I am facing with problem of Submission scoring error. I have already tried too many times considering many things suggested online about format of submission csv file. I am unaware of what is causing this problem.
\n\nCommit works without any problem.
\n\nI am left to try only one option. Should i need to read test.csv file for submitting or sample_submission.csv?
\n\nwhy is this happening and could anyone suggest what else I could try?
", "messages": [ { "id": "745885", "postDate": "02/14/2020 10:13:08", "content": "I am facing with problem of Submission scoring error. I have already tried too many times considering many things suggested online about format of submission csv file. I am unaware of what is causing this problem.
\n\nCommit works without any problem.
\n\nI am left to try only one option. Should i need to read test.csv file for submitting or sample_submission.csv?
\n\nwhy is this happening and could anyone suggest what else I could try?
", "rawMarkdown": "I am facing with problem of Submission scoring error. I have already tried too many times considering many things suggested online about format of submission csv file. I am unaware of what is causing this problem.\n\nCommit works without any problem.\n\nI am left to try only one option. Should i need to read test.csv file for submitting or sample_submission.csv?\n\nwhy is this happening and could anyone suggest what else I could try?", "votes": null }, { "id": "745959", "postDate": "02/14/2020 12:30:11", "content": "This is why I wrote this notebook. 😄
", "rawMarkdown": "This is why I wrote [this notebook](https://www.kaggle.com/humananalog/inference-demo). 😄", "votes": null }, { "id": "746031", "postDate": "02/14/2020 14:12:22", "content": "Thank you! The kernel is helpful. I will try it. However, I am having a question.\nI am using MTCNN for face detection. Is Blazenet performs better than MTCNN in terms of memory and speed? \nAlso, for the training, do you read 17 frames from video and train before moving to 17 frames of next video on the fly or do you get 17 frames from all the videos and save before training them?
", "rawMarkdown": "Thank you! The kernel is helpful. I will try it. However, I am having a question.\nI am using MTCNN for face detection. Is Blazenet performs better than MTCNN in terms of memory and speed? \nAlso, for the training, do you read 17 frames from video and train before moving to 17 frames of next video on the fly or do you get 17 frames from all the videos and save before training them?", "votes": null }, { "id": "746085", "postDate": "02/14/2020 15:49:39", "content": "It depends on your hardware but I think BlazeFace is faster than MTCNN indeed. Usually the trade-off is that it sees fewer faces or has more false positives.
\n\nFor training I made a script that extracts 10 (random) frames from each video, makes face crops, and saves these as PNG files to a folder. This is about 1 million files in total. You do this just once, and then train on those extracted face crops.
", "rawMarkdown": "It depends on your hardware but I think BlazeFace is faster than MTCNN indeed. Usually the trade-off is that it sees fewer faces or has more false positives.\n\nFor training I made a script that extracts 10 (random) frames from each video, makes face crops, and saves these as PNG files to a folder. This is about 1 million files in total. You do this just once, and then train on those extracted face crops.", "votes": null }, { "id": "748425", "postDate": "02/17/2020 14:07:13", "content": "Thank you. Sorry to bother you again but after following the core principles from your kernel for submission, I tried submitting and now I end up getting Submission csv not found error. Although I didn't use ThreadPoolexecutor because It consumed the whole RAM even for public test. Whereas Dataloader consumed significantly less memory. When I commit the file, I see the submission file in output section. It didn't even take 3 minutes to throw that error. Any idea?
", "rawMarkdown": "Thank you. Sorry to bother you again but after following the core principles from your kernel for submission, I tried submitting and now I end up getting Submission csv not found error. Although I didn't use ThreadPoolexecutor because It consumed the whole RAM even for public test. Whereas Dataloader consumed significantly less memory. When I commit the file, I see the submission file in output section. It didn't even take 3 minutes to throw that error. Any idea?", "votes": null }, { "id": "748740", "postDate": "02/17/2020 23:05:58", "content": "I'm pretty sure one of the cell(the pd.to_csv('submission.csv',index=False')) one didn't execute because it threw an error.
This is why I wrote this notebook. 😄
", "votes": null, "replies": [ { "id": 746031, "author_name": "srinivasanvasu89", "author_url": "", "post_date": "02/14/2020 14:12:22", "content": "Thank you! The kernel is helpful. I will try it. However, I am having a question.\nI am using MTCNN for face detection. Is Blazenet performs better than MTCNN in terms of memory and speed? \nAlso, for the training, do you read 17 frames from video and train before moving to 17 frames of next video on the fly or do you get 17 frames from all the videos and save before training them?
", "votes": null, "replies": [] }, { "id": 746085, "author_name": "humananalog", "author_url": "", "post_date": "02/14/2020 15:49:39", "content": "It depends on your hardware but I think BlazeFace is faster than MTCNN indeed. Usually the trade-off is that it sees fewer faces or has more false positives.
\n\nFor training I made a script that extracts 10 (random) frames from each video, makes face crops, and saves these as PNG files to a folder. This is about 1 million files in total. You do this just once, and then train on those extracted face crops.
", "votes": null, "replies": [] }, { "id": 748425, "author_name": "srinivasanvasu89", "author_url": "", "post_date": "02/17/2020 14:07:13", "content": "Thank you. Sorry to bother you again but after following the core principles from your kernel for submission, I tried submitting and now I end up getting Submission csv not found error. Although I didn't use ThreadPoolexecutor because It consumed the whole RAM even for public test. Whereas Dataloader consumed significantly less memory. When I commit the file, I see the submission file in output section. It didn't even take 3 minutes to throw that error. Any idea?
", "votes": null, "replies": [] }, { "id": 748740, "author_name": "unkownhihi", "author_url": "", "post_date": "02/17/2020 23:05:58", "content": "I'm pretty sure one of the cell(the pd.to_csv('submission.csv',index=False')) one didn't execute because it threw an error.