We notice there are about 5%-10% real videos with multiple actors. It results in a lot of DeepFake videos with more than one face, and I think we would have to label individual face ourselves to decide which face is fake. Is there a good way to reduce the efforts?
\n\nAlso, for a real video with multiple actors (for example two actors A and B) in it and the corresponding DF videos it generates, is the swapped face always from the same actor, or it could swap actor A in DF video 1, and swap actor B's face in DF video 2?
", "messages": [ { "id": "753035", "postDate": "02/21/2020 16:39:29", "content": "We notice there are about 5%-10% real videos with multiple actors. It results in a lot of DeepFake videos with more than one face, and I think we would have to label individual face ourselves to decide which face is fake. Is there a good way to reduce the efforts?
\n\nAlso, for a real video with multiple actors (for example two actors A and B) in it and the corresponding DF videos it generates, is the swapped face always from the same actor, or it could swap actor A in DF video 1, and swap actor B's face in DF video 2?
", "rawMarkdown": "We notice there are about 5%-10% real videos with multiple actors. It results in a lot of DeepFake videos with more than one face, and I think we would have to label individual face ourselves to decide which face is fake. Is there a good way to reduce the efforts?\n\nAlso, for a real video with multiple actors (for example two actors A and B) in it and the corresponding DF videos it generates, is the swapped face always from the same actor, or it could swap actor A in DF video 1, and swap actor B's face in DF video 2?", "votes": null }, { "id": "753325", "postDate": "02/22/2020 03:23:06", "content": "Year, I just find that there are exactly 9250 videos that have multiple actors. At the very start, I thought that there was only one person manipulated, whose face is closer to a front face. However, I just noticed that there exists videos that both of actors are manipulated. (edylmobkop.mp4)\nI also wonder how to deal with this kind of videos.
", "rawMarkdown": "Year, I just find that there are exactly 9250 videos that have multiple actors. At the very start, I thought that there was only one person manipulated, whose face is closer to a front face. However, I just noticed that there exists videos that both of actors are manipulated. (edylmobkop.mp4)\nI also wonder how to deal with this kind of videos.", "votes": null }, { "id": "753328", "postDate": "02/22/2020 03:27:49", "content": "Remember, you have the original for each fake video... Hint hint
", "rawMarkdown": "Remember, you have the original for each fake video... Hint hint", "votes": null }, { "id": "755328", "postDate": "02/24/2020 17:01:28", "content": "Thanks! However, we already found due to some compression issues, not just the DeepFake face changes the pixel. Comparing the source video and the fake video, we found the pixels are updated everywhere.
", "rawMarkdown": "Thanks! However, we already found due to some compression issues, not just the DeepFake face changes the pixel. Comparing the source video and the fake video, we found the pixels are updated everywhere.", "votes": null }, { "id": "756961", "postDate": "02/26/2020 09:16:22", "content": "Answering for completeness. The aim of the competition is to detect whether a video contains a DeepFake or not. Of course, there may be multiple faces and not all of them will be fakes. Which face is the fake one in a given video is not within the scope of the competition, just predict that that video contains a DeepFake. How fakes are generated and what identities are borrowed to create the final video will not be disclosed for this competition. Hope this helps!
", "rawMarkdown": "Answering for completeness. The aim of the competition is to detect whether a video contains a DeepFake or not. Of course, there may be multiple faces and not all of them will be fakes. Which face is the fake one in a given video is not within the scope of the competition, just predict that that video contains a DeepFake. How fakes are generated and what identities are borrowed to create the final video will not be disclosed for this competition. Hope this helps!", "votes": null }, { "id": "757320", "postDate": "02/26/2020 16:34:51", "content": "Thanks for clarifying. I understand the rationals, but this makes it hard to use the dataset. For most of our solutions, it involves machine learning and training data. Labeling an entire frame as fake with some of the faces as real, and some of the faces are fake are not very useful and in the end we need to figure out which face is fake ourselves before we can use them for training.
", "rawMarkdown": "Thanks for clarifying. I understand the rationals, but this makes it hard to use the dataset. For most of our solutions, it involves machine learning and training data. Labeling an entire frame as fake with some of the faces as real, and some of the faces are fake are not very useful and in the end we need to figure out which face is fake ourselves before we can use them for training.", "votes": null } ], "comments": [ { "id": 753325, "author_name": "daddyjin", "author_url": "", "post_date": "02/22/2020 03:23:06", "content": "Year, I just find that there are exactly 9250 videos that have multiple actors. At the very start, I thought that there was only one person manipulated, whose face is closer to a front face. However, I just noticed that there exists videos that both of actors are manipulated. (edylmobkop.mp4)\nI also wonder how to deal with this kind of videos.
", "votes": null, "replies": [ { "id": 753328, "author_name": "moshel", "author_url": "", "post_date": "02/22/2020 03:27:49", "content": "Remember, you have the original for each fake video... Hint hint
", "votes": null, "replies": [] }, { "id": 755328, "author_name": "yipinzhou", "author_url": "", "post_date": "02/24/2020 17:01:28", "content": "Thanks! However, we already found due to some compression issues, not just the DeepFake face changes the pixel. Comparing the source video and the fake video, we found the pixels are updated everywhere.
", "votes": null, "replies": [] } ] }, { "id": 756961, "author_name": "cristiancanton", "author_url": "", "post_date": "02/26/2020 09:16:22", "content": "Answering for completeness. The aim of the competition is to detect whether a video contains a DeepFake or not. Of course, there may be multiple faces and not all of them will be fakes. Which face is the fake one in a given video is not within the scope of the competition, just predict that that video contains a DeepFake. How fakes are generated and what identities are borrowed to create the final video will not be disclosed for this competition. Hope this helps!
", "votes": null, "replies": [ { "id": 757320, "author_name": "yipinzhou", "author_url": "", "post_date": "02/26/2020 16:34:51", "content": "Thanks for clarifying. I understand the rationals, but this makes it hard to use the dataset. For most of our solutions, it involves machine learning and training data. Labeling an entire frame as fake with some of the faces as real, and some of the faces are fake are not very useful and in the end we need to figure out which face is fake ourselves before we can use them for training.
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "753035": "We notice there are about 5%-10% real videos with multiple actors. It results in a lot of DeepFake videos with more than one face, and I think we would have to label individual face ourselves to decide which face is fake. Is there a good way to reduce the efforts?\n\nAlso, for a real video with multiple actors (for example two actors A and B) in it and the corresponding DF videos it generates, is the swapped face always from the same actor, or it could swap actor A in DF video 1, and swap actor B's face in DF video 2?", "753325": "Year, I just find that there are exactly 9250 videos that have multiple actors. At the very start, I thought that there was only one person manipulated, whose face is closer to a front face. However, I just noticed that there exists videos that both of actors are manipulated. (edylmobkop.mp4)\nI also wonder how to deal with this kind of videos.", "753328": "Remember, you have the original for each fake video... Hint hint", "755328": "Thanks! However, we already found due to some compression issues, not just the DeepFake face changes the pixel. Comparing the source video and the fake video, we found the pixels are updated everywhere.", "756961": "Answering for completeness. The aim of the competition is to detect whether a video contains a DeepFake or not. Of course, there may be multiple faces and not all of them will be fakes. Which face is the fake one in a given video is not within the scope of the competition, just predict that that video contains a DeepFake. How fakes are generated and what identities are borrowed to create the final video will not be disclosed for this competition. Hope this helps!", "757320": "Thanks for clarifying. I understand the rationals, but this makes it hard to use the dataset. For most of our solutions, it involves machine learning and training data. Labeling an entire frame as fake with some of the faces as real, and some of the faces are fake are not very useful and in the end we need to figure out which face is fake ourselves before we can use them for training." }, "source": "meta" }