.com/cdeotte/isic2019-768x768
\nhttps://www.kaggle.com/cdeotte/isic2019-512x512
\nhttps://www.kaggle.com/aziz69/data-pseudo-tf-records-09576-09526?select=train_pseudov200-1063.tfrec
I am using efficient-net, Chris notebooks and external data
\nhttps://www.kaggle.com/cdeotte/isic2019-128x128
\nhttps://www.kaggle.com/cdeotte/isic2019-192x192
\nhttps://www.kaggle.com/cdeotte/isic2019-192x192
\nhttps://www.kaggle.com/cdeotte/melanoma-768x768
\nhttps://www.kaggle.com/cdeotte/melanoma-512x512
\nhttps://www.kaggle.com/cdeotte/melanoma-384x384
\nhttps://www.kaggle.com/cdeotte/melanoma-256x256
\nhttps://www.kaggle.com/cdeotte/isic2019-768x768
\nhttps://www.kaggle.com/cdeotte/isic2019-512x512
I am using efficient-net, Chris notebooks and external data
\nhttps://www.kaggle.com/cdeotte/isic2019-128x128
\nhttps://www.kaggle.com/cdeotte/isic2019-192x192
\nhttps://www.kaggle.com/cdeotte/isic2019-192x192
\nhttps://www.kaggle.com/cdeotte/melanoma-768x768
\nhttps://www.kaggle.com/cdeotte/melanoma-512x512
\nhttps://www.kaggle.com/cdeotte/melanoma-384x384
\nhttps://www.kaggle.com/cdeotte/melanoma-256x256
\nhttps://www.kaggle.com/cdeotte/isic2019-768x768
\nhttps://www.kaggle.com/cdeotte/isic2019-512x512
I am using efficient-net, Chris notebooks and external data
\nhttps://www.kaggle.com/cdeotte/isic2019-128x128
\nhttps://www.kaggle.com/cdeotte/isic2019-192x192
\nhttps://www.kaggle.com/cdeotte/isic2019-192x192
\nhttps://www.kaggle.com/cdeotte/melanoma-768x768
\nhttps://www.kaggle.com/cdeotte/melanoma-512x512
\nhttps://www.kaggle.com/cdeotte/melanoma-384x384
\nhttps://www.kaggle.com/cdeotte/melanoma-256x256
\nhttps://www.kaggle.com/cdeotte/isic2019-768x768
\nhttps://www.kaggle.com/cdeotte/isic2019-512x512
https://www.tensorflow.org/api_docs/python/tf/keras/applications
\nI'm using the pretrained models on Kaggle's Tensorflow, specifically Xception, InceptionV3, and DenseNet201. They have been trained on ImageNet as with other declared models. I intend to compete fairly and I don't think using Tensorflow competes unfairly. I will not be using these submissions for the final leaderboard score.
update: I chose not to take a 0.9384 private score and bronze medal out of consideration for possible rule interpretations, and I'm happy nonetheless.
", "rawMarkdown": "https://www.tensorflow.org/api_docs/python/tf/keras/applications\nI'm using the pretrained models on Kaggle's Tensorflow, specifically Xception, InceptionV3, and DenseNet201. They have been trained on ImageNet as with other declared models. I intend to compete fairly and I don't think using Tensorflow competes unfairly. I will not be using these submissions for the final leaderboard score.\n\nupdate: I chose not to take a 0.9384 private score and bronze medal out of consideration for possible rule interpretations, and I'm happy nonetheless." }, { "id": 965053, "postDate": "2020-08-10T11:00:55.337Z", "content": "I'm using the segmentation data from the ISIC 2018 challenge (task 1)
\nhttps://challenge.isic-archive.com/data#2018
https://challenge2019.isic-archive.com/data.html
\nhttps://github.com/lukemelas/EfficientNet-PyTorch
\nhttps://github.com/Cadene/pretrained-models.pytorch
\nhttps://github.com/pudae/kaggle-hpa
\nhttps://github.com/filipradenovic/cnnimageretrieval-pytorch
PH2_dataset: - https://www.kaggle.com/synked/ph2-modified and 7-point :- https://www.kaggle.com/synked/7pointcriteria
", "rawMarkdown": "\nPH2_dataset: - https://www.kaggle.com/synked/ph2-modified and 7-point :- https://www.kaggle.com/synked/7pointcriteria" }, { "id": 962989, "postDate": "2020-08-08T15:30:42.347Z", "content": "https://github.com/osmr/imgclsmob
\nhttps://github.com/zhanghang1989/ResNeSt
\nhttps://github.com/rwightman/pytorch-image-models
\nhttps://github.com/hszhao/SAN
\nhttps://github.com/google-research/big_transfer
\nhttps://github.com/clovaai/rexnet
\nhttps://github.com/iduta/pyconv
Shall I mention the private kaggle dataset which contains my model output csv or it's fine?
", "rawMarkdown": "Shall I mention the private kaggle dataset which contains my model output csv or it's fine?" }, { "id": 961287, "postDate": "2020-08-07T04:43:43.233Z", "content": "https://github.com/qubvel/efficientnet
", "rawMarkdown": "https://github.com/qubvel/efficientnet" }, { "id": 960580, "postDate": "2020-08-06T14:13:40.683Z", "content": "https://isic-archive.com/api/v1/#!/image/image_doSegmentation\nhttps://github.com/Cadene/pretrained-models.pytorch
", "rawMarkdown": "https://isic-archive.com/api/v1/#!/image/image_doSegmentation\nhttps://github.com/Cadene/pretrained-models.pytorch" }, { "id": 959593, "postDate": "2020-08-05T18:21:53.963Z", "content": "Resized 2020 Comp Data\n768x768 TFRecords with targets, meta, and sample submission\n512x512 TFRecords with targets, meta, and sample submission \nResized 2019 2018 2017 Comp Data\n768x768 TFRecords with target and meta\n512x512 TFRecords with targets and meta
", "rawMarkdown": "Resized 2020 Comp Data\n[768x768 TFRecords with targets, meta, and sample submission](https://www.kaggle.com/cdeotte/melanoma-768x768)\n[512x512 TFRecords with targets, meta, and sample submission ](https://www.kaggle.com/cdeotte/melanoma-512x512)\nResized 2019 2018 2017 Comp Data\n[768x768 TFRecords with target and meta](https://www.kaggle.com/cdeotte/isic2019-768x768)\n[512x512 TFRecords with targets and meta](https://www.kaggle.com/cdeotte/isic2019-512x512)" }, { "id": 954367, "postDate": "2020-08-01T17:22:44.337Z", "rawMarkdown": "" }, { "id": 954364, "postDate": "2020-08-01T17:17:32.103Z", "content": "Here is a link for ISIC archive data\nhttps://www.isic-archive.com/#!/topWithHeader/onlyHeaderTop/gallery
\n\nIs it good as external data?
", "rawMarkdown": "Here is a link for ISIC archive data\nhttps://www.isic-archive.com/#!/topWithHeader/onlyHeaderTop/gallery\n\nIs it good as external data?" }, { "id": 944718, "postDate": "2020-07-25T09:36:03.347Z", "content": "https://github.com/qubvel/classification_models
", "rawMarkdown": "https://github.com/qubvel/classification_models" }, { "id": 942445, "postDate": "2020-07-23T18:41:50.400Z", "content": "https://github.com/rwightman/pytorch-image-models\nhttps://github.com/machine-perception-robotics-group/attention_branch_network
", "rawMarkdown": "https://github.com/rwightman/pytorch-image-models\nhttps://github.com/machine-perception-robotics-group/attention_branch_network" }, { "id": 936664, "postDate": "2020-07-20T12:22:05.073Z", "content": "PyTorch Image Models\nhttps://github.com/rwightman/pytorch-image-models
", "rawMarkdown": "PyTorch Image Models\nhttps://github.com/rwightman/pytorch-image-models" }, { "id": 934876, "postDate": "2020-07-18T21:00:18.510Z", "content": "For anyone looking to use PyTorch with the JPEG Images.\nhttps://www.kaggle.com/whatsthevariance/melanoma-image-data
\n\nI sorted the training images into subfolders of their respective classes so that they can be easily loaded into a PyTorch ImageFolder.
", "rawMarkdown": "For anyone looking to use PyTorch with the JPEG Images.\nhttps://www.kaggle.com/whatsthevariance/melanoma-image-data\n\nI sorted the training images into subfolders of their respective classes so that they can be easily loaded into a PyTorch ImageFolder." }, { "id": 934086, "postDate": "2020-07-18T08:16:49.113Z", "content": "https://github.com/Cadene/pretrained-models.pytorch\nhttps://github.com/lukemelas/EfficientNet-PyTorch\nhttps://github.com/zhanghang1989/ResNeSt
", "rawMarkdown": "https://github.com/Cadene/pretrained-models.pytorch\nhttps://github.com/lukemelas/EfficientNet-PyTorch\nhttps://github.com/zhanghang1989/ResNeSt" }, { "id": 898777, "postDate": "2020-06-23T18:28:05.553Z", "content": "@juliaelliott, can we use images from a standard google image search ? \nFor example : https://www.google.com/search?q=melanoma&source=lnms&tbm=isch&sa=X&ved=2ahUKEwjF--TKxpjqAhUJVN8KHeyXAF8Q_AUoAXoECBEQAw&biw=1563&bih=874
\n\nThey are freely and publicly available. Research/academic licensing restrictions are presumably unknown but I believe usage of Imagenet images is based on the same assumption. Thanks.
", "rawMarkdown": "@juliaelliott, can we use images from a standard google image search ? \nFor example : https://www.google.com/search?q=melanoma&source=lnms&tbm=isch&sa=X&ved=2ahUKEwjF--TKxpjqAhUJVN8KHeyXAF8Q_AUoAXoECBEQAw&biw=1563&bih=874\n\nThey are freely and publicly available. Research/academic licensing restrictions are presumably unknown but I believe usage of Imagenet images is based on the same assumption. Thanks.\n\n", "replies": [ { "id": 903260, "postDate": "2020-06-26T17:23:05.370Z", "content": "@alexandrecc Your question is being discussed by the host along with a few other nuanced external data questions. They will follow-up with a response.
", "rawMarkdown": "@alexandrecc Your question is being discussed by the host along with a few other nuanced external data questions. They will follow-up with a response.", "votes": 2 }, { "id": 913139, "postDate": "2020-07-03T02:42:44.610Z", "content": "Great thank you. Looking forward to see the response.
", "rawMarkdown": "Great thank you. Looking forward to see the response.", "votes": 1 }, { "id": 920502, "postDate": "2020-07-08T16:28:45.097Z", "content": "@alexandrecc Google search (or any search engine results) is an aggregation of web-published images, but does not disclaim the provenance or licensing of each individual image and does not guarantee such licensing abides by the rules of the competition. Therefore, it is still required that you validate the original source datasets' compliance for use in this competition. Imagenet is not equivalent to a Google search, as it is a compiled dataset whose labels are served under the dataset's license for research use. To be clear, it is not permitted to state a Google search as your external data source; you must specify the origin dataset source in declaring any external data.
", "rawMarkdown": "@alexandrecc Google search (or any search engine results) is an aggregation of web-published images, but does not disclaim the provenance or licensing of each individual image and does not guarantee such licensing abides by the rules of the competition. Therefore, it is still required that you validate the original source datasets' compliance for use in this competition. Imagenet is not equivalent to a Google search, as it is a compiled dataset whose labels are served under the dataset's license for research use. To be clear, it is not permitted to state a Google search as your external data source; you must specify the origin dataset source in declaring any external data.", "votes": 2 }, { "id": 922292, "postDate": "2020-07-10T02:17:42.153Z", "content": "Thanks Julia and to the hosts for the answer. Imagenet images are still an aggregation of web-published images. If a publicly available compiled Kaggle dataset is created with a list of URL links to melanoma images across the web, would it be valid ?
", "rawMarkdown": "Thanks Julia and to the hosts for the answer. Imagenet *images* are still an aggregation of web-published images. If a publicly available compiled Kaggle dataset is created with a list of URL links to melanoma images across the web, would it be valid ? " }, { "id": 923515, "postDate": "2020-07-10T23:57:55.633Z", "content": "@alexandrecc If the dataset is compiled and licensed appropriately (publicly, freely, research purposes) for its contents, then it would be valid. It is incumbent on participants to ensure their data use practices are in adherence to rules and law.
", "rawMarkdown": "@alexandrecc If the dataset is compiled and licensed appropriately (publicly, freely, research purposes) for its contents, then it would be valid. It is incumbent on participants to ensure their data use practices are in adherence to rules and law." } ] }, { "id": 888254, "postDate": "2020-06-16T08:24:20.570Z", "content": "so,is it meaning that I can use the data of ISIC2019?
", "rawMarkdown": "so,is it meaning that I can use the data of ISIC2019?", "replies": [ { "id": 897168, "postDate": "2020-06-22T17:01:53.057Z", "content": "@xincai1998 Yes.
", "rawMarkdown": "@xincai1998 Yes." } ] }, { "id": 884800, "postDate": "2020-06-13T16:19:04.777Z", "content": "pretrained models from https://github.com/rwightman/gen-efficientnet-pytorch
", "rawMarkdown": "pretrained models from [https://github.com/rwightman/gen-efficientnet-pytorch](https://github.com/rwightman/gen-efficientnet-pytorch)" }, { "id": 966922, "postDate": "2020-08-11T19:03:08.400Z", "rawMarkdown": "", "isDeleted": true }, { "id": 944419, "postDate": "2020-07-25T05:31:34.357Z", "rawMarkdown": "", "isDeleted": true }, { "id": 941420, "postDate": "2020-07-23T08:03:17.553Z", "rawMarkdown": "", "isDeleted": true, "replies": [ { "id": 941483, "postDate": "2020-07-23T08:38:24.317Z", "content": "@synked No, YoloV5 which uses license GPLv3 is not allowed.\nhttps://github.com/ultralytics/yolov5/blob/master/LICENSE\nhttps://github.com/ultralytics/yolov3/blob/master/LICENSE
\n\nRules section A.4. WINNER LICENSE requires that the source code should have a license allowing \"commercial use\".
", "rawMarkdown": "@synked No, YoloV5 which uses license GPLv3 is not allowed.\nhttps://github.com/ultralytics/yolov5/blob/master/LICENSE\nhttps://github.com/ultralytics/yolov3/blob/master/LICENSE\n\n[Rules](https://www.kaggle.com/c/siim-isic-melanoma-classification/rules) section A.4. WINNER LICENSE requires that the source code should have a license allowing \"commercial use\"." }, { "id": 941542, "postDate": "2020-07-23T09:36:08.637Z", "rawMarkdown": "", "isDeleted": true }, { "id": 942385, "postDate": "2020-07-23T18:06:16.123Z", "content": "@synked I echo @sirishks 's helpful response. You may also refer to my response to his above question in this thread.
", "rawMarkdown": "@synked I echo @sirishks 's helpful response. You may also refer to my response to his [above question in this thread](https://www.kaggle.com/c/siim-isic-melanoma-classification/discussion/154296#942379)." }, { "id": 942782, "postDate": "2020-07-24T02:08:58.353Z", "rawMarkdown": "", "isDeleted": true }, { "id": 942800, "postDate": "2020-07-24T02:18:48.893Z", "rawMarkdown": "", "isDeleted": true } ] }, { "id": 937316, "postDate": "2020-07-20T23:35:38.993Z", "rawMarkdown": "", "isDeleted": true, "replies": [ { "id": 937324, "postDate": "2020-07-20T23:45:31.373Z", "rawMarkdown": "", "isDeleted": true } ] }, { "id": 920520, "postDate": "2020-07-08T16:39:34.323Z", "rawMarkdown": "", "votes": 1, "isDeleted": true }, { "id": 910180, "postDate": "2020-07-01T03:20:31.707Z", "rawMarkdown": "", "votes": 2, "isDeleted": true, "replies": [ { "id": 910695, "postDate": "2020-07-01T10:03:20.787Z", "content": "@aroraaman still no guarantee that it's the proper dataset but it looks like you can download ISIC18 task 1 here https://peltarion.com/knowledge-center/documentation/terms/dataset-licenses/skin-lesion-segmentation
\nSome more can be found here: https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/DBW86T&version=2.0
", "rawMarkdown": "@aroraaman still no guarantee that it's the proper dataset but it looks like you can download ISIC18 task 1 here https://peltarion.com/knowledge-center/documentation/terms/dataset-licenses/skin-lesion-segmentation\n\nSome more can be found here: https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/DBW86T&version=2.0", "votes": 2 }, { "id": 911106, "postDate": "2020-07-01T15:12:52.440Z", "rawMarkdown": "", "votes": 1, "isDeleted": true }, { "id": 911506, "postDate": "2020-07-01T19:28:41.380Z", "rawMarkdown": "", "isDeleted": true } ] }, { "id": 902893, "postDate": "2020-06-26T12:31:14.303Z", "rawMarkdown": "", "isDeleted": true, "replies": [ { "id": 903256, "postDate": "2020-06-26T17:21:26.417Z", "content": "If you are in prize winning standing, your full model and code are expected in order to earn the prize money. Those will be inspected by the host.
\n\nThere is also an open invitation to all competitors, even those not in prize standing, to submit a manuscript/description of their approach and code, if desired at the end of the competition.
", "rawMarkdown": "If you are in prize winning standing, your full model and code are expected in order to earn the prize money. Those will be inspected by the host.\n\nThere is also an open invitation to all competitors, even those not in prize standing, to submit a manuscript/description of their approach and code, if desired at the end of the competition." }, { "id": 903646, "postDate": "2020-06-27T02:55:37.220Z", "rawMarkdown": "", "isDeleted": true } ] }, { "id": 896292, "postDate": "2020-06-22T03:59:36.627Z", "rawMarkdown": "", "isDeleted": true } ], "comments": [ { "id": 886037, "author_name": "raddar", "author_url": "", "post_date": "2020-06-14T17:00:36.730000", "content": "@juliaelliott \nGiven the situation what happened in deepfake challenge, could we get a clarification about this competition? From my personal experience it is practically impossible to get the data with the consent of the patients unless you were to have consent forms with a paragraph that your (as a patient) data can be used for research.
\n\nDoes this mean that all external data here would be considered against the rules as well (tbh have not read it thoroughly yet, but assume similar to deepfake one)
\n\nIf external data was allowed, it is going to look very bad if double standards would be applied to different competitions involving personal data...
", "votes": 27, "replies": [ { "id": 887024, "author_name": "Giba", "author_url": "", "post_date": "2020-06-15T12:40:15.150000", "content": "What I can say is to use external data at your own risk, even if its allowed in competition rules. My own experience.
", "votes": 16, "replies": [] }, { "id": 887949, "author_name": "Julia Elliott", "author_url": "", "post_date": "2020-06-16T02:56:35.190000", "content": "@raddar A quick reply to let you know your question is not being ignored; the host is evaluating their response and will post a clarification as soon as possible.
", "votes": 5, "replies": [] }, { "id": 920461, "author_name": "Veronica Rotemberg", "author_url": "", "post_date": "2020-07-08T15:59:06.990000", "content": "We apologize for the delay on this - we are still discussing internally how to proceed. Thank you!
", "votes": 3, "replies": [] }, { "id": 963580, "author_name": "Janek Idziak", "author_url": "", "post_date": "2020-08-09T06:34:24.573000", "content": "Is there any decision about that yet? @veronicarotemberg @juliaelliott
", "votes": 0, "replies": [] } ] }, { "id": 877677, "author_name": "Alex Shonenkov", "author_url": "", "post_date": "2020-06-07T20:56:04.337000", "content": "I added raw datasets from last years ISIC:\n- ISIC2016: https://www.kaggle.com/shonenkov/isic2016\n- ISIC2017: https://www.kaggle.com/shonenkov/isic2017\n- ISIC2018: https://www.kaggle.com/shonenkov/isic2018
", "votes": 14, "replies": [ { "id": 886409, "author_name": "Aman Arora", "author_url": "", "post_date": "2020-06-15T03:05:01.183000", "content": "Thanks, this is a great place to start using external data!
\n\nThanks for uploading the datasets to Kaggle, really helps.
\n\nUnfortunately, I am unable to access the website, but any chance you have a copy of 2018 Lesion Semgentation dataset please?
@shonenkov did you find duplicates between those 3 datasets using DBSCAN approach?
", "votes": 2, "replies": [] }, { "id": 887407, "author_name": "Alex Shonenkov", "author_url": "", "post_date": "2020-06-15T16:55:25.203000", "content": "@optimo I didn't try. isic16 + isic17 + isic18 gives isic19, maybe it helps you
", "votes": 3, "replies": [] }, { "id": 887627, "author_name": "Optimo", "author_url": "", "post_date": "2020-06-15T19:20:41.717000", "content": "Well I read that isic19 contained previous competition data too but @hengck23 in this comment https://www.kaggle.com/c/siim-isic-melanoma-classification/discussion/155201#880993 seems to be using isic2019 + 16 + archive, so I was wondering whether 2019 = 2017 + 2018 only or 2019 = 2018 + 2017 + 2016. Maybe it's easier to ask him directly.
\n\n@hengck23 would you mind explaining what isic2019 + 16 + archive stands for in your experiments? 😃
", "votes": 0, "replies": [] } ] }, { "id": 886521, "author_name": "datasaurus", "author_url": "", "post_date": "2020-06-15T05:22:59.040000", "content": "I'm confused about this statement in the rules:\n> Participants who do not use Public External Data are not eligible to receive a Prize, but may retain their non-monetary winning features in accordance with Section 10 of the General Rules
\n\nSo does that mean to be eligible for a prize, the use of external data is mandatory?
\n\nAs raddar mentioned, given what happened in the Deepfake competition, this could create issues for us
", "votes": 11, "replies": [ { "id": 886740, "author_name": "Alex Shonenkov", "author_url": "", "post_date": "2020-06-15T08:45:30.390000", "content": "@anjum48 Beautiful rule! 😅🤣
", "votes": 2, "replies": [] }, { "id": 886775, "author_name": "Miroslav Valan", "author_url": "", "post_date": "2020-06-15T09:13:08.870000", "content": "@juliaelliott the same paragraph includes.\nHowever, you will post any External Data used and the provenance of the data to the official competition forum prior to the Entry Deadline, and if the data is publicly available, include the link to such access.
So the external data DOES NOT need to be public.
", "votes": 2, "replies": [] }, { "id": 886787, "author_name": "datasaurus", "author_url": "", "post_date": "2020-06-15T09:19:16.200000", "content": "I think I understand now. You can use private data, but you will be ineligible for the prize.
\n\nI guess this will require very clear guidance around what is public and what is not.
", "votes": 0, "replies": [] }, { "id": 887498, "author_name": "Julia Elliott", "author_url": "", "post_date": "2020-06-15T17:42:09.320000", "content": "It is not required to use external data. If you do use external data and wish to be eligible for prizes, the dataset must be publicly and freely available for use that includes academic/research use with provenance of the data clearly posted to this thread. If you do use external data and it does not meet these requirements to be considered \"Public External Data\" then it will not be prize-eligible.
", "votes": 3, "replies": [] } ] }, { "id": 958676, "author_name": "Chris Deotte", "author_url": "", "post_date": "2020-08-05T05:34:31.420000", "content": "128x128 JPEGs with CSV target, meta, sample submission (240MB)
\n\n1024x1024 TFRecords with targets, meta, and sample submission (8.9GB)
128x128 JPEGs with CSV target and meta (150MB)
\n\nHello,
\n\nHere's some external(ish) datasets:\n- ISIC 2019;\n- ISIC 2018;\n- ISIC 2017
\n\n(I'm aware that the 2019 versions contains both datasets from 2018 and 2017 as well)
\n\nHope it helps\n😄
", "votes": 8, "replies": [ { "id": 864906, "author_name": "raddar", "author_url": "", "post_date": "2020-05-28T09:02:10.913000", "content": "I think all of these are used in competition train set, right?
", "votes": 0, "replies": [] }, { "id": 864989, "author_name": "Larxel", "author_url": "", "post_date": "2020-05-28T10:17:08.773000", "content": "External(ish)....Figured the best place to discover would be this very thread. With only 584 positive cases, this will be crucial.
\n\nSo far, official information doesn't clarify this:
\n\n\n\n\nThe dataset was generated by the International Skin Imaging Collaboration (ISIC) and images are from the following sources: Hospital Clínic de Barcelona, Medical University of Vienna, Memorial Sloan Kettering Cancer Center, Melanoma Institute Australia, The University of Queensland, and the University of Athens Medical School.
\n
It could be the aggregate training data from 2016 to 2019, It could also be train + test of 2019 (which sums to 33k images) and on top of that there's also their online archive with some 24k images to it.
\n\nIf its the first option, that might mean that the testing set is an aggregate from those years as well...and thus teams who participated in the past have probed what will be our private testing set.
\n\nI haven't participated in the past, but I would like to know anyway.
\n\ntl;dr;\nI don't know, I'm asking by suggesting.
", "votes": 2, "replies": [] }, { "id": 865350, "author_name": "Veronica Rotemberg", "author_url": "", "post_date": "2020-05-28T15:23:53.890000", "content": "Hi all - great questions! We've done the best we can to make the train and test datasets here unique. While there may be a few overlapping lesions between the training set here and prior ISIC training datasets and images available at https://isic-archive.com/ or via the ISIC API due to the retrospective nature of our clinical database queries, we expect that they will have been taken with different photographic equipment and be very few overall.
\n\nWe did this because prior datasets don't contain the patient-contextual information (aka other images from the same patient) which we hypothesize might improve performance.
", "votes": 6, "replies": [] }, { "id": 866234, "author_name": "Miroslav Valan", "author_url": "", "post_date": "2020-05-29T08:09:07.233000", "content": "thanks for setting up this challenge @veronicarotemberg. Here are my two cents.
\n\nIt would be super super interesting if other metadata such as location is available. Maybe something to think about in the future.
\n\nYou could use a large radius (i.e. 5km) to avoid privacy issues. Already this could be sufficient to link it to many other sources of information for which evidence exists to have an effect on skin cancer. To name a few with location you get access to a number of sunny days and UV radiation, altitude, air pollution etc.
", "votes": 13, "replies": [] } ] }, { "id": 866123, "author_name": "Amirreza Mahbod", "author_url": "", "post_date": "2020-05-29T06:06:19.723000", "content": "Other skin lesion datasets that may help (excluding the ISIC archive):\nDermofit\nAtlas of Dermotoscopy\nPH2\nDerm7pt
", "votes": 6, "replies": [ { "id": 910204, "author_name": "sajwankit", "author_url": "", "post_date": "2020-07-01T03:51:55.270000", "content": "@ipateam Dermofit library seems not to be public, can we use this then?
", "votes": 1, "replies": [] }, { "id": 910613, "author_name": "Amirreza Mahbod", "author_url": "", "post_date": "2020-07-01T08:51:22.267000", "content": "Yes, you are right. I was not aware of it at the beginning.
", "votes": 1, "replies": [] }, { "id": 918773, "author_name": "chao", "author_url": "", "post_date": "2020-07-07T13:39:14.120000", "content": "In PH2 website:
\n The data included in the PH² database can be used for research and educational purposes. It is important to note that redistribution and commercial use is not allowed.
Using PH2 in kaggle belongs to the condition of 'commercial use' ?
", "votes": 0, "replies": [] }, { "id": 919017, "author_name": "Julia Elliott", "author_url": "", "post_date": "2020-07-07T16:45:06.857000", "content": "@bubble94 In this competition, external data that is licensed \"for use that includes research or academic purposes\" still qualifies as permitted for prize eligibility.\n@ipateam & @ankitsajwan Correct, any dataset that comes at any cost is not permitted for use as \"publicly and freely available\" if you wish to be prize eligible.
", "votes": 1, "replies": [] } ] }, { "id": 942760, "author_name": "Rajnish Chauhan", "author_url": "", "post_date": "2020-07-24T01:36:17.710000", "content": "Following datasets :\nJPEG image dataset :\n2019 and 2020 dataset merged:\n- https://www.kaggle.com/shonenkov/melanoma-merged-external-data-512x512-jpeg
\n\nTFrecords\n- https://www.kaggle.com/cdeotte/isic2019-768x768 \n- https://www.kaggle.com/cdeotte/melanoma-768x768
\n- https://www.kaggle.com/cdeotte/melanoma-512x512\n- https://www.kaggle.com/cdeotte/melanoma-384x384
@juliaelliott: could you please revisit the competition rules and make some correction. Competition rules made confusion again to the participants. Kaggle Team \"MUST\" learn from their mistakes in DeepFake Competition. You must address it with the highest priority. Please respect competitors' free time and hard work they dedicated to this competition @anjum48 thanks for the information regarding conflicting rules.
", "votes": 3, "replies": [ { "id": 887947, "author_name": "Julia Elliott", "author_url": "", "post_date": "2020-06-16T02:55:34.543000", "content": "@projdev The rules in this competition are distinct from other competitions. Please specify where you are confused, and we will work with the hosts to help with clarification.
", "votes": 0, "replies": [] } ] }, { "id": 964082, "author_name": "Guanshuo Xu", "author_url": "", "post_date": "2020-08-09T15:17:31.843000", "content": "https://challenge2019.isic-archive.com/data.html\nhttps://github.com/lukemelas/EfficientNet-PyTorch\nhttps://github.com/Cadene/pretrained-models.pytorch\nhttps://github.com/facebookresearch/semi-supervised-ImageNet1K-models\nhttps://github.com/rwightman/pytorch-image-models
", "votes": 1, "replies": [] }, { "id": 951755, "author_name": "عثمان", "author_url": "", "post_date": "2020-07-30T11:45:08.577000", "content": "Wow, just fumbled across this very nice repo: https://github.com/CoinCheung/pytorch-loss
", "votes": 1, "replies": [] }, { "id": 950325, "author_name": "Dracarys", "author_url": "", "post_date": "2020-07-29T10:38:53.400000", "content": "@juliaelliott @veronicarotemberg Is GAN based melanoma generation allowed in this competition?
\n\nI have also created a separate discussion for this here.
", "votes": 1, "replies": [ { "id": 950862, "author_name": "Julia Elliott", "author_url": "", "post_date": "2020-07-29T17:17:18.567000", "content": "See my response here: https://www.kaggle.com/c/siim-isic-melanoma-classification/discussion/170846#950860 -- You may use generated images, as long as they are not generated from the test set, as that effectively constitutes hand-labeling.
", "votes": 0, "replies": [] }, { "id": 950943, "author_name": "Chris Deotte", "author_url": "", "post_date": "2020-07-29T18:34:39.977000", "content": "@juliaelliott Can we use GANs to generate images without target labels? Is that still considered \"hand-labeling\"? (We can then use these images without labels in unsupervised learning)
", "votes": 1, "replies": [] } ] }, { "id": 940387, "author_name": "Dieter", "author_url": "", "post_date": "2020-07-22T22:49:57.687000", "content": "https://github.com/osmr/imgclsmob/tree/master/pytorch
", "votes": 1, "replies": [] }, { "id": 934882, "author_name": "Innat", "author_url": "", "post_date": "2020-07-18T21:07:51.480000", "content": "GhostNet:\nhttps://www.kaggle.com/ipythonx/ghostnetpretrained
", "votes": 1, "replies": [] }, { "id": 913008, "author_name": "Asad Ali", "author_url": "", "post_date": "2020-07-02T22:07:10.737000", "content": "@juliaelliott One of the fellow participant ( @cdeotte , Huge Thanks) resized the images and combined the meta features into a TFRecords dataset. Does this count as external data?\nLink:https://www.kaggle.com/cdeotte/melanoma-256x256
", "votes": 1, "replies": [ { "id": 913043, "author_name": "Julia Elliott", "author_url": "", "post_date": "2020-07-02T23:06:19.503000", "content": "@spideysloth Yes.
", "votes": 2, "replies": [] }, { "id": 915040, "author_name": "AgeofPerils?", "author_url": "", "post_date": "2020-07-04T12:43:12.667000", "content": "Isn't this just preprocessing? How is it external data?
", "votes": 2, "replies": [] }, { "id": 919012, "author_name": "Julia Elliott", "author_url": "", "post_date": "2020-07-07T16:41:21.093000", "content": "Apologies for not being clear in my prior response, as I don't think I understood the question. What you describe is external data by virtue of being loaded for use in your modeling. However, the dataset itself does not need to be declared separately as external data, as it already constitutes the dataset provided, as @jonathanrayner suggests.
", "votes": 0, "replies": [] } ] }, { "id": 888476, "author_name": "Loulou", "author_url": "", "post_date": "2020-06-16T11:40:56.913000", "content": "Here is a dataset aggregating (some) of the public submissions : https://www.kaggle.com/louise2001/melanoma
", "votes": 1, "replies": [] }, { "id": 878823, "author_name": "Zakirov Jamil", "author_url": "", "post_date": "2020-06-08T22:09:54.283000", "content": "", "votes": 1, "replies": [] }, { "id": 867882, "author_name": "raddar", "author_url": "", "post_date": "2020-05-30T17:47:04.767000", "content": "https://skincancer.net/images/types/
", "votes": 1, "replies": [ { "id": 868003, "author_name": "Innat", "author_url": "", "post_date": "2020-05-30T20:09:28.043000", "content": "Is it OK?
\n\n\n", "votes": 1, "replies": [] }, { "id": 878994, "author_name": "Rony K Roy", "author_url": "", "post_date": "2020-06-09T05:22:49.860000", "content": "The skin cancer images gallery is not all-encompassing and should not be used for diagnostic purposes.
\n
I too have the same question...
", "votes": 0, "replies": [] }, { "id": 879986, "author_name": "Julia Elliott", "author_url": "", "post_date": "2020-06-09T22:18:11.957000", "content": "Hi all, good question. Upon closer inspection, we suspect this message is a disclaimer. We haven't identified any licensing restrictions, and so the host has confirmed skincancer.net is permitted for use from a rules perspective. A side note that we weren't able to determine whether these are dermoscopic images, which could impact how useful they are.
", "votes": 0, "replies": [] } ] }, { "id": 964563, "author_name": "Gary", "author_url": "", "post_date": "2020-08-10T02:25:36.903000", "content": "https://github.com/rwightman/pytorch-image-models
\nhttps://github.com/rwightman/gen-efficientnet-pytorch
\nhttps://github.com/lukemelas/EfficientNet-PyTorch
\nhttps://github.com/clovaai/rexnet
\nhttps://github.com/Cadene/pretrained-models.pytorch
@juliaelliott @veronicarotemberg The Rules section A.1. EXTERNAL DATA defines \"Public External Data\" as External Data available to use by all participants of the competition for purposes of the competition, for use that includes research or academic purposes, at no cost to the other participants.\nNOTE: \"Public External Data\" does not mention dataset \"commercial use\" anywhere!\nHowever, the A.4. WINNER LICENSE section requires that the source code (not dataset) should have a license allowing \"commercial use\"
\n\nIn a new forum post by Chris More Kaggle Datasets - 4000 Malignant Images!, I see participants referring to PH2 and 7-point datasets.
\n\nThese datasets are released as \"CC BY-NC-SA 4.0\", which allows research and educational purposes, but not redistribution and commercial use.
\n\nCould you please clarify whether we can use these Ph2 and 7-point datasets because they meet the \"research, academic, no cost\" requirements of the \"Public External Data\", even though the datasets do not allow commercial use?
", "votes": 0, "replies": [ { "id": 942379, "author_name": "Julia Elliott", "author_url": "", "post_date": "2020-07-23T18:04:24.060000", "content": "@sirishks Thanks for raising this question. As you have observed, the host is permitting use of external data that is public as eligible for prizes, as long as it has been made available, minimally, for research/academic use. However, this is also provided that the resulting solution can also be licensed per the WINNERS LICENSE requirements specified in the competition's rules, which does include commercial use. Therefore, the \"CC BY-NC-SA 4.0\" license (which, incidentally, is similar to how the provided competition dataset is also licensed) would fall under eligible, conditional on the final model ultimately being licensed under one of the eligible licenses in the rules. I'll also add that it is not a requirement to use external data at all.
", "votes": 0, "replies": [ { "id": 942428, "author_name": "Sirish Somanchi", "author_url": "", "post_date": "2020-07-23T18:27:51.960000", "content": "@juliaelliott Thank you for your reply.
\n\nJust to avoid a DFDC situation, I quickly checked the license details and found that one of the licensing terms means that \"If you remix, transform, or build upon the material, you must distribute your contributions under the same license as the original\" i.e. the model built using a dataset licensed like this can not be used for commercial purposes.
\n\nCould you please confirm this conclusion?
\n\nPS: I am also confused because last year winners DAISYLab wrote in their report section 2.1 that they have used \"dermoscopic images from the 7-point dataset\" (which does not allow commercial use).
", "votes": 0, "replies": [] } ] }, { "id": 944995, "author_name": "Janek Idziak", "author_url": "", "post_date": "2020-07-25T13:45:26.520000", "content": "\n\nDo I understand correctly that *we are eligible to use those data * for the final model training and using it wouldn't disqualify us from the competition?
", "votes": 1, "replies": [] }, { "id": 948437, "author_name": "Julia Elliott", "author_url": "", "post_date": "2020-07-27T23:59:52.813000", "content": "@janidziak Yes, datasets that are licensed as available for research/academic use can be used and remain eligible to receive the competition's prizes, as long as you will license your final solution and code as open source, including for commercial use, per the winners' licensing requirements in the rules.
", "votes": 3, "replies": [] } ] }, { "id": 888596, "author_name": "Roman", "author_url": "", "post_date": "2020-06-16T13:00:26.150000", "content": "I have created a dataset which uses original data from this competition with an addition of only malignan cases from 2019 ISIC cometition.\nDataset: https://www.kaggle.com/nroman/melanoma-external-malignant-256\nKernel using it: https://www.kaggle.com/nroman/melanoma-pytorch-starter-efficientnet/comments\nA post: https://www.kaggle.com/c/siim-isic-melanoma-classification/discussion/159101
", "votes": 2, "replies": [ { "id": 942199, "author_name": "Rony", "author_url": "", "post_date": "2020-07-23T16:26:05.143000", "content": "Hey can i use your data ? also any update regarding it. I am a novice
", "votes": 0, "replies": [] } ] }, { "id": 872033, "author_name": "David J. Slate", "author_url": "", "post_date": "2020-06-02T20:45:07.377000", "content": "Am using Keras model weights: https://github.com/fchollet/deep-learning-models/releases/download/v0.7/inception_resnet_v2_weights_tf_dim_ordering_tf_kernels_notop.h5
", "votes": 0, "replies": [] }, { "id": 969925, "author_name": "Hilal Shaath", "author_url": "", "post_date": "2020-08-14T04:16:13.453000", "content": "pie torch efficient
\n2009 cancer data
https://www.isic-archive.com/#!/onlyHeaderTop/gallery
\nISIC melanoma pictures (might be redundant though, see post by Chris Deotte)
", "votes": -1, "replies": [] }, { "id": 966878, "author_name": "Gena", "author_url": "", "post_date": "2020-08-11T18:00:51.513000", "content": "https://github.com/anindox8/Ensemble-of-Multi-Scale-CNN-for-Dermatoscopy-Classification.git
", "votes": -1, "replies": [] }, { "id": 965753, "author_name": "Austin Dibble", "author_url": "", "post_date": "2020-08-10T20:20:56.693000", "content": "https://github.com/pytorch/vision/tree/master/torchvision/models
", "votes": -1, "replies": [] }, { "id": 868032, "author_name": "Innat", "author_url": "", "post_date": "2020-05-30T20:27:53.790000", "content": "oki-toki 😏 \nhttps://www.kaggle.com/kmader/skin-cancer-mnist-ham10000
", "votes": -1, "replies": [] }, { "id": 974292, "author_name": "yash chaudhary", "author_url": "", "post_date": "2020-08-17T21:13:38.990000", "content": "I am using efficient-net, Chris notebooks and external data
\nhttps://www.kaggle.com/cdeotte/isic2019-128x128
\nhttps://www.kaggle.com/cdeotte/isic2019-192x192
\nhttps://www.kaggle.com/cdeotte/isic2019-192x192
\nhttps://www.kaggle.com/cdeotte/melanoma-768x768
\nhttps://www.kaggle.com/cdeotte/melanoma-512x512
\nhttps://www.kaggle.com/cdeotte/melanoma-384x384
\nhttps://www.kaggle.com/cdeotte/melanoma-256x256
\nhttps://www.kaggle.com/cdeotte/isic2019-768x768
\nhttps://www.kaggle.com/cdeotte/isic2019-512x512
\nhttps://www.kaggle.com/aziz69/data-pseudo-tf-records-09576-09526?select=train_pseudov200-1063.tfrec
I am using efficient-net, Chris notebooks and external data
\nhttps://www.kaggle.com/cdeotte/isic2019-128x128
\nhttps://www.kaggle.com/cdeotte/isic2019-192x192
\nhttps://www.kaggle.com/cdeotte/isic2019-192x192
\nhttps://www.kaggle.com/cdeotte/melanoma-768x768
\nhttps://www.kaggle.com/cdeotte/melanoma-512x512
\nhttps://www.kaggle.com/cdeotte/melanoma-384x384
\nhttps://www.kaggle.com/cdeotte/melanoma-256x256
\nhttps://www.kaggle.com/cdeotte/isic2019-768x768
\nhttps://www.kaggle.com/cdeotte/isic2019-512x512
I am using efficient-net, Chris notebooks and external data
\nhttps://www.kaggle.com/cdeotte/isic2019-128x128
\nhttps://www.kaggle.com/cdeotte/isic2019-192x192
\nhttps://www.kaggle.com/cdeotte/isic2019-192x192
\nhttps://www.kaggle.com/cdeotte/melanoma-768x768
\nhttps://www.kaggle.com/cdeotte/melanoma-512x512
\nhttps://www.kaggle.com/cdeotte/melanoma-384x384
\nhttps://www.kaggle.com/cdeotte/melanoma-256x256
\nhttps://www.kaggle.com/cdeotte/isic2019-768x768
\nhttps://www.kaggle.com/cdeotte/isic2019-512x512