![\"img](\"https://storage.googleapis.com/kaggle-forum-message-attachments/563184/13684/previous.jpeg\")
https://www.kaggle.com/c/diabetic-retinopathy-detection\nI just found this competition which was released 4 years ago, maybe it can bring us some insights.
\n\n1st place solution: https://www.kaggle.com/c/diabetic-retinopathy-detection/discussion/15801#latest-370950\n2nd place solution: https://www.kaggle.com/c/diabetic-retinopathy-detection/discussion/15617#latest-373487\n3rd place solution: https://www.kaggle.com/c/diabetic-retinopathy-detection/discussion/15845#latest-147181
\n\nUpdate:\nhttps://www.kaggle.com/c/diabetic-retinopathy-detection/discussion/16149\nAs we can see, the labels for testset (pre competition) is disclosed, thus we can use full dataset.
", "messages": [ { "id": 563184, "postDate": "2019-06-28T02:15:28.307Z", "content": "https://www.kaggle.com/c/diabetic-retinopathy-detection\nI just found this competition which was released 4 years ago, maybe it can bring us some insights.
\n\n1st place solution: https://www.kaggle.com/c/diabetic-retinopathy-detection/discussion/15801#latest-370950\n2nd place solution: https://www.kaggle.com/c/diabetic-retinopathy-detection/discussion/15617#latest-373487\n3rd place solution: https://www.kaggle.com/c/diabetic-retinopathy-detection/discussion/15845#latest-147181
\n\nUpdate:\nhttps://www.kaggle.com/c/diabetic-retinopathy-detection/discussion/16149\nAs we can see, the labels for testset (pre competition) is disclosed, thus we can use full dataset.
", "rawMarkdown": "https://www.kaggle.com/c/diabetic-retinopathy-detection\nI just found this competition which was released 4 years ago, maybe it can bring us some insights.\n\n1st place solution: https://www.kaggle.com/c/diabetic-retinopathy-detection/discussion/15801#latest-370950\n2nd place solution: https://www.kaggle.com/c/diabetic-retinopathy-detection/discussion/15617#latest-373487\n3rd place solution: https://www.kaggle.com/c/diabetic-retinopathy-detection/discussion/15845#latest-147181\n\nUpdate:\nhttps://www.kaggle.com/c/diabetic-retinopathy-detection/discussion/16149\nAs we can see, the labels for testset (pre competition) is disclosed, thus we can use full dataset.\n", "votes": 141 }, { "id": 564286, "postDate": "2019-06-29T08:33:28.280Z", "content": "other materials are,
\n\n5th: https://www.kaggle.com/c/diabetic-retinopathy-detection/discussion/15585#latest-88261\n6th: https://www.kaggle.com/c/diabetic-retinopathy-detection/discussion/18411#latest-104787\n20th: https://www.kaggle.com/c/diabetic-retinopathy-detection/discussion/15589#latest-87760
", "rawMarkdown": "other materials are, \n\n5th: https://www.kaggle.com/c/diabetic-retinopathy-detection/discussion/15585#latest-88261\n6th: https://www.kaggle.com/c/diabetic-retinopathy-detection/discussion/18411#latest-104787\n20th: https://www.kaggle.com/c/diabetic-retinopathy-detection/discussion/15589#latest-87760\n", "votes": 5 }, { "id": 563194, "postDate": "2019-06-28T02:36:58.910Z", "content": "Both of these competitions have QWK metric and 0-4 scale of metric, maybe we can pre-train under that dataset first.
", "rawMarkdown": "Both of these competitions have QWK metric and 0-4 scale of metric, maybe we can pre-train under that dataset first.", "votes": 6 }, { "id": 563613, "postDate": "2019-06-28T13:47:32.367Z", "content": "I fear that this extra data may have a huge impact on this competition. Its transfer learning potential looks promising and it's huge, so not many people can fully take advantage of it, which is a bad thing.
", "rawMarkdown": "I fear that this extra data may have a huge impact on this competition. Its transfer learning potential looks promising and it's huge, so not many people can fully take advantage of it, which is a bad thing.", "votes": 4, "replies": [ { "id": 563940, "postDate": "2019-06-28T19:44:04.997Z", "content": "I feel the same. But at the same time this extra data gives opportunitiy to make one more step for complex life problem.
\n\nSuch competitions have special places....
", "rawMarkdown": "I feel the same. But at the same time this extra data gives opportunitiy to make one more step for complex life problem. \n\nSuch competitions have special places.... ", "votes": 4 } ] }, { "id": 566638, "postDate": "2019-07-02T12:07:32.650Z", "content": "Thanks for sharing this. It will be helpful.\nmany healthcare companies have implemented this use case. The biggest challenge these companies had - it was accuracy.
\n\nhttps://ai.googleblog.com/2018/12/improving-effectiveness-of-diabetic.html
", "rawMarkdown": "Thanks for sharing this. It will be helpful.\nmany healthcare companies have implemented this use case. The biggest challenge these companies had - it was accuracy. \n\nhttps://ai.googleblog.com/2018/12/improving-effectiveness-of-diabetic.html ", "votes": 1 }, { "id": 565202, "postDate": "2019-06-30T15:23:58.983Z", "content": "great work!!
", "rawMarkdown": "great work!!\n", "votes": 1 }, { "id": 563910, "postDate": "2019-06-28T19:16:44.397Z", "content": "Thanks for compiling the list together Venn!
", "rawMarkdown": "Thanks for compiling the list together Venn! ", "votes": 1 }, { "id": 565135, "postDate": "2019-06-30T13:41:22.837Z", "content": "It is a nice dataset, I put a few more visualizations of it here as well. Maybe some small differences but for transfer learning it seems like something very good to look at. You can also find my dataset there, it provides a combined set of the previous data. There seems to be some issue (at least for me) with using the previous competition data directly as a data source..
", "rawMarkdown": "It is a nice dataset, I put a few more visualizations of it [here](https://www.kaggle.com/donkeys/looking-in-the-eyes-of-past-and-present) as well. Maybe some small differences but for transfer learning it seems like something very good to look at. You can also find my dataset there, it provides a combined set of the previous data. There seems to be some issue (at least for me) with using the previous competition data directly as a data source..", "votes": 2 }, { "id": 563475, "postDate": "2019-06-28T10:03:19.733Z", "content": "Thanks for this and this is really helpful, infact the evaluation metric is the same too. Thanks for sharing this, the difference here is that this is the new format kernel competition and the data is 1/8 the size now..
", "rawMarkdown": "Thanks for this and this is really helpful, infact the evaluation metric is the same too. Thanks for sharing this, the difference here is that this is the new format kernel competition and the data is 1/8 the size now..", "votes": 2, "replies": [ { "id": 563585, "postDate": "2019-06-28T13:06:48.940Z", "content": "Yes Vishy. Well, as described in other posts, we can train models via our own machines, thus an 8 times larger dataset should be helpful : )
", "rawMarkdown": "Yes Vishy. Well, as described in other posts, we can train models via our own machines, thus an 8 times larger dataset should be helpful : )", "votes": 1 } ] }, { "id": 563226, "postDate": "2019-06-28T03:34:20.173Z", "content": "
![\"img](\"https://storage.googleapis.com/kaggle-forum-message-attachments/563184/13685/current.png\")
As you can see, the images are similar, thus now I am sure that we should use the previous dataset first.
", "rawMarkdown": "\n\n\n\nAs you can see, the images are similar, thus now I am sure that we should use the previous dataset first.", "votes": 2 }, { "id": 608905, "postDate": "2019-08-27T09:02:25.357Z", "content": "Good Work 🤝
", "rawMarkdown": "Good Work 🤝 " }, { "id": 608847, "postDate": "2019-08-27T07:55:17.830Z", "content": "Awesome Work!
", "rawMarkdown": "Awesome Work!" }, { "id": 582857, "postDate": "2019-07-23T17:01:37.710Z", "content": "Thank you so much for sharing, there is a lot to learn from the links.
", "rawMarkdown": "Thank you so much for sharing, there is a lot to learn from the links.\n" }, { "id": 572024, "postDate": "2019-07-10T11:05:41.243Z", "content": "Great work!
", "rawMarkdown": "Great work!" }, { "id": 570505, "postDate": "2019-07-08T11:44:32.927Z", "content": "Great Job!
", "rawMarkdown": "Great Job!" }, { "id": 608904, "postDate": "2019-08-27T09:01:37.997Z", "rawMarkdown": "", "isDeleted": true }, { "id": 566719, "postDate": "2019-07-02T14:05:40.110Z", "rawMarkdown": "", "votes": 1, "isDeleted": true }, { "id": 564819, "postDate": "2019-06-30T03:43:12.533Z", "rawMarkdown": "", "votes": 2, "isDeleted": true, "replies": [ { "id": 564957, "postDate": "2019-06-30T08:40:38.607Z", "content": "It was held 4 years ago. There is lot more evolved during this time. So, we can hope much better results this time :)
", "rawMarkdown": "It was held 4 years ago. There is lot more evolved during this time. So, we can hope much better results this time :)", "votes": 2 } ] }, { "id": 563775, "postDate": "2019-06-28T16:55:10.040Z", "rawMarkdown": "", "votes": 1, "isDeleted": true }, { "id": 569344, "postDate": "2019-07-06T13:52:37.357Z", "content": "Many thanks @yiheng
", "rawMarkdown": "Many thanks @yiheng ", "votes": 3 }, { "id": 569992, "postDate": "2019-07-07T16:34:29.583Z", "content": "Thanks for the information @yiheng
", "rawMarkdown": "Thanks for the information @yiheng", "votes": 1 }, { "id": 569800, "postDate": "2019-07-07T11:19:48.593Z", "content": "Thanks for your job.
", "rawMarkdown": "Thanks for your job.", "votes": 1 }, { "id": 569413, "postDate": "2019-07-06T15:42:39.900Z", "content": "That's great! Thanks a lot.
", "rawMarkdown": "That's great! Thanks a lot.", "votes": 1 }, { "id": 565736, "postDate": "2019-07-01T10:16:34.107Z", "content": "@Venn Thanks a lot!
", "rawMarkdown": "@Venn Thanks a lot!", "votes": 1 }, { "id": 565364, "postDate": "2019-06-30T20:29:18.530Z", "content": "Thank you @yiheng!
", "rawMarkdown": "Thank you @yiheng!", "votes": 1 }, { "id": 563722, "postDate": "2019-06-28T15:55:39.723Z", "content": "thank you @yiheng
", "rawMarkdown": "thank you @yiheng ", "votes": 1 }, { "id": 563504, "postDate": "2019-06-28T10:47:08.293Z", "content": "thank you very much for sharing
", "rawMarkdown": "thank you very much for sharing", "votes": 1 }, { "id": 563305, "postDate": "2019-06-28T05:53:56.557Z", "content": "many thanks @yiheng
", "rawMarkdown": "many thanks @yiheng ", "votes": 1 }, { "id": 563217, "postDate": "2019-06-28T03:14:00.760Z", "content": "Thanks venn
", "rawMarkdown": "Thanks venn", "votes": 1 }, { "id": 564676, "postDate": "2019-06-29T19:37:14.037Z", "content": "Thanks
", "rawMarkdown": "Thanks", "votes": 2 }, { "id": 564548, "postDate": "2019-06-29T15:33:51.620Z", "content": "Thanks for reference.
", "rawMarkdown": "Thanks for reference.", "votes": 2 }, { "id": 564184, "postDate": "2019-06-29T05:06:46.387Z", "content": "thanks for the refrence....
", "rawMarkdown": "thanks for the refrence....", "votes": 2 }, { "id": 564026, "postDate": "2019-06-28T22:56:35.970Z", "content": "thank you~
", "rawMarkdown": "thank you~", "votes": 2 } ], "comments": [ { "id": 564286, "author_name": "Maxwell", "author_url": "", "post_date": "2019-06-29T08:33:28.280000", "content": "other materials are,
\n\n5th: https://www.kaggle.com/c/diabetic-retinopathy-detection/discussion/15585#latest-88261\n6th: https://www.kaggle.com/c/diabetic-retinopathy-detection/discussion/18411#latest-104787\n20th: https://www.kaggle.com/c/diabetic-retinopathy-detection/discussion/15589#latest-87760
", "votes": 5, "replies": [] }, { "id": 563194, "author_name": "Yiheng Wang", "author_url": "", "post_date": "2019-06-28T02:36:58.910000", "content": "Both of these competitions have QWK metric and 0-4 scale of metric, maybe we can pre-train under that dataset first.
", "votes": 6, "replies": [] }, { "id": 563613, "author_name": "Khoi Nguyen", "author_url": "", "post_date": "2019-06-28T13:47:32.367000", "content": "I fear that this extra data may have a huge impact on this competition. Its transfer learning potential looks promising and it's huge, so not many people can fully take advantage of it, which is a bad thing.
", "votes": 4, "replies": [ { "id": 563940, "author_name": "Alexander Kireev", "author_url": "", "post_date": "2019-06-28T19:44:04.997000", "content": "I feel the same. But at the same time this extra data gives opportunitiy to make one more step for complex life problem.
\n\nSuch competitions have special places....
", "votes": 4, "replies": [] } ] }, { "id": 566638, "author_name": "Ankit", "author_url": "", "post_date": "2019-07-02T12:07:32.650000", "content": "Thanks for sharing this. It will be helpful.\nmany healthcare companies have implemented this use case. The biggest challenge these companies had - it was accuracy.
\n\nhttps://ai.googleblog.com/2018/12/improving-effectiveness-of-diabetic.html
", "votes": 1, "replies": [] }, { "id": 565202, "author_name": "vortexash", "author_url": "", "post_date": "2019-06-30T15:23:58.983000", "content": "great work!!
", "votes": 1, "replies": [] }, { "id": 563910, "author_name": "Kaan Donbekci", "author_url": "", "post_date": "2019-06-28T19:16:44.397000", "content": "Thanks for compiling the list together Venn!
", "votes": 1, "replies": [] }, { "id": 565135, "author_name": "averagemn", "author_url": "", "post_date": "2019-06-30T13:41:22.837000", "content": "It is a nice dataset, I put a few more visualizations of it here as well. Maybe some small differences but for transfer learning it seems like something very good to look at. You can also find my dataset there, it provides a combined set of the previous data. There seems to be some issue (at least for me) with using the previous competition data directly as a data source..
", "votes": 2, "replies": [] }, { "id": 563475, "author_name": "Vishy", "author_url": "", "post_date": "2019-06-28T10:03:19.733000", "content": "Thanks for this and this is really helpful, infact the evaluation metric is the same too. Thanks for sharing this, the difference here is that this is the new format kernel competition and the data is 1/8 the size now..
", "votes": 2, "replies": [ { "id": 563585, "author_name": "Yiheng Wang", "author_url": "", "post_date": "2019-06-28T13:06:48.940000", "content": "Yes Vishy. Well, as described in other posts, we can train models via our own machines, thus an 8 times larger dataset should be helpful : )
", "votes": 1, "replies": [] } ] }, { "id": 563226, "author_name": "Yiheng Wang", "author_url": "", "post_date": "2019-06-28T03:34:20.173000", "content": "
As you can see, the images are similar, thus now I am sure that we should use the previous dataset first.
", "votes": 2, "replies": [] }, { "id": 608905, "author_name": "Alex Jose", "author_url": "", "post_date": "2019-08-27T09:02:25.357000", "content": "Good Work 🤝
", "votes": 0, "replies": [] }, { "id": 608847, "author_name": "Chandan Tripathy", "author_url": "", "post_date": "2019-08-27T07:55:17.830000", "content": "Awesome Work!
", "votes": 0, "replies": [] }, { "id": 582857, "author_name": "purnasaiG", "author_url": "", "post_date": "2019-07-23T17:01:37.710000", "content": "Thank you so much for sharing, there is a lot to learn from the links.
", "votes": 0, "replies": [] }, { "id": 572024, "author_name": "JoshNel", "author_url": "", "post_date": "2019-07-10T11:05:41.243000", "content": "Great work!
", "votes": 0, "replies": [] }, { "id": 570505, "author_name": "Prakash Patel", "author_url": "", "post_date": "2019-07-08T11:44:32.927000", "content": "Great Job!
", "votes": 0, "replies": [] }, { "id": 608904, "author_name": "", "author_url": "", "post_date": "2019-08-27T09:01:37.997000", "content": "", "votes": 0, "replies": [] }, { "id": 566719, "author_name": "", "author_url": "", "post_date": "2019-07-02T14:05:40.110000", "content": "", "votes": 1, "replies": [] }, { "id": 564819, "author_name": "", "author_url": "", "post_date": "2019-06-30T03:43:12.533000", "content": "", "votes": 2, "replies": [ { "id": 564957, "author_name": "Filemon", "author_url": "", "post_date": "2019-06-30T08:40:38.607000", "content": "It was held 4 years ago. There is lot more evolved during this time. So, we can hope much better results this time :)
", "votes": 2, "replies": [] } ] }, { "id": 563775, "author_name": "", "author_url": "", "post_date": "2019-06-28T16:55:10.040000", "content": "", "votes": 1, "replies": [] }, { "id": 569344, "author_name": "torch", "author_url": "", "post_date": "2019-07-06T13:52:37.357000", "content": "Many thanks @yiheng
", "votes": 3, "replies": [] }, { "id": 569992, "author_name": "C4rl05/V", "author_url": "", "post_date": "2019-07-07T16:34:29.583000", "content": "Thanks for the information @yiheng
", "votes": 1, "replies": [] }, { "id": 569800, "author_name": "ynhuhu", "author_url": "", "post_date": "2019-07-07T11:19:48.593000", "content": "Thanks for your job.
", "votes": 1, "replies": [] }, { "id": 569413, "author_name": "Nguyễn Hoà", "author_url": "", "post_date": "2019-07-06T15:42:39.900000", "content": "That's great! Thanks a lot.
", "votes": 1, "replies": [] }, { "id": 565736, "author_name": "adrofa", "author_url": "", "post_date": "2019-07-01T10:16:34.107000", "content": "@Venn Thanks a lot!
", "votes": 1, "replies": [] }, { "id": 565364, "author_name": "Andrii Litvynchuk", "author_url": "", "post_date": "2019-06-30T20:29:18.530000", "content": "Thank you @yiheng!
", "votes": 1, "replies": [] }, { "id": 563722, "author_name": "Sani Kamal", "author_url": "", "post_date": "2019-06-28T15:55:39.723000", "content": "thank you @yiheng
", "votes": 1, "replies": [] }, { "id": 563504, "author_name": "shilpa.rpns", "author_url": "", "post_date": "2019-06-28T10:47:08.293000", "content": "thank you very much for sharing
", "votes": 1, "replies": [] }, { "id": 563305, "author_name": "FGPC", "author_url": "", "post_date": "2019-06-28T05:53:56.557000", "content": "many thanks @yiheng
", "votes": 1, "replies": [] }, { "id": 563217, "author_name": "seefun", "author_url": "", "post_date": "2019-06-28T03:14:00.760000", "content": "Thanks venn
", "votes": 1, "replies": [] }, { "id": 564676, "author_name": "Sumit Mishra", "author_url": "", "post_date": "2019-06-29T19:37:14.037000", "content": "Thanks
", "votes": 2, "replies": [] }, { "id": 564548, "author_name": "", "author_url": "", "post_date": "2019-06-29T15:33:51.620000", "content": "", "votes": 2, "replies": [] }, { "id": 564184, "author_name": "", "author_url": "", "post_date": "2019-06-29T05:06:46.387000", "content": "", "votes": 2, "replies": [] }, { "id": 564026, "author_name": "", "author_url": "", "post_date": "2019-06-28T22:56:35.970000", "content": "", "votes": 2, "replies": [] } ], "raw_markdown_by_id": { "563184": "https://www.kaggle.com/c/diabetic-retinopathy-detection\nI just found this competition which was released 4 years ago, maybe it can bring us some insights.\n\n1st place solution: https://www.kaggle.com/c/diabetic-retinopathy-detection/discussion/15801#latest-370950\n2nd place solution: https://www.kaggle.com/c/diabetic-retinopathy-detection/discussion/15617#latest-373487\n3rd place solution: https://www.kaggle.com/c/diabetic-retinopathy-detection/discussion/15845#latest-147181\n\nUpdate:\nhttps://www.kaggle.com/c/diabetic-retinopathy-detection/discussion/16149\nAs we can see, the labels for testset (pre competition) is disclosed, thus we can use full dataset.\n", "564286": "other materials are, \n\n5th: https://www.kaggle.com/c/diabetic-retinopathy-detection/discussion/15585#latest-88261\n6th: https://www.kaggle.com/c/diabetic-retinopathy-detection/discussion/18411#latest-104787\n20th: https://www.kaggle.com/c/diabetic-retinopathy-detection/discussion/15589#latest-87760\n", "563194": "Both of these competitions have QWK metric and 0-4 scale of metric, maybe we can pre-train under that dataset first.", "563613": "I fear that this extra data may have a huge impact on this competition. Its transfer learning potential looks promising and it's huge, so not many people can fully take advantage of it, which is a bad thing.", "566638": "Thanks for sharing this. It will be helpful.\nmany healthcare companies have implemented this use case. The biggest challenge these companies had - it was accuracy. \n\nhttps://ai.googleblog.com/2018/12/improving-effectiveness-of-diabetic.html ", "565202": "great work!!\n", "563910": "Thanks for compiling the list together Venn! ", "565135": "It is a nice dataset, I put a few more visualizations of it [here](https://www.kaggle.com/donkeys/looking-in-the-eyes-of-past-and-present) as well. Maybe some small differences but for transfer learning it seems like something very good to look at. You can also find my dataset there, it provides a combined set of the previous data. There seems to be some issue (at least for me) with using the previous competition data directly as a data source..", "563475": "Thanks for this and this is really helpful, infact the evaluation metric is the same too. Thanks for sharing this, the difference here is that this is the new format kernel competition and the data is 1/8 the size now..", "563226": "\n\n\n\nAs you can see, the images are similar, thus now I am sure that we should use the previous dataset first.", "608905": "Good Work 🤝 ", "608847": "Awesome Work!", "582857": "Thank you so much for sharing, there is a lot to learn from the links.\n", "572024": "Great work!", "570505": "Great Job!", "608904": "", "566719": "", "564819": "", "563775": "", "569344": "Many thanks @yiheng ", "569992": "Thanks for the information @yiheng", "569800": "Thanks for your job.", "569413": "That's great! Thanks a lot.", "565736": "@Venn Thanks a lot!", "565364": "Thank you @yiheng!", "563722": "thank you @yiheng ", "563504": "thank you very much for sharing", "563305": "many thanks @yiheng ", "563217": "Thanks venn", "564676": "Thanks", "564548": "Thanks for reference.", "564184": "thanks for the refrence....", "564026": "thank you~" } }