Hello,
\n\nFrom doing some basic EDA to me it seems that the differentiating factor for wheat is not it's colour, more about texture. To reduce computation requirements could we reliably change all images to grayscale before feeding into a model?
", "messages": [ { "id": "836908", "postDate": "05/07/2020 11:03:38", "content": "Hello,
\n\nFrom doing some basic EDA to me it seems that the differentiating factor for wheat is not it's colour, more about texture. To reduce computation requirements could we reliably change all images to grayscale before feeding into a model?
", "rawMarkdown": "Hello,\n\nFrom doing some basic EDA to me it seems that the differentiating factor for wheat is not it's colour, more about texture. To reduce computation requirements could we reliably change all images to grayscale before feeding into a model?", "votes": null }, { "id": "837226", "postDate": "05/07/2020 16:11:37", "content": "I mean, you're definitely allowed to. Nothing about that goes against the competition rules.
\n\nHowever, it doesn't seem like it would help much. The training set we are provided with is relatively small and shouldn't be too computationally taxing to use in color.
", "rawMarkdown": "I mean, you're definitely allowed to. Nothing about that goes against the competition rules.\n\nHowever, it doesn't seem like it would help much. The training set we are provided with is relatively small and shouldn't be too computationally taxing to use in color.", "votes": null }, { "id": "837286", "postDate": "05/07/2020 16:58:25", "content": "I should have been clearer. My question was more of a \"is it useful\" rather than an \"is it allowed\". However, you answered that too. :) I am new to objects/pictures in ML so I have little knowledge of what is a small or large dataset. To me it seemed big, maybe I am in for a shock then when I discover truly large datasets! Many thanks.
", "rawMarkdown": "I should have been clearer. My question was more of a \"is it useful\" rather than an \"is it allowed\". However, you answered that too. :) I am new to objects/pictures in ML so I have little knowledge of what is a small or large dataset. To me it seemed big, maybe I am in for a shock then when I discover truly large datasets! Many thanks.", "votes": null }, { "id": "837337", "postDate": "05/07/2020 17:41:19", "content": "Most computer vision competitions have over 1GB of data. For example, another competition on detecting pancreatic cancer right now has nearly 400GB of image data. In this competition, we only have 600MB (compressed). So yes, it is a very reasonable dataset size, especially considering the power of Kaggle notebooks.
\n\nGood luck!
", "rawMarkdown": "Most computer vision competitions have over 1GB of data. For example, another competition on detecting pancreatic cancer right now has nearly 400GB of image data. In this competition, we only have 600MB (compressed). So yes, it is a very reasonable dataset size, especially considering the power of Kaggle notebooks.\n\nGood luck!", "votes": null } ], "comments": [ { "id": 837226, "author_name": "matthewmasters", "author_url": "", "post_date": "05/07/2020 16:11:37", "content": "I mean, you're definitely allowed to. Nothing about that goes against the competition rules.
\n\nHowever, it doesn't seem like it would help much. The training set we are provided with is relatively small and shouldn't be too computationally taxing to use in color.
", "votes": null, "replies": [ { "id": 837286, "author_name": "blueturtle", "author_url": "", "post_date": "05/07/2020 16:58:25", "content": "I should have been clearer. My question was more of a \"is it useful\" rather than an \"is it allowed\". However, you answered that too. :) I am new to objects/pictures in ML so I have little knowledge of what is a small or large dataset. To me it seemed big, maybe I am in for a shock then when I discover truly large datasets! Many thanks.
", "votes": null, "replies": [] }, { "id": 837337, "author_name": "matthewmasters", "author_url": "", "post_date": "05/07/2020 17:41:19", "content": "Most computer vision competitions have over 1GB of data. For example, another competition on detecting pancreatic cancer right now has nearly 400GB of image data. In this competition, we only have 600MB (compressed). So yes, it is a very reasonable dataset size, especially considering the power of Kaggle notebooks.
\n\nGood luck!
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "836908": "Hello,\n\nFrom doing some basic EDA to me it seems that the differentiating factor for wheat is not it's colour, more about texture. To reduce computation requirements could we reliably change all images to grayscale before feeding into a model?", "837226": "I mean, you're definitely allowed to. Nothing about that goes against the competition rules.\n\nHowever, it doesn't seem like it would help much. The training set we are provided with is relatively small and shouldn't be too computationally taxing to use in color.", "837286": "I should have been clearer. My question was more of a \"is it useful\" rather than an \"is it allowed\". However, you answered that too. :) I am new to objects/pictures in ML so I have little knowledge of what is a small or large dataset. To me it seemed big, maybe I am in for a shock then when I discover truly large datasets! Many thanks.", "837337": "Most computer vision competitions have over 1GB of data. For example, another competition on detecting pancreatic cancer right now has nearly 400GB of image data. In this competition, we only have 600MB (compressed). So yes, it is a very reasonable dataset size, especially considering the power of Kaggle notebooks.\n\nGood luck!" }, "source": "meta" }