{ "id": 388791, "title": "[Prime time] Cutting-edge enhancement methods that could further improve your LB Score (Augmentations)", "url": "/competitions/rsna-breast-cancer-detection/discussion/388791", "author_name": "Dheeraj Pandey", "post_date": "2023-02-19T14:33:23.906000", "votes": 6, "comment_count": 6, "views": 0, "content": "

These methods were borrowed from Notebook, which is located at: here . Thank you @kyleberdy for sharing cutting-edge augmentation methods.

\n

These augmentation methods helped my experiments and increased the score on my Public LB.

\n

This one describes a model of CLAHE, and some of the limits of traditional models: https://towardsdatascience.com/increase-your-face-recognition-models-accuracy-by-improving-face-contrast-a3e71bb6b9fb

\n

This one describes the opencv application of adaptive Histogram equalization on grayscale images: https://opencv-python-tutroals.readthedocs.io/en/latest/py_tutorials/py_imgproc/py_histograms/py_histogram_equalization/py_histogram_equalization.html

\n

Same as above, but in \"layman's terms\" (is anything AI ever in layman's terms): https://www.geeksforgeeks.org/clahe-histogram-eqalization-opencv/

", "messages": [ { "id": 2150733, "postDate": "2023-02-19T14:33:23.907Z", "content": "

These methods were borrowed from Notebook, which is located at: here . Thank you @kyleberdy for sharing cutting-edge augmentation methods.

\n

These augmentation methods helped my experiments and increased the score on my Public LB.

\n

This one describes a model of CLAHE, and some of the limits of traditional models: https://towardsdatascience.com/increase-your-face-recognition-models-accuracy-by-improving-face-contrast-a3e71bb6b9fb

\n

This one describes the opencv application of adaptive Histogram equalization on grayscale images: https://opencv-python-tutroals.readthedocs.io/en/latest/py_tutorials/py_imgproc/py_histograms/py_histogram_equalization/py_histogram_equalization.html

\n

Same as above, but in \"layman's terms\" (is anything AI ever in layman's terms): https://www.geeksforgeeks.org/clahe-histogram-eqalization-opencv/

", "rawMarkdown": "These methods were borrowed from Notebook, which is located at: [here](https://www.kaggle.com/code/kyleberdy/advanced-image-preprocessing-with-clahe) . Thank you @kyleberdy for sharing cutting-edge augmentation methods.\n\nThese augmentation methods helped my experiments and increased the score on my Public LB. \n\nThis one describes a model of **CLAHE**, and some of the limits of traditional models: https://towardsdatascience.com/increase-your-face-recognition-models-accuracy-by-improving-face-contrast-a3e71bb6b9fb\n\nThis one describes the opencv application of **adaptive Histogram equalization on grayscale images**: https://opencv-python-tutroals.readthedocs.io/en/latest/py_tutorials/py_imgproc/py_histograms/py_histogram_equalization/py_histogram_equalization.html\n\nSame as above, but in \"layman's terms\" (is anything AI ever in layman's terms): https://www.geeksforgeeks.org/clahe-histogram-eqalization-opencv/\n\n", "votes": 6 }, { "id": 2151239, "postDate": "2023-02-20T00:25:29.807Z", "content": "

Hey Eco,

\n

At first, I tried a few simple augmentations, but at later stage , I discovered that CLAHE works best for this type of problems.
\n(it worked for me)

\n

For me, it raised my LB from 0.53 to 0.55. (~0.02)

\n

I hope that this answered your query.

\n

Thank you!

", "rawMarkdown": "Hey Eco,\n\nAt first, I tried a few simple augmentations, but at later stage , I discovered that CLAHE works best for this type of problems.\n(it worked for me)\n\nFor me, it raised my LB from 0.53 to 0.55. (~0.02)\n\nI hope that this answered your query. \n\nThank you!", "votes": 4, "replies": [ { "id": 2153751, "postDate": "2023-02-21T16:06:52.897Z", "content": "

Thank you for sharing your experience!

\n

LB 0.02 is quite big improvement!

", "rawMarkdown": "Thank you for sharing your experience!\n\nLB 0.02 is quite big improvement!", "votes": 2, "replies": [ { "id": 2158542, "postDate": "2023-02-25T00:10:39.117Z", "content": "

Thank you!

", "rawMarkdown": "Thank you! " } ] }, { "id": 2158174, "postDate": "2023-02-24T17:08:34.793Z", "content": "

Hi Dheeraj could you please tell us how long it will take more time with the CLAHE method?

", "rawMarkdown": "Hi Dheeraj could you please tell us how long it will take more time with the CLAHE method?", "votes": 1, "replies": [ { "id": 2158540, "postDate": "2023-02-25T00:09:08.013Z", "content": "

Hi Hong,

\n

For me the impact on inference time is minimal. It took same time as simple augmentations.

\n

If you wish to speed up, you may utilize CLAHE with the most recent newest DALI nightly which is discussed: here

\n

Thank you!

", "rawMarkdown": "Hi Hong, \n\nFor me the impact on inference time is minimal. It took same time as simple augmentations.\n\nIf you wish to speed up, you may utilize **CLAHE** with the most recent newest DALI nightly which is discussed: [here](https://www.kaggle.com/competitions/rsna-breast-cancer-detection/discussion/384231) \n\nThank you!", "votes": 1 } ] } ] }, { "id": 2150904, "postDate": "2023-02-19T17:27:46.477Z", "content": "

Could you tell me how much the method improved LB?
\nBased on my experience, CLAHE usually does not improve the final performance.

", "rawMarkdown": "Could you tell me how much the method improved LB? \nBased on my experience, CLAHE usually does not improve the final performance.", "votes": 1 } ], "comments": [ { "id": 2151239, "author_name": "Dheeraj Pandey", "author_url": "", "post_date": "2023-02-20T00:25:29.807000", "content": "

Hey Eco,

\n

At first, I tried a few simple augmentations, but at later stage , I discovered that CLAHE works best for this type of problems.
\n(it worked for me)

\n

For me, it raised my LB from 0.53 to 0.55. (~0.02)

\n

I hope that this answered your query.

\n

Thank you!

", "votes": 4, "replies": [ { "id": 2153751, "author_name": "ECO", "author_url": "", "post_date": "2023-02-21T16:06:52.897000", "content": "

Thank you for sharing your experience!

\n

LB 0.02 is quite big improvement!

", "votes": 2, "replies": [ { "id": 2158542, "author_name": "Dheeraj Pandey", "author_url": "", "post_date": "2023-02-25T00:10:39.117000", "content": "

Thank you!

", "votes": 0, "replies": [] } ] }, { "id": 2158174, "author_name": "HongCheng", "author_url": "", "post_date": "2023-02-24T17:08:34.793000", "content": "

Hi Dheeraj could you please tell us how long it will take more time with the CLAHE method?

", "votes": 1, "replies": [ { "id": 2158540, "author_name": "Dheeraj Pandey", "author_url": "", "post_date": "2023-02-25T00:09:08.013000", "content": "

Hi Hong,

\n

For me the impact on inference time is minimal. It took same time as simple augmentations.

\n

If you wish to speed up, you may utilize CLAHE with the most recent newest DALI nightly which is discussed: here

\n

Thank you!

", "votes": 1, "replies": [] } ] } ] }, { "id": 2150904, "author_name": "ECO", "author_url": "", "post_date": "2023-02-19T17:27:46.477000", "content": "

Could you tell me how much the method improved LB?
\nBased on my experience, CLAHE usually does not improve the final performance.

", "votes": 1, "replies": [] } ], "raw_markdown_by_id": { "2150733": "These methods were borrowed from Notebook, which is located at: [here](https://www.kaggle.com/code/kyleberdy/advanced-image-preprocessing-with-clahe) . Thank you @kyleberdy for sharing cutting-edge augmentation methods.\n\nThese augmentation methods helped my experiments and increased the score on my Public LB. \n\nThis one describes a model of **CLAHE**, and some of the limits of traditional models: https://towardsdatascience.com/increase-your-face-recognition-models-accuracy-by-improving-face-contrast-a3e71bb6b9fb\n\nThis one describes the opencv application of **adaptive Histogram equalization on grayscale images**: https://opencv-python-tutroals.readthedocs.io/en/latest/py_tutorials/py_imgproc/py_histograms/py_histogram_equalization/py_histogram_equalization.html\n\nSame as above, but in \"layman's terms\" (is anything AI ever in layman's terms): https://www.geeksforgeeks.org/clahe-histogram-eqalization-opencv/\n\n", "2151239": "Hey Eco,\n\nAt first, I tried a few simple augmentations, but at later stage , I discovered that CLAHE works best for this type of problems.\n(it worked for me)\n\nFor me, it raised my LB from 0.53 to 0.55. (~0.02)\n\nI hope that this answered your query. \n\nThank you!", "2150904": "Could you tell me how much the method improved LB? \nBased on my experience, CLAHE usually does not improve the final performance." } }