Thanks to all of you that participated in this challenge!
\n\nWe are very happy with the results so far and look forward to diving into the details of your solutions. We are convinced of that they will contribute to an improvement of data in the Human Protein Atlas database and help to accelerate basic biological science.
\n\nAs announced in the challenge description, we aim to write an Analysis article for Nature Methods where we compare approaches and discuss the outcome of the challenge. For this purpose, all teams are hereby invited to submit details about their solutions through this survey. Based on this survey, a maximum of 10 teams will be selected for participation as co-authors. It will be required that the solutions from the selected teams are uploaded to the team page.
\n\nThe deadline to fill out the survey is by the midnight of January 16th. We will notify the selected teams on January 22nd.
\n\nThank you all again!\nEmma and the Human Protein Atlas team
\n\nUpdate Dec 1 2019:\nThe resulting article \"Analysis of the Human Protein Atlas Image Classification Competition\" is now published in Nature Methods. We would like to thank all the contributing teams, it has been a pleasure working with you! We now have models that perform 20% better than previous models for protein subcellular pattern classification, a great step forward for cell biology research. The top ranking models can also be found in our model zoo.
\n\nA free pdf copy of the paper can be found here.
\n\nThank you all for a great competition!\nEmma and the Human Protein Atlas team
", "messages": [ { "id": 454645, "postDate": "2019-01-11T23:10:44.707Z", "content": "Thanks to all of you that participated in this challenge!
\n\nWe are very happy with the results so far and look forward to diving into the details of your solutions. We are convinced of that they will contribute to an improvement of data in the Human Protein Atlas database and help to accelerate basic biological science.
\n\nAs announced in the challenge description, we aim to write an Analysis article for Nature Methods where we compare approaches and discuss the outcome of the challenge. For this purpose, all teams are hereby invited to submit details about their solutions through this survey. Based on this survey, a maximum of 10 teams will be selected for participation as co-authors. It will be required that the solutions from the selected teams are uploaded to the team page.
\n\nThe deadline to fill out the survey is by the midnight of January 16th. We will notify the selected teams on January 22nd.
\n\nThank you all again!\nEmma and the Human Protein Atlas team
\n\nUpdate Dec 1 2019:\nThe resulting article \"Analysis of the Human Protein Atlas Image Classification Competition\" is now published in Nature Methods. We would like to thank all the contributing teams, it has been a pleasure working with you! We now have models that perform 20% better than previous models for protein subcellular pattern classification, a great step forward for cell biology research. The top ranking models can also be found in our model zoo.
\n\nA free pdf copy of the paper can be found here.
\n\nThank you all for a great competition!\nEmma and the Human Protein Atlas team
", "rawMarkdown": "Thanks to all of you that participated in this challenge! \n\nWe are very happy with the results so far and look forward to diving into the details of your solutions. We are convinced of that they will contribute to an improvement of data in the Human Protein Atlas database and help to accelerate basic biological science.\n\nAs announced in the challenge description, we aim to write an Analysis article for Nature Methods where we compare approaches and discuss the outcome of the challenge. For this purpose, all teams are hereby invited to submit details about their solutions through this [survey][1]. Based on this survey, a maximum of 10 teams will be selected for participation as co-authors. It will be required that the solutions from the selected teams are uploaded to the team page.\n\nThe deadline to fill out the survey is by the midnight of January 16th. We will notify the selected teams on January 22nd.\n\nThank you all again!\nEmma and the Human Protein Atlas team\n\n-------------\nUpdate Dec 1 2019:\nThe resulting article \"[Analysis of the Human Protein Atlas Image Classification Competition](https://www.nature.com/articles/s41592-019-0658-6)\" is now published in Nature Methods. We would like to thank all the contributing teams, it has been a pleasure working with you! We now have models that perform 20% better than previous models for protein subcellular pattern classification, a great step forward for cell biology research. The top ranking models can also be found in our [model zoo](https://modelzoo.cellprofiling.org/#/).\n\nA free pdf copy of the paper can be found [here](https://rdcu.be/bX6Oq).\n\nThank you all for a great competition!\nEmma and the Human Protein Atlas team\n\n [1]: https://docs.google.com/forms/d/e/1FAIpQLScFRTpX57XyW0nn0K1NVfk8oCQEMd9wD_9ALwlMD3yN4cDoVA/viewform?usp=sf_link%20'survey'", "votes": 16 }, { "id": 454652, "postDate": "2019-01-11T23:28:37.647Z", "content": "Hi Emma,
\n\nThank you for hosting this wonderful and useful competition. My teammates and I are preparing a brief overview of our solution that we'll post in the discussion section, and we are very interested in helping out with the survey and potentially collaborating on the Nature Methods paper. However, I've taken a look at the survey, and seems like most of the questions presuppose a relatively simple single model (\"Please describe your solution using less than 100 words.\", \"How many parameters did you use for your model? Please answer with a number. \") We've built a fairly involved ensemble model, and we fear that it would be difficult to convey most of its complexity in answering such questions. If you would be kind enough to provide us with some guidance on this it would be most helpful.
\n\nBojan and the Wienerschnitzelgemeinschaft team.
", "rawMarkdown": "Hi Emma,\n\nThank you for hosting this wonderful and useful competition. My teammates and I are preparing a brief overview of our solution that we'll post in the discussion section, and we are very interested in helping out with the survey and potentially collaborating on the Nature Methods paper. However, I've taken a look at the survey, and seems like most of the questions presuppose a relatively simple single model (\"Please describe your solution using less than 100 words.\", \"How many parameters did you use for your model? Please answer with a number. \") We've built a fairly involved ensemble model, and we fear that it would be difficult to convey most of its complexity in answering such questions. If you would be kind enough to provide us with some guidance on this it would be most helpful.\n\nBojan and the Wienerschnitzelgemeinschaft team.", "votes": 5, "replies": [ { "id": 454657, "postDate": "2019-01-11T23:41:51.933Z", "content": "Hi Bojan,
\n\nThank you so much for your participation! We understand that this is a very limited space for explaining complex solutions. However, we need to keep the descriptions short so that we can review many submissions. The final selected teams will be given the opportunity to provide more details.
\n\nThat said, we still increased the word limit to 200. Hope this helps!
", "rawMarkdown": "Hi Bojan,\n\nThank you so much for your participation! We understand that this is a very limited space for explaining complex solutions. However, we need to keep the descriptions short so that we can review many submissions. The final selected teams will be given the opportunity to provide more details.\n\nThat said, we still increased the word limit to 200. Hope this helps!", "votes": 1 }, { "id": 454659, "postDate": "2019-01-11T23:46:51.553Z", "content": "Thank you, that's helpful. How about \"How many parameters did you use for your model? Please answer with a number. \"
\n\nDoes this mean how many parameters did we use for all of our submodels? Each one of them had many different parameters - number of layers, number of epochs, lr schedule, etc.
\n\nAnd the same question for the number of epochs - each one of our submodels was different.
", "rawMarkdown": "Thank you, that's helpful. How about \"How many parameters did you use for your model? Please answer with a number. \" \n\nDoes this mean how many parameters did we use for *all* of our submodels? Each one of them had many different parameters - number of layers, number of epochs, lr schedule, etc. \n\nAnd the same question for the number of epochs - each one of our submodels was different. " }, { "id": 455021, "postDate": "2019-01-12T19:05:47.030Z", "content": "We have now clarified the question to \"How many trainable parameters did you use for your model/network? If multiple models were used, please list the number of trainable parameters for each model. (e.g. 23034 + 537988 + 450104)\" and removed the question about epochs due to the complexity to give one answer.
\n\nThanks for asking!
", "rawMarkdown": "We have now clarified the question to \"How many trainable parameters did you use for your model/network? If multiple models were used, please list the number of trainable parameters for each model. (e.g. 23034 + 537988 + 450104)\" and removed the question about epochs due to the complexity to give one answer. \n\nThanks for asking!" }, { "id": 604514, "postDate": "2019-08-21T13:22:41.710Z", "content": "Hi Emma,\nThanks for holding this wonderful competition, now I'd like to use this dataset for research purpose, but I can't find the labels of the test set. Would you please tell me how to get access to the test set labels ? Thank you a lot, best wishes!
", "rawMarkdown": "Hi Emma,\nThanks for holding this wonderful competition, now I'd like to use this dataset for research purpose, but I can't find the labels of the test set. Would you please tell me how to get access to the test set labels ? Thank you a lot, best wishes!" }, { "id": 606401, "postDate": "2019-08-23T14:27:41.213Z", "content": "Please see our answer here:\nhttps://www.kaggle.com/c/human-protein-atlas-image-classification/discussion/77367#606400
", "rawMarkdown": "Please see our answer here:\nhttps://www.kaggle.com/c/human-protein-atlas-image-classification/discussion/77367#606400" } ] }, { "id": 460058, "postDate": "2019-01-22T22:46:59.640Z", "content": "Dear Organizers,
\n\nAre the results of your evaluation already out yet? Thank you!
\n\nSincerely,\nAlexander
", "rawMarkdown": "Dear Organizers, \n\nAre the results of your evaluation already out yet? Thank you!\n\nSincerely,\nAlexander", "votes": 1, "replies": [ { "id": 460125, "postDate": "2019-01-23T03:33:47.453Z", "content": "Dear Alexander,\nWe have made our selection of teams. Since we do not have access to contact information, Kaggle will reach out directly to these teams. If not today, I hope tomorrow.\nBest,\nEmma
", "rawMarkdown": "Dear Alexander,\nWe have made our selection of teams. Since we do not have access to contact information, Kaggle will reach out directly to these teams. If not today, I hope tomorrow.\nBest,\nEmma", "votes": 2 } ] }, { "id": 455302, "postDate": "2019-01-13T14:49:20.817Z", "content": "2.10 Did you scale the images before feeding them into your model? If yes, what was the scale factor? Please use a number to answer, e.g., 1 means the image was not scaled, and 0.5 means the image size was shrunk by half.\n\n
How should we answer this question, if we used three different scale factors?\nAlso, which original image size does this question imply?
", "rawMarkdown": "2.10 Did you scale the images before feeding them into your model? If yes, what was the scale factor? Please use a number to answer, e.g., 1 means the image was not scaled, and 0.5 means the image size was shrunk by half.\nHow should we answer this question, if we used three different scale factors?\nAlso, which original image size does this question imply?", "votes": 2, "replies": [ { "id": 455303, "postDate": "2019-01-13T14:51:58.040Z", "content": "
2.11 Did you crop the image into smaller patches before feeding them into your model? If yes, what patch size did you use (e.g., 128x128)?\n\n
Same question here.
", "rawMarkdown": "2.11 Did you crop the image into smaller patches before feeding them into your model? If yes, what patch size did you use (e.g., 128x128)?\nSame question here.", "votes": 1 }, { "id": 455405, "postDate": "2019-01-13T18:24:01.260Z", "content": "
Please answer with all the scale factors use (2.10) and all patch sizes (2.11)
", "rawMarkdown": "Please answer with all the scale factors use (2.10) and all patch sizes (2.11)", "votes": 2 }, { "id": 455407, "postDate": "2019-01-13T18:31:04.890Z", "content": "Thank you!\nAnd what initial size does the scale factor imply? I mean if we trained on 512x512, does it mean scale factor equal to 4?
", "rawMarkdown": "Thank you!\nAnd what initial size does the scale factor imply? I mean if we trained on 512x512, does it mean scale factor equal to 4?" }, { "id": 455434, "postDate": "2019-01-13T21:30:50.760Z", "content": "Correct!
", "rawMarkdown": "Correct!", "votes": 1 } ] }, { "id": 454699, "postDate": "2019-01-12T03:23:20.217Z", "content": "Hi Emma,
\n\nFor the question \"How many parameters did you use for your model? Please answer with a number.\" Could you clarify what parameters are you referring to? Should we report total parameters to be trained in the network or some higher level parameters like layers and epochs? If we used ensemble models, should we report the total number of parameters in all the model used to ensemble?
", "rawMarkdown": "Hi Emma,\n\nFor the question \"How many parameters did you use for your model? Please answer with a number.\" Could you clarify what parameters are you referring to? Should we report total parameters to be trained in the network or some higher level parameters like layers and epochs? If we used ensemble models, should we report the total number of parameters in all the model used to ensemble? ", "votes": 2, "replies": [ { "id": 455020, "postDate": "2019-01-12T19:04:33.327Z", "content": "Hi,
\n\nWe have now clarified the questions to be \"How many trainable parameters did you use for your model/network? If multiple models were used, please list the number of trainable parameters for each model. (e.g. 23034 + 537988 + 450104)\"
\n\nThanks for asking!
", "rawMarkdown": "Hi,\n\nWe have now clarified the questions to be \"How many trainable parameters did you use for your model/network? If multiple models were used, please list the number of trainable parameters for each model. (e.g. 23034 + 537988 + 450104)\"\n\nThanks for asking!" } ] }, { "id": 599944, "postDate": "2019-08-15T13:28:44.300Z", "content": "nice
", "rawMarkdown": "nice" }, { "id": 571697, "postDate": "2019-07-10T01:39:57.380Z", "content": "Dear Emma,
\n\nSorry to disturb you so long after this competition ended. I am a new programmer and I am doing a 29 subcellular classification problem, this competition has been of great help to me in finding tricks and methods of classification problem.\nI wanted to ask if there exist the 'corrected' version of the sample_submission now that the competition is finished so that we can compare our predictions with the solution to know how far did we go. This would greatly increase the size of this awesome open source dataset.
\n\nBest Regards,\nYanis Yang
", "rawMarkdown": "Dear Emma,\n\nSorry to disturb you so long after this competition ended. I am a new programmer and I am doing a 29 subcellular classification problem, this competition has been of great help to me in finding tricks and methods of classification problem.\nI wanted to ask if there exist the 'corrected' version of the sample_submission now that the competition is finished so that we can compare our predictions with the solution to know how far did we go. This would greatly increase the size of this awesome open source dataset.\n\nBest Regards,\nYanis Yang", "replies": [ { "id": 606400, "postDate": "2019-08-23T14:26:59.440Z", "content": "Hi!
\n\nWe are at this time not planning to release the correct answers for the test set. We want to keep it possible to try this challenge without knowing the answers.
", "rawMarkdown": "Hi!\n\nWe are at this time not planning to release the correct answers for the test set. We want to keep it possible to try this challenge without knowing the answers." } ] }, { "id": 491944, "postDate": "2019-03-16T12:29:06.973Z", "content": "So, is the analysis article out? If it is, could you please share the link here? That would be very helpful, thanks.
", "rawMarkdown": "So, is the analysis article out? If it is, could you please share the link here? That would be very helpful, thanks.", "replies": [ { "id": 524926, "postDate": "2019-04-29T19:08:05.780Z", "content": "The article is not yet out. We will announce it here when it is out, and also post the link to the paper.
\n\nBest,\nEmma
", "rawMarkdown": "The article is not yet out. We will announce it here when it is out, and also post the link to the paper.\n\nBest,\nEmma", "votes": 1 } ] }, { "id": 457040, "postDate": "2019-01-16T21:41:49.957Z", "content": "Hi Emma,
\n\nIt took our time a while to gather and distill all the relevant information for the survey, which we finally did today, only to discover that there is also the 3rd page that requires even more information in more detail! This last page was not accessible before we were able to fill out the 2nd page, and thus we only found out about it now. We don't feel it would be feasible to do justice to this last page in the remaining amount of time. We would be happy to share more information later, but don't want to be adversely penalized right now.
\n\nBest,
\n\nBojan
", "rawMarkdown": "Hi Emma,\n\nIt took our time a while to gather and distill all the relevant information for the survey, which we finally did today, only to discover that there is also the 3rd page that requires even more information in more detail! This last page was not accessible before we were able to fill out the 2nd page, and thus we only found out about it now. We don't feel it would be feasible to do justice to this last page in the remaining amount of time. We would be happy to share more information later, but don't want to be adversely penalized right now. \n\nBest,\n\nBojan", "replies": [ { "id": 457081, "postDate": "2019-01-16T23:05:54.447Z", "content": "Thanks for this notification! No worries, the third page is optional and we will remember that you are willing to share this information.
", "rawMarkdown": "Thanks for this notification! No worries, the third page is optional and we will remember that you are willing to share this information. ", "votes": 1 }, { "id": 457540, "postDate": "2019-01-17T16:54:03.967Z", "content": "Thank you, we managed to submit the 3rd page anyways, but we can certainly improve on it and provide more details if needed!
", "rawMarkdown": "Thank you, we managed to submit the 3rd page anyways, but we can certainly improve on it and provide more details if needed!" } ] }, { "id": 454665, "postDate": "2019-01-12T00:16:00.770Z", "content": "Hi, Can we get the labels of the test set for further research on your dataset? Thanks.
", "rawMarkdown": "Hi, Can we get the labels of the test set for further research on your dataset? Thanks." }, { "id": 456063, "postDate": "2019-01-15T03:44:38.270Z", "rawMarkdown": "", "isDeleted": true, "replies": [ { "id": 456065, "postDate": "2019-01-15T03:57:21.487Z", "content": "The deadline is Midnight of January 16th Pacific time.
", "rawMarkdown": "The deadline is Midnight of January 16th Pacific time.", "votes": 1 }, { "id": 456323, "postDate": "2019-01-15T15:09:05.390Z", "content": "Thanks!
", "rawMarkdown": "Thanks!" } ] }, { "id": 454712, "postDate": "2019-01-12T04:19:55.760Z", "rawMarkdown": "", "isDeleted": true } ], "comments": [ { "id": 454652, "author_name": "Bojan Tunguz", "author_url": "", "post_date": "2019-01-11T23:28:37.647000", "content": "Hi Emma,
\n\nThank you for hosting this wonderful and useful competition. My teammates and I are preparing a brief overview of our solution that we'll post in the discussion section, and we are very interested in helping out with the survey and potentially collaborating on the Nature Methods paper. However, I've taken a look at the survey, and seems like most of the questions presuppose a relatively simple single model (\"Please describe your solution using less than 100 words.\", \"How many parameters did you use for your model? Please answer with a number. \") We've built a fairly involved ensemble model, and we fear that it would be difficult to convey most of its complexity in answering such questions. If you would be kind enough to provide us with some guidance on this it would be most helpful.
\n\nBojan and the Wienerschnitzelgemeinschaft team.
", "votes": 5, "replies": [ { "id": 454657, "author_name": "Emma Lundberg", "author_url": "", "post_date": "2019-01-11T23:41:51.933000", "content": "Hi Bojan,
\n\nThank you so much for your participation! We understand that this is a very limited space for explaining complex solutions. However, we need to keep the descriptions short so that we can review many submissions. The final selected teams will be given the opportunity to provide more details.
\n\nThat said, we still increased the word limit to 200. Hope this helps!
", "votes": 1, "replies": [] }, { "id": 454659, "author_name": "Bojan Tunguz", "author_url": "", "post_date": "2019-01-11T23:46:51.553000", "content": "Thank you, that's helpful. How about \"How many parameters did you use for your model? Please answer with a number. \"
\n\nDoes this mean how many parameters did we use for all of our submodels? Each one of them had many different parameters - number of layers, number of epochs, lr schedule, etc.
\n\nAnd the same question for the number of epochs - each one of our submodels was different.
", "votes": 0, "replies": [] }, { "id": 455021, "author_name": "Emma Lundberg", "author_url": "", "post_date": "2019-01-12T19:05:47.030000", "content": "We have now clarified the question to \"How many trainable parameters did you use for your model/network? If multiple models were used, please list the number of trainable parameters for each model. (e.g. 23034 + 537988 + 450104)\" and removed the question about epochs due to the complexity to give one answer.
\n\nThanks for asking!
", "votes": 0, "replies": [] }, { "id": 604514, "author_name": "Caiqing Jian", "author_url": "", "post_date": "2019-08-21T13:22:41.710000", "content": "Hi Emma,\nThanks for holding this wonderful competition, now I'd like to use this dataset for research purpose, but I can't find the labels of the test set. Would you please tell me how to get access to the test set labels ? Thank you a lot, best wishes!
", "votes": 0, "replies": [] }, { "id": 606401, "author_name": "Martin Hjelmare", "author_url": "", "post_date": "2019-08-23T14:27:41.213000", "content": "Please see our answer here:\nhttps://www.kaggle.com/c/human-protein-atlas-image-classification/discussion/77367#606400
", "votes": 0, "replies": [] } ] }, { "id": 460058, "author_name": "Peiyuan Liao", "author_url": "", "post_date": "2019-01-22T22:46:59.640000", "content": "Dear Organizers,
\n\nAre the results of your evaluation already out yet? Thank you!
\n\nSincerely,\nAlexander
", "votes": 1, "replies": [ { "id": 460125, "author_name": "Emma Lundberg", "author_url": "", "post_date": "2019-01-23T03:33:47.453000", "content": "Dear Alexander,\nWe have made our selection of teams. Since we do not have access to contact information, Kaggle will reach out directly to these teams. If not today, I hope tomorrow.\nBest,\nEmma
", "votes": 2, "replies": [] } ] }, { "id": 455302, "author_name": "Dmytro Panchenko", "author_url": "", "post_date": "2019-01-13T14:49:20.817000", "content": "2.10 Did you scale the images before feeding them into your model? If yes, what was the scale factor? Please use a number to answer, e.g., 1 means the image was not scaled, and 0.5 means the image size was shrunk by half.\n\n
How should we answer this question, if we used three different scale factors?\nAlso, which original image size does this question imply?
", "votes": 2, "replies": [ { "id": 455303, "author_name": "Dmytro Panchenko", "author_url": "", "post_date": "2019-01-13T14:51:58.040000", "content": "2.11 Did you crop the image into smaller patches before feeding them into your model? If yes, what patch size did you use (e.g., 128x128)?\n\n
Same question here.
", "votes": 1, "replies": [] }, { "id": 455405, "author_name": "Emma Lundberg", "author_url": "", "post_date": "2019-01-13T18:24:01.260000", "content": "Please answer with all the scale factors use (2.10) and all patch sizes (2.11)
", "votes": 2, "replies": [] }, { "id": 455407, "author_name": "Dmytro Panchenko", "author_url": "", "post_date": "2019-01-13T18:31:04.890000", "content": "Thank you!\nAnd what initial size does the scale factor imply? I mean if we trained on 512x512, does it mean scale factor equal to 4?
", "votes": 0, "replies": [] }, { "id": 455434, "author_name": "Emma Lundberg", "author_url": "", "post_date": "2019-01-13T21:30:50.760000", "content": "Correct!
", "votes": 1, "replies": [] } ] }, { "id": 454699, "author_name": "Xuan Cao", "author_url": "", "post_date": "2019-01-12T03:23:20.217000", "content": "Hi Emma,
\n\nFor the question \"How many parameters did you use for your model? Please answer with a number.\" Could you clarify what parameters are you referring to? Should we report total parameters to be trained in the network or some higher level parameters like layers and epochs? If we used ensemble models, should we report the total number of parameters in all the model used to ensemble?
", "votes": 2, "replies": [ { "id": 455020, "author_name": "Emma Lundberg", "author_url": "", "post_date": "2019-01-12T19:04:33.327000", "content": "Hi,
\n\nWe have now clarified the questions to be \"How many trainable parameters did you use for your model/network? If multiple models were used, please list the number of trainable parameters for each model. (e.g. 23034 + 537988 + 450104)\"
\n\nThanks for asking!
", "votes": 0, "replies": [] } ] }, { "id": 599944, "author_name": "James Peng", "author_url": "", "post_date": "2019-08-15T13:28:44.300000", "content": "nice
", "votes": 0, "replies": [] }, { "id": 571697, "author_name": "Yanis Yang", "author_url": "", "post_date": "2019-07-10T01:39:57.380000", "content": "Dear Emma,
\n\nSorry to disturb you so long after this competition ended. I am a new programmer and I am doing a 29 subcellular classification problem, this competition has been of great help to me in finding tricks and methods of classification problem.\nI wanted to ask if there exist the 'corrected' version of the sample_submission now that the competition is finished so that we can compare our predictions with the solution to know how far did we go. This would greatly increase the size of this awesome open source dataset.
\n\nBest Regards,\nYanis Yang
", "votes": 0, "replies": [ { "id": 606400, "author_name": "Martin Hjelmare", "author_url": "", "post_date": "2019-08-23T14:26:59.440000", "content": "Hi!
\n\nWe are at this time not planning to release the correct answers for the test set. We want to keep it possible to try this challenge without knowing the answers.
", "votes": 0, "replies": [] } ] }, { "id": 491944, "author_name": "Avijit Mitra", "author_url": "", "post_date": "2019-03-16T12:29:06.973000", "content": "So, is the analysis article out? If it is, could you please share the link here? That would be very helpful, thanks.
", "votes": 0, "replies": [ { "id": 524926, "author_name": "Emma Lundberg", "author_url": "", "post_date": "2019-04-29T19:08:05.780000", "content": "The article is not yet out. We will announce it here when it is out, and also post the link to the paper.
\n\nBest,\nEmma
", "votes": 1, "replies": [] } ] }, { "id": 457040, "author_name": "Bojan Tunguz", "author_url": "", "post_date": "2019-01-16T21:41:49.957000", "content": "Hi Emma,
\n\nIt took our time a while to gather and distill all the relevant information for the survey, which we finally did today, only to discover that there is also the 3rd page that requires even more information in more detail! This last page was not accessible before we were able to fill out the 2nd page, and thus we only found out about it now. We don't feel it would be feasible to do justice to this last page in the remaining amount of time. We would be happy to share more information later, but don't want to be adversely penalized right now.
\n\nBest,
\n\nBojan
", "votes": 0, "replies": [ { "id": 457081, "author_name": "Emma Lundberg", "author_url": "", "post_date": "2019-01-16T23:05:54.447000", "content": "Thanks for this notification! No worries, the third page is optional and we will remember that you are willing to share this information.
", "votes": 1, "replies": [] }, { "id": 457540, "author_name": "Bojan Tunguz", "author_url": "", "post_date": "2019-01-17T16:54:03.967000", "content": "Thank you, we managed to submit the 3rd page anyways, but we can certainly improve on it and provide more details if needed!
", "votes": 0, "replies": [] } ] }, { "id": 454665, "author_name": "Erfun", "author_url": "", "post_date": "2019-01-12T00:16:00.770000", "content": "Hi, Can we get the labels of the test set for further research on your dataset? Thanks.
", "votes": 0, "replies": [] }, { "id": 456063, "author_name": "", "author_url": "", "post_date": "2019-01-15T03:44:38.270000", "content": "", "votes": 0, "replies": [ { "id": 456065, "author_name": "Emma Lundberg", "author_url": "", "post_date": "2019-01-15T03:57:21.487000", "content": "The deadline is Midnight of January 16th Pacific time.
", "votes": 1, "replies": [] }, { "id": 456323, "author_name": "Peiyuan Liao", "author_url": "", "post_date": "2019-01-15T15:09:05.390000", "content": "Thanks!
", "votes": 0, "replies": [] } ] }, { "id": 454712, "author_name": "", "author_url": "", "post_date": "2019-01-12T04:19:55.760000", "content": "", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "454645": "Thanks to all of you that participated in this challenge! \n\nWe are very happy with the results so far and look forward to diving into the details of your solutions. We are convinced of that they will contribute to an improvement of data in the Human Protein Atlas database and help to accelerate basic biological science.\n\nAs announced in the challenge description, we aim to write an Analysis article for Nature Methods where we compare approaches and discuss the outcome of the challenge. For this purpose, all teams are hereby invited to submit details about their solutions through this [survey][1]. Based on this survey, a maximum of 10 teams will be selected for participation as co-authors. It will be required that the solutions from the selected teams are uploaded to the team page.\n\nThe deadline to fill out the survey is by the midnight of January 16th. We will notify the selected teams on January 22nd.\n\nThank you all again!\nEmma and the Human Protein Atlas team\n\n-------------\nUpdate Dec 1 2019:\nThe resulting article \"[Analysis of the Human Protein Atlas Image Classification Competition](https://www.nature.com/articles/s41592-019-0658-6)\" is now published in Nature Methods. We would like to thank all the contributing teams, it has been a pleasure working with you! We now have models that perform 20% better than previous models for protein subcellular pattern classification, a great step forward for cell biology research. The top ranking models can also be found in our [model zoo](https://modelzoo.cellprofiling.org/#/).\n\nA free pdf copy of the paper can be found [here](https://rdcu.be/bX6Oq).\n\nThank you all for a great competition!\nEmma and the Human Protein Atlas team\n\n [1]: https://docs.google.com/forms/d/e/1FAIpQLScFRTpX57XyW0nn0K1NVfk8oCQEMd9wD_9ALwlMD3yN4cDoVA/viewform?usp=sf_link%20'survey'", "454652": "Hi Emma,\n\nThank you for hosting this wonderful and useful competition. My teammates and I are preparing a brief overview of our solution that we'll post in the discussion section, and we are very interested in helping out with the survey and potentially collaborating on the Nature Methods paper. However, I've taken a look at the survey, and seems like most of the questions presuppose a relatively simple single model (\"Please describe your solution using less than 100 words.\", \"How many parameters did you use for your model? Please answer with a number. \") We've built a fairly involved ensemble model, and we fear that it would be difficult to convey most of its complexity in answering such questions. If you would be kind enough to provide us with some guidance on this it would be most helpful.\n\nBojan and the Wienerschnitzelgemeinschaft team.", "460058": "Dear Organizers, \n\nAre the results of your evaluation already out yet? Thank you!\n\nSincerely,\nAlexander", "455302": "2.10 Did you scale the images before feeding them into your model? If yes, what was the scale factor? Please use a number to answer, e.g., 1 means the image was not scaled, and 0.5 means the image size was shrunk by half.\nHow should we answer this question, if we used three different scale factors?\nAlso, which original image size does this question imply?", "454699": "Hi Emma,\n\nFor the question \"How many parameters did you use for your model? Please answer with a number.\" Could you clarify what parameters are you referring to? Should we report total parameters to be trained in the network or some higher level parameters like layers and epochs? If we used ensemble models, should we report the total number of parameters in all the model used to ensemble? ", "599944": "nice", "571697": "Dear Emma,\n\nSorry to disturb you so long after this competition ended. I am a new programmer and I am doing a 29 subcellular classification problem, this competition has been of great help to me in finding tricks and methods of classification problem.\nI wanted to ask if there exist the 'corrected' version of the sample_submission now that the competition is finished so that we can compare our predictions with the solution to know how far did we go. This would greatly increase the size of this awesome open source dataset.\n\nBest Regards,\nYanis Yang", "491944": "So, is the analysis article out? If it is, could you please share the link here? That would be very helpful, thanks.", "457040": "Hi Emma,\n\nIt took our time a while to gather and distill all the relevant information for the survey, which we finally did today, only to discover that there is also the 3rd page that requires even more information in more detail! This last page was not accessible before we were able to fill out the 2nd page, and thus we only found out about it now. We don't feel it would be feasible to do justice to this last page in the remaining amount of time. We would be happy to share more information later, but don't want to be adversely penalized right now. \n\nBest,\n\nBojan", "454665": "Hi, Can we get the labels of the test set for further research on your dataset? Thanks.", "456063": "", "454712": "" } }