I haven't entered the steel comp, but it looks there are more people in that comp. I wonder what the main differences between these two comps are. Which one do you think is more difficult?
\n\nI have found one: There can be 'no mask' in the steel comp. In this comp, each figure at least contains one mask.
", "messages": [ { "id": "620373", "postDate": "09/07/2019 11:59:15", "content": "I haven't entered the steel comp, but it looks there are more people in that comp. I wonder what the main differences between these two comps are. Which one do you think is more difficult?
\n\nI have found one: There can be 'no mask' in the steel comp. In this comp, each figure at least contains one mask.
", "rawMarkdown": "I haven't entered the steel comp, but it looks there are more people in that comp. I wonder what the main differences between these two comps are. Which one do you think is more difficult?\n\nI have found one: There can be 'no mask' in the steel comp. In this comp, each figure at least contains one mask.", "votes": null }, { "id": "620420", "postDate": "09/07/2019 13:36:33", "content": "Zero mask ratio is very different\n- Cloud: train = 0.467, public = 0.477\n- Steel: train = 0.85887 , public = 0.85674
", "rawMarkdown": "Zero mask ratio is very different\n- Cloud: train = 0.467, public = 0.477\n- Steel: train = 0.85887 , public = 0.85674", "votes": null }, { "id": "620556", "postDate": "09/07/2019 17:09:08", "content": "I would be active in steel myself but I am voting against the submission method by lack of participating.
", "rawMarkdown": "1. Prize money is different.\n2. IMO it's a cleaner problem to solve.\n3. In clouds the measurement error in the TRUTH is terrible. During my 6 Sigma training two decades ago the first thing we were taught to do is to evaluate the measurement method to insure it was capable of making measurements with enough resolution to solve the problem. This competition would have never gotten off the ground with the existing fish/flower/etc definitions and lack of measurement resolution.\n4. Segmentation makes a lot of sense IF the test data was segmented - it's not - it's just almost bounding boxes turned into segments to help account for the huge black band.\n5. A huge shakeup in the leader board at the end can be expected with this data set. \n6. While there is some challenge in taking a poor data set and making something out of it - IMO in real life I would spend 2 hours on the data set and realize that a lot more work was needed to create a data set worthy of spending time in analysis. This reminds me of the belief that if you throw enough crap at the wall some of it will stick.\n7. The authors/sponsors seem to have already solved the problem and are looking for an incremental improvement. \n8. I think the satellites were launched almost two decades ago - so lots of possible unsupervised data available which would be an interesting bit of work - but that gets you to huge investment in download time and hard drive space. \n\nI would be active in steel myself but I am voting against the submission method by lack of participating.", "votes": null }, { "id": "620578", "postDate": "09/07/2019 17:43:11", "content": "Yes. I also concern about the label shape.
", "rawMarkdown": "Yes. I also concern about the label shape.", "votes": null }, { "id": "620879", "postDate": "09/08/2019 03:01:01", "content": "steel has $40,000 for first place, this has $5,000 for first place. I think that explains most of the difference in entrance rates. Most people enter a competition like this before they have time to find out how hard it is, so the difficulty just determines how many stay active instead of giving up early.
", "rawMarkdown": "steel has $40,000 for first place, this has $5,000 for first place. I think that explains most of the difference in entrance rates. Most people enter a competition like this before they have time to find out how hard it is, so the difficulty just determines how many stay active instead of giving up early.", "votes": null }, { "id": "622622", "postDate": "09/09/2019 22:35:39", "content": "I agree that the prize could be the main reason.
", "rawMarkdown": "I agree that the prize could be the main reason.", "votes": null }, { "id": "622623", "postDate": "09/09/2019 22:35:59", "content": "Thanks. This is important.
", "rawMarkdown": "Thanks. This is important.", "votes": null }, { "id": "622878", "postDate": "09/10/2019 07:46:02", "content": "What does this exactly imply?
", "rawMarkdown": "What does this exactly imply?", "votes": null }, { "id": "623056", "postDate": "09/10/2019 12:35:53", "content": "You may need to first classify which figure has no mask because of the high zero-mask ratio in the steel comp.
", "rawMarkdown": "You may need to first classify which figure has no mask because of the high zero-mask ratio in the steel comp.", "votes": null } ], "comments": [ { "id": 620420, "author_name": "tikutiku", "author_url": "", "post_date": "09/07/2019 13:36:33", "content": "Zero mask ratio is very different\n- Cloud: train = 0.467, public = 0.477\n- Steel: train = 0.85887 , public = 0.85674
", "votes": null, "replies": [ { "id": 622623, "author_name": "gogo827jz", "author_url": "", "post_date": "09/09/2019 22:35:59", "content": "Thanks. This is important.
", "votes": null, "replies": [] }, { "id": 622878, "author_name": "philippsinger", "author_url": "", "post_date": "09/10/2019 07:46:02", "content": "What does this exactly imply?
", "votes": null, "replies": [] }, { "id": 623056, "author_name": "gogo827jz", "author_url": "", "post_date": "09/10/2019 12:35:53", "content": "You may need to first classify which figure has no mask because of the high zero-mask ratio in the steel comp.
", "votes": null, "replies": [] } ] }, { "id": 620556, "author_name": "pcjimmmy", "author_url": "", "post_date": "09/07/2019 17:09:08", "content": "I would be active in steel myself but I am voting against the submission method by lack of participating.
", "votes": null, "replies": [ { "id": 620578, "author_name": "gogo827jz", "author_url": "", "post_date": "09/07/2019 17:43:11", "content": "Yes. I also concern about the label shape.
", "votes": null, "replies": [] } ] }, { "id": 620879, "author_name": "angrim", "author_url": "", "post_date": "09/08/2019 03:01:01", "content": "steel has $40,000 for first place, this has $5,000 for first place. I think that explains most of the difference in entrance rates. Most people enter a competition like this before they have time to find out how hard it is, so the difficulty just determines how many stay active instead of giving up early.
", "votes": null, "replies": [ { "id": 622622, "author_name": "gogo827jz", "author_url": "", "post_date": "09/09/2019 22:35:39", "content": "I agree that the prize could be the main reason.
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "620373": "I haven't entered the steel comp, but it looks there are more people in that comp. I wonder what the main differences between these two comps are. Which one do you think is more difficult?\n\nI have found one: There can be 'no mask' in the steel comp. In this comp, each figure at least contains one mask.", "620420": "Zero mask ratio is very different\n- Cloud: train = 0.467, public = 0.477\n- Steel: train = 0.85887 , public = 0.85674", "620556": "1. Prize money is different.\n2. IMO it's a cleaner problem to solve.\n3. In clouds the measurement error in the TRUTH is terrible. During my 6 Sigma training two decades ago the first thing we were taught to do is to evaluate the measurement method to insure it was capable of making measurements with enough resolution to solve the problem. This competition would have never gotten off the ground with the existing fish/flower/etc definitions and lack of measurement resolution.\n4. Segmentation makes a lot of sense IF the test data was segmented - it's not - it's just almost bounding boxes turned into segments to help account for the huge black band.\n5. A huge shakeup in the leader board at the end can be expected with this data set. \n6. While there is some challenge in taking a poor data set and making something out of it - IMO in real life I would spend 2 hours on the data set and realize that a lot more work was needed to create a data set worthy of spending time in analysis. This reminds me of the belief that if you throw enough crap at the wall some of it will stick.\n7. The authors/sponsors seem to have already solved the problem and are looking for an incremental improvement. \n8. I think the satellites were launched almost two decades ago - so lots of possible unsupervised data available which would be an interesting bit of work - but that gets you to huge investment in download time and hard drive space. \n\nI would be active in steel myself but I am voting against the submission method by lack of participating.", "620578": "Yes. I also concern about the label shape.", "620879": "steel has $40,000 for first place, this has $5,000 for first place. I think that explains most of the difference in entrance rates. Most people enter a competition like this before they have time to find out how hard it is, so the difficulty just determines how many stay active instead of giving up early.", "622622": "I agree that the prize could be the main reason.", "622623": "Thanks. This is important.", "622878": "What does this exactly imply?", "623056": "You may need to first classify which figure has no mask because of the high zero-mask ratio in the steel comp." }, "source": "meta" }