We used an end-to-end model for feature matching and successfully ran it on the 41 samples from test set after submission.
\nHowever, after running on the actual test set for a few hours, we encountered the error \"Submission CSV NOT Found\".
\nWe have successfully obtained scores in the following attempts(we use the original create_submission function as in post https://www.kaggle.com/competitions/image-matching-challenge-2024/discussion/491321#2778205):
The scores were as low as expected, which proves that the function create_submission works well.
\nWe also used try-except to handle exceptions in image reading and processing, but the error still exists. After local testing, the submission file can always be obtained, even if the h5 files for colmap (keypoints, matches) do not exist.
\nCan anyone share some experience of submission or give some advice?
", "messages": [ { "id": "2782613", "postDate": "04/29/2024 11:43:39", "content": "We used an end-to-end model for feature matching and successfully ran it on the 41 samples from test set after submission.
\nHowever, after running on the actual test set for a few hours, we encountered the error \"Submission CSV NOT Found\".
\nWe have successfully obtained scores in the following attempts(we use the original create_submission function as in post https://www.kaggle.com/competitions/image-matching-challenge-2024/discussion/491321#2778205):
The scores were as low as expected, which proves that the function create_submission works well.
\nWe also used try-except to handle exceptions in image reading and processing, but the error still exists. After local testing, the submission file can always be obtained, even if the h5 files for colmap (keypoints, matches) do not exist.
\nCan anyone share some experience of submission or give some advice?
", "rawMarkdown": "We used an end-to-end model for feature matching and successfully ran it on the 41 samples from test set after submission. \nHowever, after running on the actual test set for a few hours, we encountered the error \"Submission CSV NOT Found\". \nWe have successfully obtained scores in the following attempts(we use the original create_submission function as in post https://www.kaggle.com/competitions/image-matching-challenge-2024/discussion/491321#2778205):\n- run our pipeline only in the first three datasets,\n- traverse all scenes without any processing\n\nThe scores were as low as expected, which proves that the function create_submission works well.\n\nWe also used try-except to handle exceptions in image reading and processing, but the error still exists. After local testing, the submission file can always be obtained, even if the h5 files for colmap (keypoints, matches) do not exist.\n\nCan anyone share some experience of submission or give some advice?", "votes": null }, { "id": "2783472", "postDate": "04/29/2024 18:58:21", "content": "When you make a submission, a version of the notebook will be saved to run on the sample test data. You can check if there is a submission.csv or any typo of the name from the output tab.
\n![](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F3964695%2F324e1da61629da261e86e436c53fea1e%2FScreenshot%20from%202024-04-29%2014-35-39.png?generation=1714415756585089&alt=media\")
I have seen errors when the submission.csv is not the first (alphabetical order) file in the output (/kaggle/working/). I think the problem was solved by moving all other intermediate files to a /kaggle/tmp/ directory.
\n\nWe have successfully obtained scores in the following attempts(we use the original create_submission function as in post https://www.kaggle.com/competitions/image-matching-challenge-2024/discussion/491321#2778205):
\n\n
\n- run our pipeline only in the first three datasets,
\n- traverse all scenes without any processing
\nCan anyone share some experience of submission or give some advice?
\n
Since your pipeline works work the first three dataset, would that be possible that the intermediate files generated by your pipeline jammed the disk space so that your \"submissions.csv\" cannot be saved properly due to full disk space?
", "rawMarkdown": "> We have successfully obtained scores in the following attempts(we use the original create_submission function as in post https://www.kaggle.com/competitions/image-matching-challenge-2024/discussion/491321#2778205):\n> - run our pipeline only in the first three datasets,\n> - traverse all scenes without any processing\n\n> Can anyone share some experience of submission or give some advice?\n\nSince your pipeline works work the first three dataset, would that be possible that the intermediate files generated by your pipeline jammed the disk space so that your \"submissions.csv\" cannot be saved properly due to full disk space?", "votes": null }, { "id": "2784553", "postDate": "04/30/2024 10:51:02", "content": "It's strange. Our pipeline does not produce much intermediate data.
\nThanks for your comment.
Thanks. We will try your second suggestion. Our pipeline makes a folder in the /kaggle/working/.
Hello. I would like to ask you about same problem.
\nHow was the result of 2nd suggestion trying? Did it work?
That problem may not be the key one. We added more exceptions handling, such as input check before colmap, image reading check, and we solved it.
", "rawMarkdown": "That problem may not be the key one. We added more exceptions handling, such as input check before colmap, image reading check, and we solved it.", "votes": null }, { "id": "2825455", "postDate": "05/20/2024 12:13:04", "content": "I see that it was not important thing.
\nI will consider your comment. Thank you so much!!!
When you make a submission, a version of the notebook will be saved to run on the sample test data. You can check if there is a submission.csv or any typo of the name from the output tab.
\n
I have seen errors when the submission.csv is not the first (alphabetical order) file in the output (/kaggle/working/). I think the problem was solved by moving all other intermediate files to a /kaggle/tmp/ directory.
Thanks. We will try your second suggestion. Our pipeline makes a folder in the /kaggle/working/.
Hello. I would like to ask you about same problem.
\nHow was the result of 2nd suggestion trying? Did it work?
That problem may not be the key one. We added more exceptions handling, such as input check before colmap, image reading check, and we solved it.
", "votes": null, "replies": [ { "id": 2825455, "author_name": "katsuhisamasuda7", "author_url": "", "post_date": "05/20/2024 12:13:04", "content": "I see that it was not important thing.
\nI will consider your comment. Thank you so much!!!
\n\nWe have successfully obtained scores in the following attempts(we use the original create_submission function as in post https://www.kaggle.com/competitions/image-matching-challenge-2024/discussion/491321#2778205):
\n\n
\n- run our pipeline only in the first three datasets,
\n- traverse all scenes without any processing
\nCan anyone share some experience of submission or give some advice?
\n
Since your pipeline works work the first three dataset, would that be possible that the intermediate files generated by your pipeline jammed the disk space so that your \"submissions.csv\" cannot be saved properly due to full disk space?
", "votes": null, "replies": [ { "id": 2784553, "author_name": "xikegy", "author_url": "", "post_date": "04/30/2024 10:51:02", "content": "It's strange. Our pipeline does not produce much intermediate data.
\nThanks for your comment.