![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F3117946%2F0740325ffbd1a7d7b44acad7d3301b48%2F1.png?generation=1578223219042480&alt=media\")
Hi all!
\n\nWe'll be updating the private test set shortly to eliminate a metadata leak via reported aspect ratio strings. On Monday, when we have full engineering resources, we'll kick off a rescore. Note that the rescore could take some time - there's a large backlog of submissions to get through. Please bear with us! Thanks for your patience. In the meantime, any new submissions will be evaluated using the updated data.
\n\nWe investigated a number of other reported leaks and found that they either were not leaks, or existed only in the training set. Please use caution when attempting to use metadata for predictions!
\n\nThanks to all of the Kagglers who reported issues - we very much appreciate it.
\n\nLet us know if you have any questions!
", "messages": [ { "id": 709002, "postDate": "2020-01-02T23:15:07.127Z", "content": "Hi all!
\n\nWe'll be updating the private test set shortly to eliminate a metadata leak via reported aspect ratio strings. On Monday, when we have full engineering resources, we'll kick off a rescore. Note that the rescore could take some time - there's a large backlog of submissions to get through. Please bear with us! Thanks for your patience. In the meantime, any new submissions will be evaluated using the updated data.
\n\nWe investigated a number of other reported leaks and found that they either were not leaks, or existed only in the training set. Please use caution when attempting to use metadata for predictions!
\n\nThanks to all of the Kagglers who reported issues - we very much appreciate it.
\n\nLet us know if you have any questions!
", "rawMarkdown": "Hi all!\n\nWe'll be updating the private test set shortly to eliminate a metadata leak via reported aspect ratio strings. On Monday, when we have full engineering resources, we'll kick off a rescore. Note that the rescore could take some time - there's a large backlog of submissions to get through. Please bear with us! Thanks for your patience. In the meantime, any new submissions will be evaluated using the updated data.\n\nWe investigated a number of other reported leaks and found that they either were not leaks, or existed only in the training set. Please use caution when attempting to use metadata for predictions!\n\nThanks to all of the Kagglers who reported issues - we very much appreciate it.\n\nLet us know if you have any questions!", "votes": 35 }, { "id": 709718, "postDate": "2020-01-03T21:45:46.923Z", "content": "On Monday, on the stroke of twelve, the spell will be broken, and all these 0.49274 submissions will turn into pumpkin (1.13121 to be precise).
", "rawMarkdown": "On Monday, on the stroke of twelve, the spell will be broken, and all these 0.49274 submissions will turn into pumpkin (1.13121 to be precise). ", "votes": 14, "replies": [ { "id": 710695, "postDate": "2020-01-05T04:54:47.570Z", "content": "> Note that the rescore could take some time - there's a large backlog of submissions to get through.
\n\nNot that fast! 😄
", "rawMarkdown": "> Note that the rescore could take some time - there's a large backlog of submissions to get through.\n\nNot that fast! 😄 " } ] }, { "id": 712431, "postDate": "2020-01-07T08:32:44.220Z", "content": "I propose to move 1 new updated video from private to public test set. So we have better chance to understand code problems on initial run (without submit).
", "rawMarkdown": "I propose to move 1 new updated video from private to public test set. So we have better chance to understand code problems on initial run (without submit).", "votes": 5, "replies": [ { "id": 712511, "postDate": "2020-01-07T10:10:42.227Z", "content": "Good idea, my last 4 submissions returned \"Submission Scoring Error\" while I iterate through potential problems. I am almost sure it is something special on the public test.
\n\nBut really any help from Kaggle in relation to the submission errors that multiple people report will be welcomed.
", "rawMarkdown": "Good idea, my last 4 submissions returned \"Submission Scoring Error\" while I iterate through potential problems. I am almost sure it is something special on the public test.\n\nBut really any help from Kaggle in relation to the submission errors that multiple people report will be welcomed." } ] }, { "id": 712384, "postDate": "2020-01-07T07:07:59.887Z", "content": "Since nothing has been said - is it correct that you made no modifications to the files in the Test Public set?\nIs it correct that your going to leave us completely in the dark with regards the general nature of your fix?
", "rawMarkdown": "Since nothing has been said - is it correct that you made no modifications to the files in the Test Public set?\nIs it correct that your going to leave us completely in the dark with regards the general nature of your fix?", "votes": 1, "replies": [ { "id": 712395, "postDate": "2020-01-07T07:23:01.337Z", "content": "I also agree that giving more details is prompted. What fields have been cleared, and how? I noticed the same submission gave different results already after the update, what is it? We now have different files structure in train and test, and statements like the following don't help:
\n\n\n", "rawMarkdown": "I also agree that giving more details is prompted. What fields have been cleared, and how? I noticed the same submission gave different results already after the update, what is it? We now have different files structure in train and test, and statements like the following don't help:\n\n> Please use caution when attempting to use metadata for predictions", "votes": 1 }, { "id": 712653, "postDate": "2020-01-07T13:31:15.223Z", "content": "Please use caution when attempting to use metadata for predictions
\n
Thanks for your questions!
\n\n@pcjimmmy - what do you mean by the public test set? You mean the small 400 video test set?
\n\n@zaharch - what do you mean by different file structures in train and test?
\n\nAgreed about giving out details - more are definitely needed. My plan is to provide more detail when the rescore is complete. Right now we may need to make further modifications if issues arise.
\n\nSorry about the general nature of the statement noted above - I think it's fair to say that being careful about using metadata is not a bad idea, though. In examining this and other reports I noted that exiftools, MediaInfo (and its Python library), and ffprobe all surfaced things slightly differently, and that's likely to be even more pervasive in the wild. I'd focus on the content of the videos, not what ffprobe or something else surfaces.
I did mean the 400 video test set - my experience and lots of posts in discussion show that pesky submission errors are going to be pretty common in the competition. I think I am at about a 50% rate. Stepping thru the small 400 does not promise success with the full private set, but if it at least was \"fixed\" in a manner matching what you have done to the full private it would help keep that 50% submission failure rate of mine in check.
\n\nOf course, the real help would be a lot better error feedback beyond the current - sorry you goofed message we now get.
\n\nI easily see the potential that I walk away from this competition because I can't get a successful submission.
", "rawMarkdown": "I did mean the 400 video test set - my experience and lots of posts in discussion show that pesky submission errors are going to be pretty common in the competition. I think I am at about a 50% rate. Stepping thru the small 400 does not promise success with the full private set, but if it at least was \"fixed\" in a manner matching what you have done to the full private it would help keep that 50% submission failure rate of mine in check.\n\nOf course, the real help would be a lot better error feedback beyond the current - sorry you goofed message we now get.\n\nI easily see the potential that I walk away from this competition because I can't get a successful submission. ", "votes": 2 } ] }, { "id": 710877, "postDate": "2020-01-05T11:21:07.107Z", "content": "Does this update change the content of the test video? After this update, I used the same code and pre-trained model, but got a different score.
\n\nBefore update:\n
After update:\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F3117946%2F1e80385f688f4cc3d7afb8665e463379%2F2.png?generation=1578223236163324&alt=media\")
I think they did. Without doing so it is not possible to re-evaluate the score of previous submissions done using leaks.
", "rawMarkdown": "I think they did. Without doing so it is not possible to re-evaluate the score of previous submissions done using leaks." }, { "id": 712232, "postDate": "2020-01-07T01:09:11.587Z", "content": "BTW, I didn't use data leak in the code. It's just a image-based binary classification model.
", "rawMarkdown": "BTW, I didn't use data leak in the code. It's just a image-based binary classification model.", "votes": 1 }, { "id": 712250, "postDate": "2020-01-07T01:49:18.527Z", "content": "Make sure your script runs even if one video loading fails. I abandoned torch dataset/dataloader and just did like the following pseudo code.\n\nfor path in paths:\n try:\n # load video and predict\n except:\n # predict 0.5\n
![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F3117946%2F4cba6149f340b72310350c7683cdc485%2F_20200110103835.png?generation=1578623942867406&alt=media\")
@philculliton I found version4 has been rescore and the result is the same as before. Does this mean that there was a problem with the test set at some point after metadata was updated?
", "rawMarkdown": "\n\n@philculliton I found version4 has been rescore and the result is the same as before. Does this mean that there was a problem with the test set at some point after metadata was updated?\n\n" }, { "id": 715005, "postDate": "2020-01-10T02:40:47.483Z", "content": "I suddenly found that the score of Version1 was different too.
", "rawMarkdown": "I suddenly found that the score of Version1 was different too." } ] }, { "id": 728187, "postDate": "2020-01-24T13:45:08.383Z", "content": "Hi Phil,
\n\nWould you be able to let me know whether all the frames in deep faked videos are altered? or Some of the frames only? Also, the given data for kaggle kernel, we could not find most of the original files as named in the .json file's 'original' column. (error code below)
\n\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1345685%2F7e925f29f4d7578fcba1e526db762219%2FScreenshot_2020-01-19_00-10-35.png?generation=1579873440795683&alt=media\")
Your answer is much appreciated.
\n\nThanks
\n\nMichael
", "rawMarkdown": "Hi Phil,\n\nWould you be able to let me know whether all the frames in deep faked videos are altered? or Some of the frames only? Also, the given data for kaggle kernel, we could not find most of the original files as named in the .json file's 'original' column. (error code below)\n\n\n\nYour answer is much appreciated.\n\nThanks\n\nMichael\n " }, { "id": 714182, "postDate": "2020-01-09T06:48:11.463Z", "content": "seems the rescoring is done? the LB is useful again 😆
", "rawMarkdown": "seems the rescoring is done? the LB is useful again 😆 ", "replies": [ { "id": 714194, "postDate": "2020-01-09T06:59:20.797Z", "content": "Maybe not yet. Like @unkownhihi said, scores from deleted leak kernels are still up ;)
", "rawMarkdown": "Maybe not yet. Like @unkownhihi said, scores from deleted leak kernels are still up ;)" }, { "id": 714307, "postDate": "2020-01-09T09:33:44.543Z", "content": "lol, I am just happy to get a boost on the LB 😄
", "rawMarkdown": "lol, I am just happy to get a boost on the LB 😄 ", "votes": 2 } ] }, { "id": 713859, "postDate": "2020-01-08T18:14:56.613Z", "content": "Just to get little more clarity, Is there any change to the 474GB training set? . Do we need to download it again?
", "rawMarkdown": "Just to get little more clarity, Is there any change to the 474GB training set? . Do we need to download it again?", "replies": [ { "id": 713862, "postDate": "2020-01-08T18:18:11.937Z", "content": "The training set has not changed, just the test set that's used when you submit your notebook.
", "rawMarkdown": "The training set has not changed, just the test set that's used when you submit your notebook." } ] }, { "id": 712285, "postDate": "2020-01-07T03:12:55.910Z", "content": "When will the rescore end? I was using the leak and my position in leaderboard got better. Is there a bug or something?
", "rawMarkdown": "When will the rescore end? I was using the leak and my position in leaderboard got better. Is there a bug or something?", "replies": [ { "id": 712612, "postDate": "2020-01-07T12:54:23.640Z", "content": "@unkownhihi - The rescore may go on for some time. What's happening with your code may be a bug, though. Could you share your notebook with me, please?
", "rawMarkdown": "@unkownhihi - The rescore may go on for some time. What's happening with your code may be a bug, though. Could you share your notebook with me, please?" }, { "id": 713106, "postDate": "2020-01-07T22:48:47.610Z", "content": "I deleted that notebook on accident. I think it was just a fork of some public notebook using the leak.
", "rawMarkdown": "I deleted that notebook on accident. I think it was just a fork of some public notebook using the leak." }, { "id": 713207, "postDate": "2020-01-08T02:52:42.480Z", "content": "A screen shot FYI\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F3319985%2Fb3b6857d48c49c38ce73ed15183bced9%2FScreen%20Shot%202020-01-07%20at%206.52.09%20PM.png?generation=1578451960504267&alt=media\")
@philculliton I think I figured out the bug. The score will keep the same if the notebook is deleted because there's nothing to rerun.
", "rawMarkdown": "@philculliton I think I figured out the bug. The score will keep the same if the notebook is deleted because there's nothing to rerun.", "votes": 1 }, { "id": 714403, "postDate": "2020-01-09T12:08:11.427Z", "content": "That's correct! It's less a bug than an artifact of the process. At the very end of the rescore those will be cleaned up. Thanks for checking in about it!
", "rawMarkdown": "That's correct! It's less a bug than an artifact of the process. At the very end of the rescore those will be cleaned up. Thanks for checking in about it!" }, { "id": 721357, "postDate": "2020-01-17T09:49:12.333Z", "content": "Hello @philculliton,
\n\nIs there any plan or timetable to remove those artifacts?
\n\nThank you.
", "rawMarkdown": "Hello @philculliton,\n\nIs there any plan or timetable to remove those artifacts?\n\nThank you." }, { "id": 721997, "postDate": "2020-01-18T01:08:26.230Z", "content": "Hi! did the rescore end? I hoping the rescore will end before 1/30 so that my leak score will be canceled and won't count for AWS credit distribution. My leak submission was on 12/27/19, if the score is not canceled, I won't be able to receive the credit for Jan.
", "rawMarkdown": "Hi! did the rescore end? I hoping the rescore will end before 1/30 so that my leak score will be canceled and won't count for AWS credit distribution. My leak submission was on 12/27/19, if the score is not canceled, I won't be able to receive the credit for Jan. " } ] }, { "id": 709276, "postDate": "2020-01-03T10:02:46.887Z", "content": "Apdated already? Are the public score calculated on the new test set now?
", "rawMarkdown": "Apdated already? Are the public score calculated on the new test set now?", "replies": [ { "id": 709598, "postDate": "2020-01-03T17:53:43.023Z", "content": "The public score for any new submissions will be calculated on the new test set, correct.
", "rawMarkdown": "The public score for any _new_ submissions will be calculated on the new test set, correct.", "votes": 1 } ] }, { "id": 713093, "postDate": "2020-01-07T21:56:57.433Z", "rawMarkdown": "", "isDeleted": true }, { "id": 712230, "postDate": "2020-01-07T01:08:47.077Z", "rawMarkdown": "", "isDeleted": true } ], "comments": [ { "id": 709718, "author_name": "nosound", "author_url": "", "post_date": "2020-01-03T21:45:46.923000", "content": "On Monday, on the stroke of twelve, the spell will be broken, and all these 0.49274 submissions will turn into pumpkin (1.13121 to be precise).
", "votes": 14, "replies": [ { "id": 710695, "author_name": "Shangqiu Li", "author_url": "", "post_date": "2020-01-05T04:54:47.570000", "content": "> Note that the rescore could take some time - there's a large backlog of submissions to get through.
\n\nNot that fast! 😄
", "votes": 0, "replies": [] } ] }, { "id": 712431, "author_name": "ZFTurbo", "author_url": "", "post_date": "2020-01-07T08:32:44.220000", "content": "I propose to move 1 new updated video from private to public test set. So we have better chance to understand code problems on initial run (without submit).
", "votes": 5, "replies": [ { "id": 712511, "author_name": "nosound", "author_url": "", "post_date": "2020-01-07T10:10:42.227000", "content": "Good idea, my last 4 submissions returned \"Submission Scoring Error\" while I iterate through potential problems. I am almost sure it is something special on the public test.
\n\nBut really any help from Kaggle in relation to the submission errors that multiple people report will be welcomed.
", "votes": 0, "replies": [] } ] }, { "id": 712384, "author_name": "PC Jimmmy", "author_url": "", "post_date": "2020-01-07T07:07:59.887000", "content": "Since nothing has been said - is it correct that you made no modifications to the files in the Test Public set?\nIs it correct that your going to leave us completely in the dark with regards the general nature of your fix?
", "votes": 1, "replies": [ { "id": 712395, "author_name": "nosound", "author_url": "", "post_date": "2020-01-07T07:23:01.337000", "content": "I also agree that giving more details is prompted. What fields have been cleared, and how? I noticed the same submission gave different results already after the update, what is it? We now have different files structure in train and test, and statements like the following don't help:
\n\n\n", "votes": 1, "replies": [] }, { "id": 712653, "author_name": "Phil Culliton", "author_url": "", "post_date": "2020-01-07T13:31:15.223000", "content": "Please use caution when attempting to use metadata for predictions
\n
Thanks for your questions!
\n\n@pcjimmmy - what do you mean by the public test set? You mean the small 400 video test set?
\n\n@zaharch - what do you mean by different file structures in train and test?
\n\nAgreed about giving out details - more are definitely needed. My plan is to provide more detail when the rescore is complete. Right now we may need to make further modifications if issues arise.
\n\nSorry about the general nature of the statement noted above - I think it's fair to say that being careful about using metadata is not a bad idea, though. In examining this and other reports I noted that exiftools, MediaInfo (and its Python library), and ffprobe all surfaced things slightly differently, and that's likely to be even more pervasive in the wild. I'd focus on the content of the videos, not what ffprobe or something else surfaces.
I did mean the 400 video test set - my experience and lots of posts in discussion show that pesky submission errors are going to be pretty common in the competition. I think I am at about a 50% rate. Stepping thru the small 400 does not promise success with the full private set, but if it at least was \"fixed\" in a manner matching what you have done to the full private it would help keep that 50% submission failure rate of mine in check.
\n\nOf course, the real help would be a lot better error feedback beyond the current - sorry you goofed message we now get.
\n\nI easily see the potential that I walk away from this competition because I can't get a successful submission.
", "votes": 2, "replies": [] } ] }, { "id": 710877, "author_name": "Chason", "author_url": "", "post_date": "2020-01-05T11:21:07.107000", "content": "Does this update change the content of the test video? After this update, I used the same code and pre-trained model, but got a different score.
\n\nBefore update:\n
After update:\n
I think they did. Without doing so it is not possible to re-evaluate the score of previous submissions done using leaks.
", "votes": 0, "replies": [] }, { "id": 712232, "author_name": "Chason", "author_url": "", "post_date": "2020-01-07T01:09:11.587000", "content": "BTW, I didn't use data leak in the code. It's just a image-based binary classification model.
", "votes": 1, "replies": [] }, { "id": 712250, "author_name": "YoonSoo", "author_url": "", "post_date": "2020-01-07T01:49:18.527000", "content": "Make sure your script runs even if one video loading fails. I abandoned torch dataset/dataloader and just did like the following pseudo code.\n\nfor path in paths:\n try:\n # load video and predict\n except:\n # predict 0.5\n
@philculliton I found version4 has been rescore and the result is the same as before. Does this mean that there was a problem with the test set at some point after metadata was updated?
", "votes": 0, "replies": [] }, { "id": 715005, "author_name": "Chason", "author_url": "", "post_date": "2020-01-10T02:40:47.483000", "content": "I suddenly found that the score of Version1 was different too.
", "votes": 0, "replies": [] } ] }, { "id": 728187, "author_name": "Schroter", "author_url": "", "post_date": "2020-01-24T13:45:08.383000", "content": "Hi Phil,
\n\nWould you be able to let me know whether all the frames in deep faked videos are altered? or Some of the frames only? Also, the given data for kaggle kernel, we could not find most of the original files as named in the .json file's 'original' column. (error code below)
\n\n
Your answer is much appreciated.
\n\nThanks
\n\nMichael
", "votes": 0, "replies": [] }, { "id": 714182, "author_name": "Yifan Xie", "author_url": "", "post_date": "2020-01-09T06:48:11.463000", "content": "seems the rescoring is done? the LB is useful again 😆
", "votes": 0, "replies": [ { "id": 714194, "author_name": "YoonSoo", "author_url": "", "post_date": "2020-01-09T06:59:20.797000", "content": "Maybe not yet. Like @unkownhihi said, scores from deleted leak kernels are still up ;)
", "votes": 0, "replies": [] }, { "id": 714307, "author_name": "Yifan Xie", "author_url": "", "post_date": "2020-01-09T09:33:44.543000", "content": "lol, I am just happy to get a boost on the LB 😄
", "votes": 2, "replies": [] } ] }, { "id": 713859, "author_name": "Al Warrior", "author_url": "", "post_date": "2020-01-08T18:14:56.613000", "content": "Just to get little more clarity, Is there any change to the 474GB training set? . Do we need to download it again?
", "votes": 0, "replies": [ { "id": 713862, "author_name": "Phil Culliton", "author_url": "", "post_date": "2020-01-08T18:18:11.937000", "content": "The training set has not changed, just the test set that's used when you submit your notebook.
", "votes": 0, "replies": [] } ] }, { "id": 712285, "author_name": "Shangqiu Li", "author_url": "", "post_date": "2020-01-07T03:12:55.910000", "content": "When will the rescore end? I was using the leak and my position in leaderboard got better. Is there a bug or something?
", "votes": 0, "replies": [ { "id": 712612, "author_name": "Phil Culliton", "author_url": "", "post_date": "2020-01-07T12:54:23.640000", "content": "@unkownhihi - The rescore may go on for some time. What's happening with your code may be a bug, though. Could you share your notebook with me, please?
", "votes": 0, "replies": [] }, { "id": 713106, "author_name": "Shangqiu Li", "author_url": "", "post_date": "2020-01-07T22:48:47.610000", "content": "I deleted that notebook on accident. I think it was just a fork of some public notebook using the leak.
", "votes": 0, "replies": [] }, { "id": 713207, "author_name": "Shangqiu Li", "author_url": "", "post_date": "2020-01-08T02:52:42.480000", "content": "A screen shot FYI\n
@philculliton I think I figured out the bug. The score will keep the same if the notebook is deleted because there's nothing to rerun.
", "votes": 1, "replies": [] }, { "id": 714403, "author_name": "Phil Culliton", "author_url": "", "post_date": "2020-01-09T12:08:11.427000", "content": "That's correct! It's less a bug than an artifact of the process. At the very end of the rescore those will be cleaned up. Thanks for checking in about it!
", "votes": 0, "replies": [] }, { "id": 721357, "author_name": "deepware.ai", "author_url": "", "post_date": "2020-01-17T09:49:12.333000", "content": "Hello @philculliton,
\n\nIs there any plan or timetable to remove those artifacts?
\n\nThank you.
", "votes": 0, "replies": [] }, { "id": 721997, "author_name": "Shangqiu Li", "author_url": "", "post_date": "2020-01-18T01:08:26.230000", "content": "Hi! did the rescore end? I hoping the rescore will end before 1/30 so that my leak score will be canceled and won't count for AWS credit distribution. My leak submission was on 12/27/19, if the score is not canceled, I won't be able to receive the credit for Jan.
", "votes": 0, "replies": [] } ] }, { "id": 709276, "author_name": "Kevin Zheng", "author_url": "", "post_date": "2020-01-03T10:02:46.887000", "content": "Apdated already? Are the public score calculated on the new test set now?
", "votes": 0, "replies": [ { "id": 709598, "author_name": "Phil Culliton", "author_url": "", "post_date": "2020-01-03T17:53:43.023000", "content": "The public score for any new submissions will be calculated on the new test set, correct.
", "votes": 1, "replies": [] } ] }, { "id": 713093, "author_name": "", "author_url": "", "post_date": "2020-01-07T21:56:57.433000", "content": "", "votes": 0, "replies": [] }, { "id": 712230, "author_name": "", "author_url": "", "post_date": "2020-01-07T01:08:47.077000", "content": "", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "709002": "Hi all!\n\nWe'll be updating the private test set shortly to eliminate a metadata leak via reported aspect ratio strings. On Monday, when we have full engineering resources, we'll kick off a rescore. Note that the rescore could take some time - there's a large backlog of submissions to get through. Please bear with us! Thanks for your patience. In the meantime, any new submissions will be evaluated using the updated data.\n\nWe investigated a number of other reported leaks and found that they either were not leaks, or existed only in the training set. Please use caution when attempting to use metadata for predictions!\n\nThanks to all of the Kagglers who reported issues - we very much appreciate it.\n\nLet us know if you have any questions!", "709718": "On Monday, on the stroke of twelve, the spell will be broken, and all these 0.49274 submissions will turn into pumpkin (1.13121 to be precise). ", "712431": "I propose to move 1 new updated video from private to public test set. So we have better chance to understand code problems on initial run (without submit).", "712384": "Since nothing has been said - is it correct that you made no modifications to the files in the Test Public set?\nIs it correct that your going to leave us completely in the dark with regards the general nature of your fix?", "710877": "Does this update change the content of the test video? After this update, I used the same code and pre-trained model, but got a different score.\n\nBefore update:\n\n\nAfter update:\n\n", "728187": "Hi Phil,\n\nWould you be able to let me know whether all the frames in deep faked videos are altered? or Some of the frames only? Also, the given data for kaggle kernel, we could not find most of the original files as named in the .json file's 'original' column. (error code below)\n\n\n\nYour answer is much appreciated.\n\nThanks\n\nMichael\n ", "714182": "seems the rescoring is done? the LB is useful again 😆 ", "713859": "Just to get little more clarity, Is there any change to the 474GB training set? . Do we need to download it again?", "712285": "When will the rescore end? I was using the leak and my position in leaderboard got better. Is there a bug or something?", "709276": "Apdated already? Are the public score calculated on the new test set now?", "713093": "", "712230": "" } }