![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F3982638%2F0467882d2af49259f537df7aa8e59293%2F2cd038d5c85feb0a8da44407e4fb6a5a_mask.tiff.png?generation=1587754711288856&alt=media\")
Recently i have trained my Resnext50 model on images dataset and it has quite decent CV of 0.713 and LB: 0.64 (kernel is attached here). I thought it will be a great idea to try training on the image-mask-overlay.
So, here i have created a dataset of image-mask overlay images resized and saved into pngs.\nhttps://www.kaggle.com/rohitsingh9990/image-mask-overlay-512x512.
\n\nThe kernel i used to create this dataset is https://www.kaggle.com/rohitsingh9990/panda-resize-and-save-image-mask-overlay/notebook?scriptVersionId=32614926
", "messages": [ { "id": 819687, "postDate": "2020-04-24T19:08:01.937Z", "content": "Also while going through the images, i found few images with markers as mentioned here.
\n\n\nNote that slightly different procedures were in place for the images used in the test set than the training set. Some of the training set images have stray pen marks on them, but the test set slides are free of pen marks.\n
![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F3982638%2F935b9b3e17ae7c818b4d2c88422b3700%2F8b90a3312d9fc702a9ec8236e23065c7_mask.tiff.png?generation=1587755200515217&alt=media\")
Maybe we should try to clean these images, as test set images do not have pen markers, it will definitely boost our score on private dataset.
", "rawMarkdown": "Also while going through the images, i found few images with markers as mentioned [here](https://www.kaggle.com/c/prostate-cancer-grade-assessment/data).\n\n```\nNote that slightly different procedures were in place for the images used in the test set than the training set. Some of the training set images have stray pen marks on them, but the test set slides are free of pen marks.\n```\n\n \n\n\n\nMaybe we should try to clean these images, as test set images do not have pen markers, it will definitely boost our score on private dataset.", "votes": 1, "replies": [ { "id": 819843, "postDate": "2020-04-24T23:31:32.437Z", "content": "Agreed. Each data provider also has a different method for labeling. I brought up some ideas on how to standardize this dataset here.
", "rawMarkdown": "Agreed. Each data provider also has a different method for labeling. I brought up some ideas on how to standardize this dataset [here](https://www.kaggle.com/c/prostate-cancer-grade-assessment/discussion/145742)." } ] }, { "id": 819681, "postDate": "2020-04-24T19:03:46.970Z", "content": "Recently i have trained my Resnext50 model on images dataset and it has quite decent CV of 0.713 and LB: 0.64 (kernel is attached here). I thought it will be a great idea to try training on the image-mask-overlay.
So, here i have created a dataset of image-mask overlay images resized and saved into pngs.\nhttps://www.kaggle.com/rohitsingh9990/image-mask-overlay-512x512.
\n\nThe kernel i used to create this dataset is https://www.kaggle.com/rohitsingh9990/panda-resize-and-save-image-mask-overlay/notebook?scriptVersionId=32614926
", "rawMarkdown": "Recently i have trained my Resnext50 model on images dataset and it has quite decent CV of 0.713 and LB: 0.64 (kernel is attached [here](https://www.kaggle.com/rohitsingh9990/panda-resnext-inference)). I thought it will be a great idea to try training on the `image-mask-overlay`. \n\nSo, here i have created a dataset of image-mask overlay images resized and saved into pngs.\nhttps://www.kaggle.com/rohitsingh9990/image-mask-overlay-512x512.\n\nThe kernel i used to create this dataset is https://www.kaggle.com/rohitsingh9990/panda-resize-and-save-image-mask-overlay/notebook?scriptVersionId=32614926\n\n\n", "votes": 1 } ], "comments": [ { "id": 819687, "author_name": "Dracarys", "author_url": "", "post_date": "2020-04-24T19:08:01.937000", "content": "Also while going through the images, i found few images with markers as mentioned here.
\n\n\nNote that slightly different procedures were in place for the images used in the test set than the training set. Some of the training set images have stray pen marks on them, but the test set slides are free of pen marks.\n
Maybe we should try to clean these images, as test set images do not have pen markers, it will definitely boost our score on private dataset.
", "votes": 1, "replies": [ { "id": 819843, "author_name": "Matt", "author_url": "", "post_date": "2020-04-24T23:31:32.437000", "content": "Agreed. Each data provider also has a different method for labeling. I brought up some ideas on how to standardize this dataset here.
", "votes": 0, "replies": [] } ] } ], "raw_markdown_by_id": { "819687": "Also while going through the images, i found few images with markers as mentioned [here](https://www.kaggle.com/c/prostate-cancer-grade-assessment/data).\n\n```\nNote that slightly different procedures were in place for the images used in the test set than the training set. Some of the training set images have stray pen marks on them, but the test set slides are free of pen marks.\n```\n\n \n\n\n\nMaybe we should try to clean these images, as test set images do not have pen markers, it will definitely boost our score on private dataset.", "819681": "Recently i have trained my Resnext50 model on images dataset and it has quite decent CV of 0.713 and LB: 0.64 (kernel is attached [here](https://www.kaggle.com/rohitsingh9990/panda-resnext-inference)). I thought it will be a great idea to try training on the `image-mask-overlay`. \n\nSo, here i have created a dataset of image-mask overlay images resized and saved into pngs.\nhttps://www.kaggle.com/rohitsingh9990/image-mask-overlay-512x512.\n\nThe kernel i used to create this dataset is https://www.kaggle.com/rohitsingh9990/panda-resize-and-save-image-mask-overlay/notebook?scriptVersionId=32614926\n\n\n" } }