Hi all,
\n\nI wrote a baseline model using one frame and a pre-trained resnet18 https://www.kaggle.com/xinyiding/deepfake-single-frame. But I keep getting this 'Submission CSV Not Found' error. My guess is there is some exception before generating the submission csv. I have a model works fine without the face detection stuff. So added some try except just for the face detection.
\n\nI have no problem committing and generating the 400 scores for the public dataset. Also, when I trained my model, I almost used all the 500G data available and didn't see any errors. Do we expect some videos from the private test set that are very different from this 500g dataset (say invalid videos or videos with less than 60 frames ?)
\n\nOr do you think it might be a memory issue? Thank you for your help!
\n\nUPDATE: I fixed this issue by enclosing a try-except block for each prediction, my guess is there are some videos that opencv will throw some errors
", "messages": [ { "id": 711352, "postDate": "2020-01-06T00:49:24.500Z", "content": "Hi all,
\n\nI wrote a baseline model using one frame and a pre-trained resnet18 https://www.kaggle.com/xinyiding/deepfake-single-frame. But I keep getting this 'Submission CSV Not Found' error. My guess is there is some exception before generating the submission csv. I have a model works fine without the face detection stuff. So added some try except just for the face detection.
\n\nI have no problem committing and generating the 400 scores for the public dataset. Also, when I trained my model, I almost used all the 500G data available and didn't see any errors. Do we expect some videos from the private test set that are very different from this 500g dataset (say invalid videos or videos with less than 60 frames ?)
\n\nOr do you think it might be a memory issue? Thank you for your help!
\n\nUPDATE: I fixed this issue by enclosing a try-except block for each prediction, my guess is there are some videos that opencv will throw some errors
", "rawMarkdown": "Hi all,\n\nI wrote a baseline model using one frame and a pre-trained resnet18 [https://www.kaggle.com/xinyiding/deepfake-single-frame](https://www.kaggle.com/xinyiding/deepfake-single-frame). But I keep getting this 'Submission CSV Not Found' error. My guess is there is some exception before generating the submission csv. I have a model works fine without the face detection stuff. So added some try except just for the face detection.\n\nI have no problem committing and generating the 400 scores for the public dataset. Also, when I trained my model, I almost used all the 500G data available and didn't see any errors. Do we expect some videos from the private test set that are very different from this 500g dataset (say invalid videos or videos with less than 60 frames ?)\n\nOr do you think it might be a memory issue? Thank you for your help!\n\n\nUPDATE: I fixed this issue by enclosing a try-except block for each prediction, my guess is there are some videos that opencv will throw some errors", "votes": 7 }, { "id": 711607, "postDate": "2020-01-06T10:00:35.830Z", "content": "I have error \"Submission CSV Not Found\" for the same code, which works ok 4 days ago. Looks like some problem at Kaggle scoring system.
", "rawMarkdown": "I have error \"Submission CSV Not Found\" for the same code, which works ok 4 days ago. Looks like some problem at Kaggle scoring system.", "votes": 5, "replies": [ { "id": 712368, "postDate": "2020-01-07T06:30:03.823Z", "content": "I had similar issues and I suspect there are some corrupt video files in the updated test dataset. I abandoned pytorch dataloader and instead did like below pseudo code to solve the issues:\n\nsubmission = pd.read_csv('...sample_submission.csv')\nsubmission['label'] = 0.5\nsubmission.set_index('filename')\nfor path in paths:\n try:\n # load video and predict\n submission.loc[path] = prediction\n except:\n pass\nsubmission.to_csv('submission.csv', index=True)\n
I didn't use try-except block in my code. I probably will give it a try.
\n\nThe other idea what could happen: after reencode or update of videos in private test set, opencv couldn't read them, because of new codec or container.
", "rawMarkdown": "I didn't use try-except block in my code. I probably will give it a try. \n\nThe other idea what could happen: after reencode or update of videos in private test set, opencv couldn't read them, because of new codec or container.", "votes": 2 }, { "id": 712844, "postDate": "2020-01-07T16:26:36.733Z", "content": "> I have error \"Submission CSV Not Found\" for the same code, which works ok 4 days ago.
\n\nI have the same problem with submission which successfully run before holidays (today it fails in first few minutes). \nI've done minor changes in video reading & frames handling and now submission has not fallen yet...
\n\nUPD : Submission run successfully.
", "rawMarkdown": "> I have error \"Submission CSV Not Found\" for the same code, which works ok 4 days ago.\n\nI have the same problem with submission which successfully run before holidays (today it fails in first few minutes). \nI've done minor changes in **video reading** & **frames handling** and now submission has not fallen yet...\n\nUPD : Submission run successfully.\n", "votes": 5, "isDeleted": true }, { "id": 713542, "postDate": "2020-01-08T12:02:56.117Z", "content": "My submissions now work fine too.
", "rawMarkdown": "My submissions now work fine too.", "votes": 2 }, { "id": 720066, "postDate": "2020-01-16T05:53:40.557Z", "rawMarkdown": "", "isDeleted": true } ] }, { "id": 725983, "postDate": "2020-01-22T17:31:03.893Z", "content": "@xinyiding thanks for sharing your notebook. reviewing your code was very insightful.
", "rawMarkdown": "@xinyiding thanks for sharing your notebook. reviewing your code was very insightful. " }, { "id": 725107, "postDate": "2020-01-21T20:12:06.903Z", "content": "So what is the solution here?\nI guess it has something to do with way we take input, and where we generate output (in the same directory??).\nShould our model look for: \"test_videos.zip\" or something else? Will it be in the same directory?
", "rawMarkdown": "So what is the solution here?\nI guess it has something to do with way we take input, and where we generate output (in the same directory??).\nShould our model look for: \"test_videos.zip\" or something else? Will it be in the same directory? " }, { "id": 713531, "postDate": "2020-01-08T11:43:13.777Z", "content": "I have exactly the same error, even when providing a submission CSV without accessing to any data :
\n\npython\nsubmission = pd.read_csv(\"../input/deepfake-detection-challenge/sample_submission.csv\")\nsubmission['label'] = 0.5\nsubmission.set_index('filename')\nsubmission.to_csv('submission.csv', index=True)\n
I tried to contact the support where I received the following answer :
\n\n\n\n\nThis is a suitable question to best ask in the discussion or competition forums. Your peers are a great resource for these types of questions. Please note that Kaggle support is reserved for technical issues affecting all participants.
\n
This is actually not anymore my problem but the problem of multiple different submissions ...
", "rawMarkdown": "I have exactly the same error, even when providing a submission CSV without accessing to any data : \n\n```python\nsubmission = pd.read_csv(\"../input/deepfake-detection-challenge/sample_submission.csv\")\nsubmission['label'] = 0.5\nsubmission.set_index('filename')\nsubmission.to_csv('submission.csv', index=True)\n```\n\nI tried to contact the support where I received the following answer : \n> This is a suitable question to best ask in the discussion or competition forums. Your peers are a great resource for these types of questions. Please note that Kaggle support is reserved for technical issues affecting all participants.\n\nThis is actually not anymore my problem but the problem of multiple different submissions ..." }, { "id": 712550, "postDate": "2020-01-07T11:14:23.683Z", "content": "Hi All,
\n\nI downloaded the large data-set (470GB). However, after unzipping I can only find the .mp4 files. Would anyone be able to let me know where is the .json (metadata) file and is the whole 470GB data-set is for training? If so how could I get the test set?
\n\nThanks
\n\nMichael
", "rawMarkdown": "Hi All, \n\nI downloaded the large data-set (470GB). However, after unzipping I can only find the .mp4 files. Would anyone be able to let me know where is the .json (metadata) file and is the whole 470GB data-set is for training? If so how could I get the test set? \n\nThanks\n\nMichael", "replies": [ { "id": 712909, "postDate": "2020-01-07T18:02:16.420Z", "content": "The json file is in the same folder as videos, just sort by file size, you will find it
", "rawMarkdown": "The json file is in the same folder as videos, just sort by file size, you will find it" } ] }, { "id": 712366, "postDate": "2020-01-07T06:26:43.087Z", "content": "you can try to kick off the internet button in your kernel and then commit. hope that can help you.
", "rawMarkdown": "you can try to kick off the internet button in your kernel and then commit. hope that can help you." }, { "id": 711447, "postDate": "2020-01-06T05:39:05.860Z", "content": "I have got the same error now, no idea what it is yet.
", "rawMarkdown": "I have got the same error now, no idea what it is yet.", "replies": [ { "id": 711452, "postDate": "2020-01-06T05:57:46.500Z", "content": "One workaround is not to use the data loader from pytorch, but iterate through each video and enclose each prediction with a try except block. This might not be the most efficient way, but it did solve my problem.
", "rawMarkdown": "One workaround is not to use the data loader from pytorch, but iterate through each video and enclose each prediction with a try except block. This might not be the most efficient way, but it did solve my problem. ", "votes": 1 } ] }, { "id": 712667, "postDate": "2020-01-07T13:46:07.793Z", "rawMarkdown": "", "votes": 1, "isDeleted": true } ], "comments": [ { "id": 711607, "author_name": "ZFTurbo", "author_url": "", "post_date": "2020-01-06T10:00:35.830000", "content": "I have error \"Submission CSV Not Found\" for the same code, which works ok 4 days ago. Looks like some problem at Kaggle scoring system.
", "votes": 5, "replies": [ { "id": 712368, "author_name": "YoonSoo", "author_url": "", "post_date": "2020-01-07T06:30:03.823000", "content": "I had similar issues and I suspect there are some corrupt video files in the updated test dataset. I abandoned pytorch dataloader and instead did like below pseudo code to solve the issues:\n\nsubmission = pd.read_csv('...sample_submission.csv')\nsubmission['label'] = 0.5\nsubmission.set_index('filename')\nfor path in paths:\n try:\n # load video and predict\n submission.loc[path] = prediction\n except:\n pass\nsubmission.to_csv('submission.csv', index=True)\n
I didn't use try-except block in my code. I probably will give it a try.
\n\nThe other idea what could happen: after reencode or update of videos in private test set, opencv couldn't read them, because of new codec or container.
", "votes": 2, "replies": [] }, { "id": 712844, "author_name": "", "author_url": "", "post_date": "2020-01-07T16:26:36.733000", "content": "> I have error \"Submission CSV Not Found\" for the same code, which works ok 4 days ago.
\n\nI have the same problem with submission which successfully run before holidays (today it fails in first few minutes). \nI've done minor changes in video reading & frames handling and now submission has not fallen yet...
\n\nUPD : Submission run successfully.
", "votes": 5, "replies": [] }, { "id": 713542, "author_name": "ZFTurbo", "author_url": "", "post_date": "2020-01-08T12:02:56.117000", "content": "My submissions now work fine too.
", "votes": 2, "replies": [] }, { "id": 720066, "author_name": "", "author_url": "", "post_date": "2020-01-16T05:53:40.557000", "content": "", "votes": 0, "replies": [] } ] }, { "id": 725983, "author_name": "Nick Sciarrilli", "author_url": "", "post_date": "2020-01-22T17:31:03.893000", "content": "@xinyiding thanks for sharing your notebook. reviewing your code was very insightful.
", "votes": 0, "replies": [] }, { "id": 725107, "author_name": "AshutoshNirala", "author_url": "", "post_date": "2020-01-21T20:12:06.903000", "content": "So what is the solution here?\nI guess it has something to do with way we take input, and where we generate output (in the same directory??).\nShould our model look for: \"test_videos.zip\" or something else? Will it be in the same directory?
", "votes": 0, "replies": [] }, { "id": 713531, "author_name": "P. Breton", "author_url": "", "post_date": "2020-01-08T11:43:13.777000", "content": "I have exactly the same error, even when providing a submission CSV without accessing to any data :
\n\npython\nsubmission = pd.read_csv(\"../input/deepfake-detection-challenge/sample_submission.csv\")\nsubmission['label'] = 0.5\nsubmission.set_index('filename')\nsubmission.to_csv('submission.csv', index=True)\n
I tried to contact the support where I received the following answer :
\n\n\n\n\nThis is a suitable question to best ask in the discussion or competition forums. Your peers are a great resource for these types of questions. Please note that Kaggle support is reserved for technical issues affecting all participants.
\n
This is actually not anymore my problem but the problem of multiple different submissions ...
", "votes": 0, "replies": [] }, { "id": 712550, "author_name": "Schroter", "author_url": "", "post_date": "2020-01-07T11:14:23.683000", "content": "Hi All,
\n\nI downloaded the large data-set (470GB). However, after unzipping I can only find the .mp4 files. Would anyone be able to let me know where is the .json (metadata) file and is the whole 470GB data-set is for training? If so how could I get the test set?
\n\nThanks
\n\nMichael
", "votes": 0, "replies": [ { "id": 712909, "author_name": "Xinyi Ding", "author_url": "", "post_date": "2020-01-07T18:02:16.420000", "content": "The json file is in the same folder as videos, just sort by file size, you will find it
", "votes": 0, "replies": [] } ] }, { "id": 712366, "author_name": "CKC_WDH", "author_url": "", "post_date": "2020-01-07T06:26:43.087000", "content": "you can try to kick off the internet button in your kernel and then commit. hope that can help you.
", "votes": 0, "replies": [] }, { "id": 711447, "author_name": "nosound", "author_url": "", "post_date": "2020-01-06T05:39:05.860000", "content": "I have got the same error now, no idea what it is yet.
", "votes": 0, "replies": [ { "id": 711452, "author_name": "Xinyi Ding", "author_url": "", "post_date": "2020-01-06T05:57:46.500000", "content": "One workaround is not to use the data loader from pytorch, but iterate through each video and enclose each prediction with a try except block. This might not be the most efficient way, but it did solve my problem.
", "votes": 1, "replies": [] } ] }, { "id": 712667, "author_name": "", "author_url": "", "post_date": "2020-01-07T13:46:07.793000", "content": "", "votes": 1, "replies": [] } ], "raw_markdown_by_id": { "711352": "Hi all,\n\nI wrote a baseline model using one frame and a pre-trained resnet18 [https://www.kaggle.com/xinyiding/deepfake-single-frame](https://www.kaggle.com/xinyiding/deepfake-single-frame). But I keep getting this 'Submission CSV Not Found' error. My guess is there is some exception before generating the submission csv. I have a model works fine without the face detection stuff. So added some try except just for the face detection.\n\nI have no problem committing and generating the 400 scores for the public dataset. Also, when I trained my model, I almost used all the 500G data available and didn't see any errors. Do we expect some videos from the private test set that are very different from this 500g dataset (say invalid videos or videos with less than 60 frames ?)\n\nOr do you think it might be a memory issue? Thank you for your help!\n\n\nUPDATE: I fixed this issue by enclosing a try-except block for each prediction, my guess is there are some videos that opencv will throw some errors", "711607": "I have error \"Submission CSV Not Found\" for the same code, which works ok 4 days ago. Looks like some problem at Kaggle scoring system.", "725983": "@xinyiding thanks for sharing your notebook. reviewing your code was very insightful. ", "725107": "So what is the solution here?\nI guess it has something to do with way we take input, and where we generate output (in the same directory??).\nShould our model look for: \"test_videos.zip\" or something else? Will it be in the same directory? ", "713531": "I have exactly the same error, even when providing a submission CSV without accessing to any data : \n\n```python\nsubmission = pd.read_csv(\"../input/deepfake-detection-challenge/sample_submission.csv\")\nsubmission['label'] = 0.5\nsubmission.set_index('filename')\nsubmission.to_csv('submission.csv', index=True)\n```\n\nI tried to contact the support where I received the following answer : \n> This is a suitable question to best ask in the discussion or competition forums. Your peers are a great resource for these types of questions. Please note that Kaggle support is reserved for technical issues affecting all participants.\n\nThis is actually not anymore my problem but the problem of multiple different submissions ...", "712550": "Hi All, \n\nI downloaded the large data-set (470GB). However, after unzipping I can only find the .mp4 files. Would anyone be able to let me know where is the .json (metadata) file and is the whole 470GB data-set is for training? If so how could I get the test set? \n\nThanks\n\nMichael", "712366": "you can try to kick off the internet button in your kernel and then commit. hope that can help you.", "711447": "I have got the same error now, no idea what it is yet.", "712667": "" } }