I've noticed that some notebooks based on the tile/patch method seem to use the label of one WSI as the label for tiles/patches that come from the same WSI during the validation phase as well, is this a really reasonable approach? If the tiles/patches chosen for validation do not actually have the ability to represent WSI features, could such a validation process be misleading or introduce noise into the model? I am still very confused after a long time of checking online…
", "messages": [ { "id": 2511234, "postDate": "2023-11-03T14:40:23.073Z", "content": "I've noticed that some notebooks based on the tile/patch method seem to use the label of one WSI as the label for tiles/patches that come from the same WSI during the validation phase as well, is this a really reasonable approach? If the tiles/patches chosen for validation do not actually have the ability to represent WSI features, could such a validation process be misleading or introduce noise into the model? I am still very confused after a long time of checking online…
", "rawMarkdown": "I've noticed that some notebooks based on the tile/patch method seem to use the label of one WSI as the label for tiles/patches that come from the same WSI during the validation phase as well, is this a really reasonable approach? If the tiles/patches chosen for validation do not actually have the ability to represent WSI features, could such a validation process be misleading or introduce noise into the model? I am still very confused after a long time of checking online...", "votes": 4 }, { "id": 2511935, "postDate": "2023-11-04T04:39:29.930Z", "content": "How many tiles from a single image you have pick in test image data , when submitting notebook. Please provide me the below details.
\n1) no. of tiles for single image 2) batch size 3) image size
\nfor your model . It will be helpful for me.
Randomly picking tiles and hoping they represent the label is not a good method. But until we get the annotations mentioned in this thread -> https://www.kaggle.com/competitions/UBC-OCEAN/discussion/452360, I suspect picking proper tiles will be the challenge in this competition.
", "rawMarkdown": "Randomly picking tiles and hoping they represent the label is not a good method. But until we get the annotations mentioned in this thread -> https://www.kaggle.com/competitions/UBC-OCEAN/discussion/452360, I suspect picking proper tiles will be *the* challenge in this competition.", "replies": [ { "id": 2511806, "postDate": "2023-11-04T01:27:48.517Z", "content": "Yeah, I totally agree with you. Are there any good ideas on tile/patch selection? After all, we don't have tile/patch level labels available so far😂
", "rawMarkdown": "Yeah, I totally agree with you. Are there any good ideas on tile/patch selection? After all, we don't have tile/patch level labels available so far😂", "replies": [ { "id": 2518049, "postDate": "2023-11-09T02:20:27.187Z", "content": "hope it's helpful: https://www.kaggle.com/competitions/UBC-OCEAN/discussion/451902
", "rawMarkdown": "hope it's helpful: https://www.kaggle.com/competitions/UBC-OCEAN/discussion/451902" } ] } ] } ], "comments": [ { "id": 2511935, "author_name": "sunil thite", "author_url": "", "post_date": "2023-11-04T04:39:29.930000", "content": "How many tiles from a single image you have pick in test image data , when submitting notebook. Please provide me the below details.
\n1) no. of tiles for single image 2) batch size 3) image size
\nfor your model . It will be helpful for me.
Randomly picking tiles and hoping they represent the label is not a good method. But until we get the annotations mentioned in this thread -> https://www.kaggle.com/competitions/UBC-OCEAN/discussion/452360, I suspect picking proper tiles will be the challenge in this competition.
", "votes": 0, "replies": [ { "id": 2511806, "author_name": "阳光开朗大男孩", "author_url": "", "post_date": "2023-11-04T01:27:48.517000", "content": "Yeah, I totally agree with you. Are there any good ideas on tile/patch selection? After all, we don't have tile/patch level labels available so far😂
", "votes": 0, "replies": [ { "id": 2518049, "author_name": "Time Master", "author_url": "", "post_date": "2023-11-09T02:20:27.187000", "content": "hope it's helpful: https://www.kaggle.com/competitions/UBC-OCEAN/discussion/451902
", "votes": 0, "replies": [] } ] } ] } ], "raw_markdown_by_id": { "2511234": "I've noticed that some notebooks based on the tile/patch method seem to use the label of one WSI as the label for tiles/patches that come from the same WSI during the validation phase as well, is this a really reasonable approach? If the tiles/patches chosen for validation do not actually have the ability to represent WSI features, could such a validation process be misleading or introduce noise into the model? I am still very confused after a long time of checking online...", "2511935": "How many tiles from a single image you have pick in test image data , when submitting notebook. Please provide me the below details.\n1) no. of tiles for single image 2) batch size 3) image size \nfor your model . It will be helpful for me.", "2511799": "Randomly picking tiles and hoping they represent the label is not a good method. But until we get the annotations mentioned in this thread -> https://www.kaggle.com/competitions/UBC-OCEAN/discussion/452360, I suspect picking proper tiles will be *the* challenge in this competition." } }