I was wondering how much time do you need for the inference part ?
\nMy script is done running in 5 minutes because I infer on the single patient_id folder but it take a lot more time to get lb score with the hidden data. After 7-8h, it is still scoring…
I was wondering how much time do you need for the inference part ?
\nMy script is done running in 5 minutes because I infer on the single patient_id folder but it take a lot more time to get lb score with the hidden data. After 7-8h, it is still scoring…
For better inference time check out: https://www.kaggle.com/competitions/rsna-breast-cancer-detection/discussion/375881
", "rawMarkdown": "For better inference time check out: https://www.kaggle.com/competitions/rsna-breast-cancer-detection/discussion/375881", "votes": 1 }, { "id": 2083697, "postDate": "2023-01-02T18:42:13.757Z", "content": "I use dicomsdl from this test: https://www.kaggle.com/code/remekkinas/fast-dicom-processing-1-6-2x-faster
\nI ran a test (only for time measurement)- 3x efficientnet_b2 models + TTA (each model). It takes about 6h to finish submission.
Indeed i am using pydicom, might not be the fastest one, thank you very much
", "rawMarkdown": "Indeed i am using pydicom, might not be the fastest one, thank you very much", "replies": [ { "id": 2087499, "postDate": "2023-01-05T16:28:59.437Z", "content": "yes, definitely - there are few notebooks with Nvidia DALI. Eventually I will use them for speed up. Now I play locally and look for single model improvement. Then I will look for speed up.
", "rawMarkdown": "yes, definitely - there are few notebooks with Nvidia DALI. Eventually I will use them for speed up. Now I play locally and look for single model improvement. Then I will look for speed up.", "votes": 1 } ] } ] }, { "id": 2083643, "postDate": "2023-01-02T17:46:46.320Z", "content": "For me it takes between 5 to 7 hours depending on the model, resolution, etc.
", "rawMarkdown": "For me it takes between 5 to 7 hours depending on the model, resolution, etc.", "votes": 1 }, { "id": 2083642, "postDate": "2023-01-02T17:44:40.910Z", "content": "Mine takes around 5-6h,I guess it can be brought down to 3-4h by using a faster library for dicom handling like dicomsdl
", "rawMarkdown": "Mine takes around 5-6h,I guess it can be brought down to 3-4h by using a faster library for dicom handling like dicomsdl", "votes": 1, "replies": [ { "id": 2083832, "postDate": "2023-01-02T22:24:57.553Z", "content": "thank you, i will have a look at that one too
", "rawMarkdown": "thank you, i will have a look at that one too" } ] }, { "id": 2087244, "postDate": "2023-01-05T13:14:49.150Z", "content": "it takes somthing between 7 to 8 hours:)
", "rawMarkdown": "it takes somthing between 7 to 8 hours:)" }, { "id": 2083648, "postDate": "2023-01-02T17:48:34.253Z", "rawMarkdown": "", "isDeleted": true } ], "comments": [ { "id": 2087930, "author_name": "Ali Abdin", "author_url": "", "post_date": "2023-01-05T22:46:34.517000", "content": "For better inference time check out: https://www.kaggle.com/competitions/rsna-breast-cancer-detection/discussion/375881
", "votes": 1, "replies": [] }, { "id": 2083697, "author_name": "Remek Kinas", "author_url": "", "post_date": "2023-01-02T18:42:13.757000", "content": "I use dicomsdl from this test: https://www.kaggle.com/code/remekkinas/fast-dicom-processing-1-6-2x-faster
\nI ran a test (only for time measurement)- 3x efficientnet_b2 models + TTA (each model). It takes about 6h to finish submission.
Indeed i am using pydicom, might not be the fastest one, thank you very much
", "votes": 0, "replies": [ { "id": 2087499, "author_name": "Remek Kinas", "author_url": "", "post_date": "2023-01-05T16:28:59.437000", "content": "yes, definitely - there are few notebooks with Nvidia DALI. Eventually I will use them for speed up. Now I play locally and look for single model improvement. Then I will look for speed up.
", "votes": 1, "replies": [] } ] } ] }, { "id": 2083643, "author_name": "Rasoul Mojtahedzadeh", "author_url": "", "post_date": "2023-01-02T17:46:46.320000", "content": "For me it takes between 5 to 7 hours depending on the model, resolution, etc.
", "votes": 1, "replies": [] }, { "id": 2083642, "author_name": "Marius ", "author_url": "", "post_date": "2023-01-02T17:44:40.910000", "content": "Mine takes around 5-6h,I guess it can be brought down to 3-4h by using a faster library for dicom handling like dicomsdl
", "votes": 1, "replies": [ { "id": 2083832, "author_name": "BryanB", "author_url": "", "post_date": "2023-01-02T22:24:57.553000", "content": "thank you, i will have a look at that one too
", "votes": 0, "replies": [] } ] }, { "id": 2087244, "author_name": "Armin Azhdehnia", "author_url": "", "post_date": "2023-01-05T13:14:49.150000", "content": "it takes somthing between 7 to 8 hours:)
", "votes": 0, "replies": [] }, { "id": 2083648, "author_name": "", "author_url": "", "post_date": "2023-01-02T17:48:34.253000", "content": "", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "2083630": "I was wondering how much time do you need for the inference part ?\nMy script is done running in 5 minutes because I infer on the single patient_id folder but it take a lot more time to get lb score with the hidden data. After 7-8h, it is still scoring...\n", "2087930": "For better inference time check out: https://www.kaggle.com/competitions/rsna-breast-cancer-detection/discussion/375881", "2083697": "I use dicomsdl from this test: https://www.kaggle.com/code/remekkinas/fast-dicom-processing-1-6-2x-faster\nI ran a test (only for time measurement)- 3x efficientnet_b2 models + TTA (each model). It takes about 6h to finish submission.", "2083643": "For me it takes between 5 to 7 hours depending on the model, resolution, etc.", "2083642": "Mine takes around 5-6h,I guess it can be brought down to 3-4h by using a faster library for dicom handling like dicomsdl", "2087244": "it takes somthing between 7 to 8 hours:)", "2083648": "" } }