Dear Kagglers,
\n\nAs you may have noticed, we have organized this competition as part of our research projets with the will to challenge our academic community with kagglers who have very little or even no knowledge in steganalysis but tremendous experience in deep learning and its practical use for solving image classification problems efficiently.
\n\nSo far, this competition has been a great success for all of us in many aspects:\n- The competition ragged\n- Several new ideas have been thrown for our community (use of EfficientNet for instance)\n- Several breakthrough have been achieved (just like we though it was impossible to merge altogether difference JPEG quality factors, embedding schemes and color channels)\nAnd several ideas seems to be confirmed with all you numerous experiments:\n- Using DCT coefficients does not seem to be a clever move, use pixels values instead\n- Using Deep nets designed for natural image classification is a good starting point\n- Multi-class detection eventually pays off\n....
\n\nI would like to call for help in order to get as much feedback as possible for our community.\nThere I kindly ask you two things\n(1) Can you please send me a message with your feedback, what you learn, how you addressed the problem and what you would have tried with more time (and possibly your code that allows getting your best results). \nThe goal is to provide our community with as much insights as possible throughout this informal pool survey. We will not keep those information to publish some research paper for our own fame (the organizers) but will share it openly with the academic community with the goal that it can shape our future research directions.\n(2) For the \"best\" competitors, consider submitting a paper to IEEE WIFS conference. The paper will be evaluated based on both the final score but also on the scientific contribution and analysis (a paper that simply describe a Deep Net architecture has little added values). I am especially thinking about people from research community that get excellent scores (ABBA team / Haneol Jang / Guanshuo Xu / Han Chen / etc. ...) but anyone else who believe have found interesting insights, ideas and findings that can be of general interest for our community, your submission will be more than welcome (note that a total 3k$ travel grant will be shared among accepted papers which can help but can hardly cover all travel costs)
\n\nAgain feel free to message me (private message via kaggle or email directly at remi.cogranne@utt.fr) any feedback, findings, comment and details about codes will be more than welcome.
", "messages": [ { "id": 921899, "postDate": "2020-07-09T16:30:46.750Z", "content": "Dear Kagglers,
\n\nAs you may have noticed, we have organized this competition as part of our research projets with the will to challenge our academic community with kagglers who have very little or even no knowledge in steganalysis but tremendous experience in deep learning and its practical use for solving image classification problems efficiently.
\n\nSo far, this competition has been a great success for all of us in many aspects:\n- The competition ragged\n- Several new ideas have been thrown for our community (use of EfficientNet for instance)\n- Several breakthrough have been achieved (just like we though it was impossible to merge altogether difference JPEG quality factors, embedding schemes and color channels)\nAnd several ideas seems to be confirmed with all you numerous experiments:\n- Using DCT coefficients does not seem to be a clever move, use pixels values instead\n- Using Deep nets designed for natural image classification is a good starting point\n- Multi-class detection eventually pays off\n....
\n\nI would like to call for help in order to get as much feedback as possible for our community.\nThere I kindly ask you two things\n(1) Can you please send me a message with your feedback, what you learn, how you addressed the problem and what you would have tried with more time (and possibly your code that allows getting your best results). \nThe goal is to provide our community with as much insights as possible throughout this informal pool survey. We will not keep those information to publish some research paper for our own fame (the organizers) but will share it openly with the academic community with the goal that it can shape our future research directions.\n(2) For the \"best\" competitors, consider submitting a paper to IEEE WIFS conference. The paper will be evaluated based on both the final score but also on the scientific contribution and analysis (a paper that simply describe a Deep Net architecture has little added values). I am especially thinking about people from research community that get excellent scores (ABBA team / Haneol Jang / Guanshuo Xu / Han Chen / etc. ...) but anyone else who believe have found interesting insights, ideas and findings that can be of general interest for our community, your submission will be more than welcome (note that a total 3k$ travel grant will be shared among accepted papers which can help but can hardly cover all travel costs)
\n\nAgain feel free to message me (private message via kaggle or email directly at remi.cogranne@utt.fr) any feedback, findings, comment and details about codes will be more than welcome.
", "rawMarkdown": "Dear Kagglers,\n\nAs you may have noticed, we have organized this competition as part of our research projets with the will to challenge our academic community with kagglers who have very little or even no knowledge in steganalysis but tremendous experience in deep learning and its practical use for solving image classification problems efficiently.\n\nSo far, this competition has been a great success for all of us in many aspects:\n- The competition ragged\n- Several new ideas have been thrown for our community (use of EfficientNet for instance)\n- Several breakthrough have been achieved (just like we though it was impossible to merge altogether difference JPEG quality factors, embedding schemes and color channels)\nAnd several ideas seems to be confirmed with all you numerous experiments:\n- Using DCT coefficients does not seem to be a clever move, use pixels values instead\n- Using Deep nets designed for natural image classification is a good starting point\n- Multi-class detection eventually pays off\n....\n\nI would like to call for help in order to get as much feedback as possible for our community.\nThere I kindly ask you two things\n(1) Can you please send me a message with your feedback, what you learn, how you addressed the problem and what you would have tried with more time (and possibly your code that allows getting your best results). \nThe goal is to provide our community with as much insights as possible throughout this informal pool survey. We will not keep those information to publish some research paper for our own fame (the organizers) but will share it openly with the academic community with the goal that it can shape our future research directions.\n(2) For the \"best\" competitors, consider submitting a paper to IEEE WIFS conference. The paper will be evaluated based on both the final score but also on the scientific contribution and analysis (a paper that simply describe a Deep Net architecture has little added values). I am especially thinking about people from research community that get excellent scores (ABBA team / Haneol Jang / Guanshuo Xu / Han Chen / etc. ...) but anyone else who believe have found interesting insights, ideas and findings that can be of general interest for our community, your submission will be more than welcome (note that a total 3k$ travel grant will be shared among accepted papers which can help but can hardly cover all travel costs)\n\nAgain feel free to message me (private message via kaggle or email directly at remi.cogranne@utt.fr) any feedback, findings, comment and details about codes will be more than welcome.", "votes": 17 }, { "id": 925206, "postDate": "2020-07-11T22:59:30.343Z", "content": "Just wish there was more time between the end of this competition and WIFS deadline. You will need to lower the bar for the acceptance, Remi. Most teams will not be able to go much beyond basically summarizing their experience during this competition (which may be still valuable). A question to you: Will you be able to access the scores on both LBs and put something meaningful about the competition itself for WIFS? I assume you are planning this, right?
", "rawMarkdown": "Just wish there was more time between the end of this competition and WIFS deadline. You will need to lower the bar for the acceptance, Remi. Most teams will not be able to go much beyond basically summarizing their experience during this competition (which may be still valuable). A question to you: Will you be able to access the scores on both LBs and put something meaningful about the competition itself for WIFS? I assume you are planning this, right?", "votes": 3 }, { "id": 927232, "postDate": "2020-07-13T09:25:02.793Z", "content": "i have a feeling that the sometimes deep network is just detecting texture rather than just hidden message. I would suggest that for future steganalysis competition, the test images should contain both the cover image and the stego image. We should also measure and compare the score difference the pair.
", "rawMarkdown": "i have a feeling that the sometimes deep network is just detecting texture rather than just hidden message. I would suggest that for future steganalysis competition, the test images should contain both the cover image and the stego image. We should also measure and compare the score difference the pair.\n\n", "votes": 1 }, { "id": 921990, "postDate": "2020-07-09T17:56:22.327Z", "content": "Am I the only one who after reading this post is thinking: \"did the competition end already\"? 😆
\n\nRe: sharing \"how you addressed the problem\" is not typically considered good etiquette in Kaggle to share key insights / code close to competition end.
", "rawMarkdown": "Am I the only one who after reading this post is thinking: \"did the competition end already\"? 😆\n\nRe: sharing \"how you addressed the problem\" is not typically considered good etiquette in Kaggle to share key insights / code close to competition end.", "votes": 1, "replies": [ { "id": 922143, "postDate": "2020-07-09T20:23:38.080Z", "content": "As long as it's DM'd to the competition host, shouldn't be a problem ;-).
", "rawMarkdown": "As long as it's DM'd to the competition host, shouldn't be a problem ;-).", "votes": 1 } ] }, { "id": 933848, "postDate": "2020-07-18T04:46:56.747Z", "content": "Hello there,
\nA short message to let you know a good news regarding the follow-up of the competition.
\nThe IEEE WIFS paper submission deadline has been \"unofficially\" extended by 10 days (to be accurate, until August 9th 23:59 AoE).
\nBy unofficially, I mean it will be a two step process ; if you want to submit a paper you have to do it before Aug 1st. However, this can be a draft that can be revised by August 10th.
\nWhat do we expect from you contributions ?
\nIn those paper we expect not you to write a dairy about how the challenge was addressed, but why you eventually ended up with your proposal, what did you learn and how, what you would recommend for our community (and fact on which it is based)
On my hand, I do not have a total access to submissions. Not even sure I can get the raw submissions you provided. This is why I am asking to kindly let me know any useful information (code, submitted data, architecture description, which net you tried and results you got, etc. …. ). This would be very usefull in order to draw meaningful conclusion about this challenge.
\nThanks in advance for you help, we remain available to help you out if you have more question about these aspect of article writing and/or information to provide me
", "rawMarkdown": "Hello there,\nA short message to let you know a good news regarding the follow-up of the competition.\nThe IEEE WIFS paper submission deadline has been \"unofficially\" extended by 10 days (to be accurate, until August 9th 23:59 AoE).\nBy unofficially, I mean it will be a two step process ; if you want to submit a paper you have to do it before Aug 1st. However, this can be a draft that can be revised by August 10th.\nWhat do we expect from you contributions ? \nIn those paper we expect not you to write a dairy about how the challenge was addressed, but why you eventually ended up with your proposal, what did you learn and how, what you would recommend for our community (and fact on which it is based) \n\nOn my hand, I do not have a total access to submissions. Not even sure I can get the raw submissions you provided. This is why I am asking to kindly let me know any useful information (code, submitted data, architecture description, which net you tried and results you got, etc. .... ). This would be very usefull in order to draw meaningful conclusion about this challenge.\n\nThanks in advance for you help, we remain available to help you out if you have more question about these aspect of article writing and/or information to provide me", "votes": 2 }, { "id": 925228, "postDate": "2020-07-11T23:51:33.047Z", "content": "Just a thought, most of us who are working hard to make one last push for this competition, taking on the additional tasks of writing an academic paper sounds quite taxing.
\n\nI guess the one feedback we could share is - for a workshop regarding a competition, making paper submission deadline so soon after finishing date of the competition doesn't seem to be a good idea... Bear in mind that potential prize winners would also need to be writing up for the winning solution documentation. and people do suffer burn out for working flat-out on these competitions on top of their day job. Would really be nice if this is taken into consideration.
\n\nPerhaps you can work with Kaggle to come up with a winning submission documentation template such that the documentation can also be treated as a potential paper submission?
\n\nAlso due to the small size public LB, it is perhaps hard to decide who might qualify as \"best\" competitors - private LB shakeup is a regular phenomenon on Kaggle. I would suggest, if @remicogranne also have access to people private's LB now, you might want to keep an eye out from insights from those team who are high on the private LB score. You must not approach them directly though because it will give unfair advantage :)
", "rawMarkdown": "Just a thought, most of us who are working hard to make one last push for this competition, taking on the additional tasks of writing an academic paper sounds quite taxing. \n\nI guess the one feedback we could share is - for a workshop regarding a competition, making paper submission deadline so soon after finishing date of the competition doesn't seem to be a good idea... Bear in mind that potential prize winners would also need to be writing up for the winning solution documentation. and people do suffer burn out for working flat-out on these competitions on top of their day job. Would really be nice if this is taken into consideration. \n\nPerhaps you can work with Kaggle to come up with a winning submission documentation template such that the documentation can also be treated as a potential paper submission? \n\nAlso due to the small size public LB, it is perhaps hard to decide who might qualify as \"best\" competitors - private LB shakeup is a regular phenomenon on Kaggle. I would suggest, if @remicogranne also have access to people private's LB now, you might want to keep an eye out from insights from those team who are high on the private LB score. You must not approach them directly though because it will give unfair advantage :) \n\n\n\n\n", "votes": 2, "replies": [ { "id": 927422, "postDate": "2020-07-13T11:23:16.503Z", "content": "insights from those team who are high on the private LB score
I would go further and say private LB of 4000 images is junk compared with highest reported CV scores across 300,000 images. Also, Private LB depends on the 2 chosen scores, whereas the organiser might be interested in solutions behind actual (non-selected) highest scoring.
", "rawMarkdown": "` insights from those team who are high on the private LB score`\n\nI would go further and say private LB of 4000 images is junk compared with highest reported CV scores across 300,000 images. Also, Private LB depends on the 2 chosen scores, whereas the organiser might be interested in solutions behind actual (non-selected) highest scoring.\n", "votes": 1 } ] }, { "id": 922019, "postDate": "2020-07-09T18:28:07.097Z", "content": "Hello Andres,\nThanks for your message, I just realize that I may have forgotten to clarify something here ...\nWhile the competition did not end yet, (scientific) conference have strict deadlines and the one for IEEE WIFS, hosting a special session on this kaggle competition, implies that paper are due latest July 31st. Time to write a paper (and get + analyze first feedbacks) time will quickly run out.\nTherefore, I would appreciate to get feedback, codes, ideas, insights, findings as soon as possible and I insist that those who want to submit a paper should already consider the writing especially if not used to this exercise.
", "rawMarkdown": "Hello Andres,\nThanks for your message, I just realize that I may have forgotten to clarify something here ...\nWhile the competition did not end yet, (scientific) conference have strict deadlines and the one for IEEE WIFS, hosting a special session on this kaggle competition, implies that paper are due latest July 31st. Time to write a paper (and get + analyze first feedbacks) time will quickly run out.\nTherefore, I would appreciate to get feedback, codes, ideas, insights, findings as soon as possible and I insist that those who want to submit a paper should already consider the writing especially if not used to this exercise.", "votes": 2 }, { "id": 926843, "postDate": "2020-07-13T03:28:47.250Z", "content": "I fully agree with Jessica Fridrich and Yifan Xie. We need more time to prepare the paper after this competition. That way, we can summarize more meaningful analysis and results.
", "rawMarkdown": "I fully agree with Jessica Fridrich and Yifan Xie. We need more time to prepare the paper after this competition. That way, we can summarize more meaningful analysis and results." }, { "id": 933840, "postDate": "2020-07-18T04:37:45.940Z", "rawMarkdown": "", "isDeleted": true } ], "comments": [ { "id": 925206, "author_name": "Jessica Fridrich", "author_url": "", "post_date": "2020-07-11T22:59:30.343000", "content": "Just wish there was more time between the end of this competition and WIFS deadline. You will need to lower the bar for the acceptance, Remi. Most teams will not be able to go much beyond basically summarizing their experience during this competition (which may be still valuable). A question to you: Will you be able to access the scores on both LBs and put something meaningful about the competition itself for WIFS? I assume you are planning this, right?
", "votes": 3, "replies": [] }, { "id": 927232, "author_name": "hengck23", "author_url": "", "post_date": "2020-07-13T09:25:02.793000", "content": "i have a feeling that the sometimes deep network is just detecting texture rather than just hidden message. I would suggest that for future steganalysis competition, the test images should contain both the cover image and the stego image. We should also measure and compare the score difference the pair.
", "votes": 1, "replies": [] }, { "id": 921990, "author_name": "Andrés Miguel Torrubia Sáez", "author_url": "", "post_date": "2020-07-09T17:56:22.327000", "content": "Am I the only one who after reading this post is thinking: \"did the competition end already\"? 😆
\n\nRe: sharing \"how you addressed the problem\" is not typically considered good etiquette in Kaggle to share key insights / code close to competition end.
", "votes": 1, "replies": [ { "id": 922143, "author_name": "عثمان", "author_url": "", "post_date": "2020-07-09T20:23:38.080000", "content": "As long as it's DM'd to the competition host, shouldn't be a problem ;-).
", "votes": 1, "replies": [] } ] }, { "id": 933848, "author_name": "Rémi Cogranne", "author_url": "", "post_date": "2020-07-18T04:46:56.747000", "content": "Hello there,
\nA short message to let you know a good news regarding the follow-up of the competition.
\nThe IEEE WIFS paper submission deadline has been \"unofficially\" extended by 10 days (to be accurate, until August 9th 23:59 AoE).
\nBy unofficially, I mean it will be a two step process ; if you want to submit a paper you have to do it before Aug 1st. However, this can be a draft that can be revised by August 10th.
\nWhat do we expect from you contributions ?
\nIn those paper we expect not you to write a dairy about how the challenge was addressed, but why you eventually ended up with your proposal, what did you learn and how, what you would recommend for our community (and fact on which it is based)
On my hand, I do not have a total access to submissions. Not even sure I can get the raw submissions you provided. This is why I am asking to kindly let me know any useful information (code, submitted data, architecture description, which net you tried and results you got, etc. …. ). This would be very usefull in order to draw meaningful conclusion about this challenge.
\nThanks in advance for you help, we remain available to help you out if you have more question about these aspect of article writing and/or information to provide me
", "votes": 2, "replies": [] }, { "id": 925228, "author_name": "Yifan Xie", "author_url": "", "post_date": "2020-07-11T23:51:33.047000", "content": "Just a thought, most of us who are working hard to make one last push for this competition, taking on the additional tasks of writing an academic paper sounds quite taxing.
\n\nI guess the one feedback we could share is - for a workshop regarding a competition, making paper submission deadline so soon after finishing date of the competition doesn't seem to be a good idea... Bear in mind that potential prize winners would also need to be writing up for the winning solution documentation. and people do suffer burn out for working flat-out on these competitions on top of their day job. Would really be nice if this is taken into consideration.
\n\nPerhaps you can work with Kaggle to come up with a winning submission documentation template such that the documentation can also be treated as a potential paper submission?
\n\nAlso due to the small size public LB, it is perhaps hard to decide who might qualify as \"best\" competitors - private LB shakeup is a regular phenomenon on Kaggle. I would suggest, if @remicogranne also have access to people private's LB now, you might want to keep an eye out from insights from those team who are high on the private LB score. You must not approach them directly though because it will give unfair advantage :)
", "votes": 2, "replies": [ { "id": 927422, "author_name": "robga", "author_url": "", "post_date": "2020-07-13T11:23:16.503000", "content": "insights from those team who are high on the private LB score
I would go further and say private LB of 4000 images is junk compared with highest reported CV scores across 300,000 images. Also, Private LB depends on the 2 chosen scores, whereas the organiser might be interested in solutions behind actual (non-selected) highest scoring.
", "votes": 1, "replies": [] } ] }, { "id": 922019, "author_name": "Rémi Cogranne", "author_url": "", "post_date": "2020-07-09T18:28:07.097000", "content": "Hello Andres,\nThanks for your message, I just realize that I may have forgotten to clarify something here ...\nWhile the competition did not end yet, (scientific) conference have strict deadlines and the one for IEEE WIFS, hosting a special session on this kaggle competition, implies that paper are due latest July 31st. Time to write a paper (and get + analyze first feedbacks) time will quickly run out.\nTherefore, I would appreciate to get feedback, codes, ideas, insights, findings as soon as possible and I insist that those who want to submit a paper should already consider the writing especially if not used to this exercise.
", "votes": 2, "replies": [] }, { "id": 926843, "author_name": "Haneol Jang", "author_url": "", "post_date": "2020-07-13T03:28:47.250000", "content": "I fully agree with Jessica Fridrich and Yifan Xie. We need more time to prepare the paper after this competition. That way, we can summarize more meaningful analysis and results.
", "votes": 0, "replies": [] }, { "id": 933840, "author_name": "", "author_url": "", "post_date": "2020-07-18T04:37:45.940000", "content": "", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "921899": "Dear Kagglers,\n\nAs you may have noticed, we have organized this competition as part of our research projets with the will to challenge our academic community with kagglers who have very little or even no knowledge in steganalysis but tremendous experience in deep learning and its practical use for solving image classification problems efficiently.\n\nSo far, this competition has been a great success for all of us in many aspects:\n- The competition ragged\n- Several new ideas have been thrown for our community (use of EfficientNet for instance)\n- Several breakthrough have been achieved (just like we though it was impossible to merge altogether difference JPEG quality factors, embedding schemes and color channels)\nAnd several ideas seems to be confirmed with all you numerous experiments:\n- Using DCT coefficients does not seem to be a clever move, use pixels values instead\n- Using Deep nets designed for natural image classification is a good starting point\n- Multi-class detection eventually pays off\n....\n\nI would like to call for help in order to get as much feedback as possible for our community.\nThere I kindly ask you two things\n(1) Can you please send me a message with your feedback, what you learn, how you addressed the problem and what you would have tried with more time (and possibly your code that allows getting your best results). \nThe goal is to provide our community with as much insights as possible throughout this informal pool survey. We will not keep those information to publish some research paper for our own fame (the organizers) but will share it openly with the academic community with the goal that it can shape our future research directions.\n(2) For the \"best\" competitors, consider submitting a paper to IEEE WIFS conference. The paper will be evaluated based on both the final score but also on the scientific contribution and analysis (a paper that simply describe a Deep Net architecture has little added values). I am especially thinking about people from research community that get excellent scores (ABBA team / Haneol Jang / Guanshuo Xu / Han Chen / etc. ...) but anyone else who believe have found interesting insights, ideas and findings that can be of general interest for our community, your submission will be more than welcome (note that a total 3k$ travel grant will be shared among accepted papers which can help but can hardly cover all travel costs)\n\nAgain feel free to message me (private message via kaggle or email directly at remi.cogranne@utt.fr) any feedback, findings, comment and details about codes will be more than welcome.", "925206": "Just wish there was more time between the end of this competition and WIFS deadline. You will need to lower the bar for the acceptance, Remi. Most teams will not be able to go much beyond basically summarizing their experience during this competition (which may be still valuable). A question to you: Will you be able to access the scores on both LBs and put something meaningful about the competition itself for WIFS? I assume you are planning this, right?", "927232": "i have a feeling that the sometimes deep network is just detecting texture rather than just hidden message. I would suggest that for future steganalysis competition, the test images should contain both the cover image and the stego image. We should also measure and compare the score difference the pair.\n\n", "921990": "Am I the only one who after reading this post is thinking: \"did the competition end already\"? 😆\n\nRe: sharing \"how you addressed the problem\" is not typically considered good etiquette in Kaggle to share key insights / code close to competition end.", "933848": "Hello there,\nA short message to let you know a good news regarding the follow-up of the competition.\nThe IEEE WIFS paper submission deadline has been \"unofficially\" extended by 10 days (to be accurate, until August 9th 23:59 AoE).\nBy unofficially, I mean it will be a two step process ; if you want to submit a paper you have to do it before Aug 1st. However, this can be a draft that can be revised by August 10th.\nWhat do we expect from you contributions ? \nIn those paper we expect not you to write a dairy about how the challenge was addressed, but why you eventually ended up with your proposal, what did you learn and how, what you would recommend for our community (and fact on which it is based) \n\nOn my hand, I do not have a total access to submissions. Not even sure I can get the raw submissions you provided. This is why I am asking to kindly let me know any useful information (code, submitted data, architecture description, which net you tried and results you got, etc. .... ). This would be very usefull in order to draw meaningful conclusion about this challenge.\n\nThanks in advance for you help, we remain available to help you out if you have more question about these aspect of article writing and/or information to provide me", "925228": "Just a thought, most of us who are working hard to make one last push for this competition, taking on the additional tasks of writing an academic paper sounds quite taxing. \n\nI guess the one feedback we could share is - for a workshop regarding a competition, making paper submission deadline so soon after finishing date of the competition doesn't seem to be a good idea... Bear in mind that potential prize winners would also need to be writing up for the winning solution documentation. and people do suffer burn out for working flat-out on these competitions on top of their day job. Would really be nice if this is taken into consideration. \n\nPerhaps you can work with Kaggle to come up with a winning submission documentation template such that the documentation can also be treated as a potential paper submission? \n\nAlso due to the small size public LB, it is perhaps hard to decide who might qualify as \"best\" competitors - private LB shakeup is a regular phenomenon on Kaggle. I would suggest, if @remicogranne also have access to people private's LB now, you might want to keep an eye out from insights from those team who are high on the private LB score. You must not approach them directly though because it will give unfair advantage :) \n\n\n\n\n", "922019": "Hello Andres,\nThanks for your message, I just realize that I may have forgotten to clarify something here ...\nWhile the competition did not end yet, (scientific) conference have strict deadlines and the one for IEEE WIFS, hosting a special session on this kaggle competition, implies that paper are due latest July 31st. Time to write a paper (and get + analyze first feedbacks) time will quickly run out.\nTherefore, I would appreciate to get feedback, codes, ideas, insights, findings as soon as possible and I insist that those who want to submit a paper should already consider the writing especially if not used to this exercise.", "926843": "I fully agree with Jessica Fridrich and Yifan Xie. We need more time to prepare the paper after this competition. That way, we can summarize more meaningful analysis and results.", "933840": "" } }