![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2234817%2F633f7e95e1816524ebb273da33fe2d35%2FDeepfake_Detection.png?generation=1576477912786964&alt=media\")
Hi All , \nThis is a quite new and interesting competition . I am planning to join this after some time , atleast when Baidu finishes . However , after reading some basic ideas about this problem, I am thinking of this pipeline in future . Right now , i dont know how to code for this , but hopefully , with a little help from open source codes and people here , I could figure out atleast few of the planned items . Whats your plan ?
\n\n
Hi All , \nThis is a quite new and interesting competition . I am planning to join this after some time , atleast when Baidu finishes . However , after reading some basic ideas about this problem, I am thinking of this pipeline in future . Right now , i dont know how to code for this , but hopefully , with a little help from open source codes and people here , I could figure out atleast few of the planned items . Whats your plan ?
\n\n
Looks interesting, I'm curious how could one fit all of these under the 9h constraints + 1 GB Model weights constraint.
", "rawMarkdown": "Looks interesting, I'm curious how could one fit all of these under the 9h constraints + 1 GB Model weights constraint.", "votes": null }, { "id": "699201", "postDate": "12/20/2019 06:38:51", "content": "Could you explain the Speech-to-text pipeline? I feel its implementation delusional.
", "rawMarkdown": "Could you explain the Speech-to-text pipeline? I feel its implementation delusional.", "votes": null }, { "id": "699207", "postDate": "12/20/2019 06:52:11", "content": "It could be . The idea behind that was not substantiated by the data we have in this comp . I had the thought let's say all videos are taken from different personalities and different personalities have different point of view ,knowledge and personal biases and inclination. If we could convert those into word vectors, then in the fake audio in real world might give away certain information . Just like how we identify in real world . Let's say you know me and you know my right/left political inclination about some issue . If you find one of my video which gives you the impression that I am talking something in a completely different inclination , you might think ..humm this can't be him . But the applicability of the same in this comp is questionable.
", "rawMarkdown": "It could be . The idea behind that was not substantiated by the data we have in this comp . I had the thought let's say all videos are taken from different personalities and different personalities have different point of view ,knowledge and personal biases and inclination. If we could convert those into word vectors, then in the fake audio in real world might give away certain information . Just like how we identify in real world . Let's say you know me and you know my right/left political inclination about some issue . If you find one of my video which gives you the impression that I am talking something in a completely different inclination , you might think ..humm this can't be him . But the applicability of the same in this comp is questionable.", "votes": null }, { "id": "699208", "postDate": "12/20/2019 07:01:05", "content": "That would be really interesting. This could turn out to be a completely new discussion-thread or contest on its own. Amazing.
", "rawMarkdown": "That would be really interesting. This could turn out to be a completely new discussion-thread or contest on its own. Amazing.", "votes": null }, { "id": "699305", "postDate": "12/20/2019 09:46:16", "content": "Use DVC if you want ML pipeline
", "rawMarkdown": "Use DVC if you want ML pipeline", "votes": null }, { "id": "700030", "postDate": "12/21/2019 10:03:05", "content": "I like it, definitely similar to what I had in mind.
\n\nIgnoring the model constraints for a moment, I wonder how adding a stacked classifier at the end would perform by comparison.
", "rawMarkdown": "I like it, definitely similar to what I had in mind. \n\nIgnoring the model constraints for a moment, I wonder how adding a stacked classifier at the end would perform by comparison.", "votes": null } ], "comments": [ { "id": 696883, "author_name": "init27", "author_url": "", "post_date": "12/17/2019 07:23:54", "content": "Looks interesting, I'm curious how could one fit all of these under the 9h constraints + 1 GB Model weights constraint.
", "votes": null, "replies": [] }, { "id": 699201, "author_name": "thanatoz", "author_url": "", "post_date": "12/20/2019 06:38:51", "content": "Could you explain the Speech-to-text pipeline? I feel its implementation delusional.
", "votes": null, "replies": [ { "id": 699207, "author_name": "phoenix9032", "author_url": "", "post_date": "12/20/2019 06:52:11", "content": "It could be . The idea behind that was not substantiated by the data we have in this comp . I had the thought let's say all videos are taken from different personalities and different personalities have different point of view ,knowledge and personal biases and inclination. If we could convert those into word vectors, then in the fake audio in real world might give away certain information . Just like how we identify in real world . Let's say you know me and you know my right/left political inclination about some issue . If you find one of my video which gives you the impression that I am talking something in a completely different inclination , you might think ..humm this can't be him . But the applicability of the same in this comp is questionable.
", "votes": null, "replies": [] }, { "id": 699208, "author_name": "thanatoz", "author_url": "", "post_date": "12/20/2019 07:01:05", "content": "That would be really interesting. This could turn out to be a completely new discussion-thread or contest on its own. Amazing.
", "votes": null, "replies": [] } ] }, { "id": 699305, "author_name": "kurianbenoy", "author_url": "", "post_date": "12/20/2019 09:46:16", "content": "Use DVC if you want ML pipeline
", "votes": null, "replies": [] }, { "id": 700030, "author_name": "thenerdycat", "author_url": "", "post_date": "12/21/2019 10:03:05", "content": "I like it, definitely similar to what I had in mind.
\n\nIgnoring the model constraints for a moment, I wonder how adding a stacked classifier at the end would perform by comparison.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "696121": "Hi All , \nThis is a quite new and interesting competition . I am planning to join this after some time , atleast when Baidu finishes . However , after reading some basic ideas about this problem, I am thinking of this pipeline in future . Right now , i dont know how to code for this , but hopefully , with a little help from open source codes and people here , I could figure out atleast few of the planned items . Whats your plan ?\n\n", "696883": "Looks interesting, I'm curious how could one fit all of these under the 9h constraints + 1 GB Model weights constraint.", "699201": "Could you explain the Speech-to-text pipeline? I feel its implementation delusional.", "699207": "It could be . The idea behind that was not substantiated by the data we have in this comp . I had the thought let's say all videos are taken from different personalities and different personalities have different point of view ,knowledge and personal biases and inclination. If we could convert those into word vectors, then in the fake audio in real world might give away certain information . Just like how we identify in real world . Let's say you know me and you know my right/left political inclination about some issue . If you find one of my video which gives you the impression that I am talking something in a completely different inclination , you might think ..humm this can't be him . But the applicability of the same in this comp is questionable.", "699208": "That would be really interesting. This could turn out to be a completely new discussion-thread or contest on its own. Amazing.", "699305": "Use DVC if you want ML pipeline", "700030": "I like it, definitely similar to what I had in mind. \n\nIgnoring the model constraints for a moment, I wonder how adding a stacked classifier at the end would perform by comparison." }, "source": "meta" }