I don't have fancy hardware yet, so training locally for me is off.\nAre there people that consider working on the cloud? Is that even feasible considering amount of data (471 GB) and the obvious economic reasons?
", "messages": [ { "id": 692983, "postDate": "2019-12-11T23:34:31.847Z", "content": "I don't have fancy hardware yet, so training locally for me is off.\nAre there people that consider working on the cloud? Is that even feasible considering amount of data (471 GB) and the obvious economic reasons?
", "rawMarkdown": "I don't have fancy hardware yet, so training locally for me is off.\nAre there people that consider working on the cloud? Is that even feasible considering amount of data (471 GB) and the obvious economic reasons?", "votes": 19 }, { "id": 693013, "postDate": "2019-12-12T00:36:12.340Z", "content": "I think it will be very difficult for people without great hardware access. I am going to start working on grabbing a small sample of frames from videos and rescaling them so it is more manageable for people as just an image classification problem rather than a video problem.
", "rawMarkdown": "I think it will be very difficult for people without great hardware access. I am going to start working on grabbing a small sample of frames from videos and rescaling them so it is more manageable for people as just an image classification problem rather than a video problem. ", "votes": 11 }, { "id": 694417, "postDate": "2019-12-13T14:52:57.633Z", "content": "hopefully, we have also access to one of the sponsors: AWS. https://www.kaggle.com/c/deepfake-detection-challenge/discussion/121462#latest-694391
", "rawMarkdown": "hopefully, we have also access to one of the sponsors: AWS. https://www.kaggle.com/c/deepfake-detection-challenge/discussion/121462#latest-694391", "votes": 1 }, { "id": 693846, "postDate": "2019-12-12T21:06:51.550Z", "content": "they might give away gcp credits (probably will). note that you will probably need one better then baseline submission to get these, which is, I agree, a chicken and egg kind of thing.
", "rawMarkdown": "they might give away gcp credits (probably will). note that you will probably need one better then baseline submission to get these, which is, I agree, a chicken and egg kind of thing.", "votes": 2, "replies": [ { "id": 696175, "postDate": "2019-12-16T08:05:32.777Z", "content": "If you have used gcp than you know that $300 in credits can be spent pretty fast - don't see it possible that they will offer enough credits that you could manage a winning model.
\n\nWhat is need is some kind soul like ryches to build and share - even than $300 will be spent fast.
\n\nI did come to understand when I spent my sign up credits that gcp makes sense for mere mortals only on the run of the very last large model.
", "rawMarkdown": "If you have used gcp than you know that $300 in credits can be spent pretty fast - don't see it possible that they will offer enough credits that you could manage a winning model.\n\nWhat is need is some kind soul like ryches to build and share - even than $300 will be spent fast. \n\nI did come to understand when I spent my sign up credits that gcp makes sense for mere mortals only on the run of the very last large model. " } ] }, { "id": 693414, "postDate": "2019-12-12T10:44:58.373Z", "content": "The data size is huge!
", "rawMarkdown": "The data size is huge!", "votes": 2 }, { "id": 693020, "postDate": "2019-12-12T00:46:16.240Z", "content": "Spin you up an extra extra large aws instance
", "rawMarkdown": "Spin you up an extra extra large aws instance " }, { "id": 698121, "postDate": "2019-12-18T20:31:47.177Z", "rawMarkdown": "", "votes": 1, "isDeleted": true } ], "comments": [ { "id": 693013, "author_name": "ryches", "author_url": "", "post_date": "2019-12-12T00:36:12.340000", "content": "I think it will be very difficult for people without great hardware access. I am going to start working on grabbing a small sample of frames from videos and rescaling them so it is more manageable for people as just an image classification problem rather than a video problem.
", "votes": 11, "replies": [] }, { "id": 694417, "author_name": "FGPC", "author_url": "", "post_date": "2019-12-13T14:52:57.633000", "content": "hopefully, we have also access to one of the sponsors: AWS. https://www.kaggle.com/c/deepfake-detection-challenge/discussion/121462#latest-694391
", "votes": 1, "replies": [] }, { "id": 693846, "author_name": "Moshel", "author_url": "", "post_date": "2019-12-12T21:06:51.550000", "content": "they might give away gcp credits (probably will). note that you will probably need one better then baseline submission to get these, which is, I agree, a chicken and egg kind of thing.
", "votes": 2, "replies": [ { "id": 696175, "author_name": "PC Jimmmy", "author_url": "", "post_date": "2019-12-16T08:05:32.777000", "content": "If you have used gcp than you know that $300 in credits can be spent pretty fast - don't see it possible that they will offer enough credits that you could manage a winning model.
\n\nWhat is need is some kind soul like ryches to build and share - even than $300 will be spent fast.
\n\nI did come to understand when I spent my sign up credits that gcp makes sense for mere mortals only on the run of the very last large model.
", "votes": 0, "replies": [] } ] }, { "id": 693414, "author_name": "Firat Gonen", "author_url": "", "post_date": "2019-12-12T10:44:58.373000", "content": "The data size is huge!
", "votes": 2, "replies": [] }, { "id": 693020, "author_name": "", "author_url": "", "post_date": "2019-12-12T00:46:16.240000", "content": "Spin you up an extra extra large aws instance
", "votes": 0, "replies": [] }, { "id": 698121, "author_name": "", "author_url": "", "post_date": "2019-12-18T20:31:47.177000", "content": "", "votes": 1, "replies": [] } ], "raw_markdown_by_id": { "692983": "I don't have fancy hardware yet, so training locally for me is off.\nAre there people that consider working on the cloud? Is that even feasible considering amount of data (471 GB) and the obvious economic reasons?", "693013": "I think it will be very difficult for people without great hardware access. I am going to start working on grabbing a small sample of frames from videos and rescaling them so it is more manageable for people as just an image classification problem rather than a video problem. ", "694417": "hopefully, we have also access to one of the sponsors: AWS. https://www.kaggle.com/c/deepfake-detection-challenge/discussion/121462#latest-694391", "693846": "they might give away gcp credits (probably will). note that you will probably need one better then baseline submission to get these, which is, I agree, a chicken and egg kind of thing.", "693414": "The data size is huge!", "693020": "Spin you up an extra extra large aws instance ", "698121": "" } }