Trust me, we debated long and hard about whether to reveal who was doing which annotations. I was outvoted so we kept it secret during the competition. So now it's time for the big reveal, for those who care:
\n\nWe used two strategies for annotation; one is described here (literally outlining each nucleus with a tablet & pen):
\nhttps://www.kaggle.com/c/data-science-bowl-2018/discussion/48347
As far as I'm aware, all scored images in test stage 2 used this approach and were done by myself.
\n\nImages in training set and test stage 1 were a mixture: roughly 10% were done my myself using the strategy above; the remaining 90% were done by a team of people using the annotation tool that Allen made - he plans to make it public now, but doesn't want to support it going forward so don't pester him about it! It was some sort of super-pixel strategy.
\n\nEither way, each image was either annotated by or had annotations checked by a PhD biologist but generally speaking there was only one independent annotation done per image (we thought it better to have 2x as much data vs. having two annotators independently do the same images and provide a consensus labeling).
", "messages": [ { "id": "316624", "postDate": "04/19/2018 14:22:05", "content": "Trust me, we debated long and hard about whether to reveal who was doing which annotations. I was outvoted so we kept it secret during the competition. So now it's time for the big reveal, for those who care:
\n\nWe used two strategies for annotation; one is described here (literally outlining each nucleus with a tablet & pen):
\nhttps://www.kaggle.com/c/data-science-bowl-2018/discussion/48347
As far as I'm aware, all scored images in test stage 2 used this approach and were done by myself.
\n\nImages in training set and test stage 1 were a mixture: roughly 10% were done my myself using the strategy above; the remaining 90% were done by a team of people using the annotation tool that Allen made - he plans to make it public now, but doesn't want to support it going forward so don't pester him about it! It was some sort of super-pixel strategy.
\n\nEither way, each image was either annotated by or had annotations checked by a PhD biologist but generally speaking there was only one independent annotation done per image (we thought it better to have 2x as much data vs. having two annotators independently do the same images and provide a consensus labeling).
", "rawMarkdown": "Trust me, we debated long and hard about whether to reveal who was doing which annotations. I was outvoted so we kept it secret during the competition. So now it's time for the big reveal, for those who care:\n\nWe used two strategies for annotation; one is described here (literally outlining each nucleus with a tablet & pen): \nhttps://www.kaggle.com/c/data-science-bowl-2018/discussion/48347\n\nAs far as I'm aware, all scored images in test stage 2 used this approach and were done by myself. \n\nImages in training set and test stage 1 were a mixture: roughly 10% were done my myself using the strategy above; the remaining 90% were done by a team of people using the annotation tool that Allen made - he plans to make it public now, but doesn't want to support it going forward so don't pester him about it! It was some sort of super-pixel strategy. \n\nEither way, each image was either annotated by or had annotations checked by a PhD biologist but generally speaking there was only one independent annotation done per image (we thought it better to have 2x as much data vs. having two annotators independently do the same images and provide a consensus labeling).", "votes": null }, { "id": "316710", "postDate": "04/19/2018 18:01:15", "content": "my( i should sat her 40 cell masks) were done by digital tablet on microscope images ,by 4th year Ph.D biologist-candidate whose arm I bent. In feb, i sent her the masks and she said they were good. She then made the comment (being even more diplomatic than I am) that it was probably done by some Ph.D candidate slave. Imagine her embarrassment when I let her know you did some them and the rest were other world-class cell biologists. I expect many research labs are downloading this set now or have already done so. what a gift. thank you.
", "rawMarkdown": "my( i should sat her 40 cell masks) were done by digital tablet on microscope images ,by 4th year Ph.D biologist-candidate whose arm I bent. In feb, i sent her the masks and she said they were good. She then made the comment (being even more diplomatic than I am) that it was probably done by some Ph.D candidate slave. Imagine her embarrassment when I let her know you did some them and the rest were other world-class cell biologists. I expect many research labs are downloading this set now or have already done so. what a gift. thank you.", "votes": null } ], "comments": [ { "id": 316710, "author_name": "bcottman", "author_url": "", "post_date": "04/19/2018 18:01:15", "content": "my( i should sat her 40 cell masks) were done by digital tablet on microscope images ,by 4th year Ph.D biologist-candidate whose arm I bent. In feb, i sent her the masks and she said they were good. She then made the comment (being even more diplomatic than I am) that it was probably done by some Ph.D candidate slave. Imagine her embarrassment when I let her know you did some them and the rest were other world-class cell biologists. I expect many research labs are downloading this set now or have already done so. what a gift. thank you.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "316624": "Trust me, we debated long and hard about whether to reveal who was doing which annotations. I was outvoted so we kept it secret during the competition. So now it's time for the big reveal, for those who care:\n\nWe used two strategies for annotation; one is described here (literally outlining each nucleus with a tablet & pen): \nhttps://www.kaggle.com/c/data-science-bowl-2018/discussion/48347\n\nAs far as I'm aware, all scored images in test stage 2 used this approach and were done by myself. \n\nImages in training set and test stage 1 were a mixture: roughly 10% were done my myself using the strategy above; the remaining 90% were done by a team of people using the annotation tool that Allen made - he plans to make it public now, but doesn't want to support it going forward so don't pester him about it! It was some sort of super-pixel strategy. \n\nEither way, each image was either annotated by or had annotations checked by a PhD biologist but generally speaking there was only one independent annotation done per image (we thought it better to have 2x as much data vs. having two annotators independently do the same images and provide a consensus labeling).", "316710": "my( i should sat her 40 cell masks) were done by digital tablet on microscope images ,by 4th year Ph.D biologist-candidate whose arm I bent. In feb, i sent her the masks and she said they were good. She then made the comment (being even more diplomatic than I am) that it was probably done by some Ph.D candidate slave. Imagine her embarrassment when I let her know you did some them and the rest were other world-class cell biologists. I expect many research labs are downloading this set now or have already done so. what a gift. thank you." }, "source": "meta" }