Hi,
\nI'm new to this type of competitions so my question might be trivial.
\nATM, I'm using data that other users have prepared so skipping the data processing step in my notebook. Do we need to have all the processing of data (e.g., dcm to png conversion) in the submitted notebook?
\nCheers
You'll want to process the images during infer similar to the way they are done in your training data. You don't need to create PNG or JPG though, just extract the pixel data and normalize it to the same bit depth of your training images (8 or 16).
", "rawMarkdown": "You'll want to process the images during infer similar to the way they are done in your training data. You don't need to create PNG or JPG though, just extract the pixel data and normalize it to the same bit depth of your training images (8 or 16).", "votes": 1, "replies": [ { "id": 2054212, "postDate": "2022-12-04T00:19:45.997Z", "content": "how about the training part? we need to reduce the size of the images and save them before starting the training phase. Should that step occur in the notebook? that step is pretty time consuming.
", "rawMarkdown": "how about the training part? we need to reduce the size of the images and save them before starting the training phase. Should that step occur in the notebook? that step is pretty time consuming. " }, { "id": 2054222, "postDate": "2022-12-04T00:37:44.420Z", "content": "Training should be done separately from infer.
", "rawMarkdown": "Training should be done separately from infer." }, { "id": 2054243, "postDate": "2022-12-04T01:05:36.720Z", "content": "so you just submit the model object with the notebook and not the actual training part. Am I correct?
", "rawMarkdown": "so you just submit the model object with the notebook and not the actual training part. Am I correct? " }, { "id": 2054250, "postDate": "2022-12-04T01:16:56.620Z", "content": "Yep. Train your model locally, on colab or here, then load it into a dataset. Reference your trained model in your inference notebook.
", "rawMarkdown": "Yep. Train your model locally, on colab or here, then load it into a dataset. Reference your trained model in your inference notebook.", "votes": 1 } ] }, { "id": 2053804, "postDate": "2022-12-03T16:14:12.777Z", "content": "Hi,
\nI'm new to this type of competitions so my question might be trivial.
\nATM, I'm using data that other users have prepared so skipping the data processing step in my notebook. Do we need to have all the processing of data (e.g., dcm to png conversion) in the submitted notebook?
\nCheers
You'll want to process the images during infer similar to the way they are done in your training data. You don't need to create PNG or JPG though, just extract the pixel data and normalize it to the same bit depth of your training images (8 or 16).
", "votes": 1, "replies": [ { "id": 2054212, "author_name": "Dance_with_Data", "author_url": "", "post_date": "2022-12-04T00:19:45.997000", "content": "how about the training part? we need to reduce the size of the images and save them before starting the training phase. Should that step occur in the notebook? that step is pretty time consuming.
", "votes": 0, "replies": [] }, { "id": 2054222, "author_name": "David Roberts", "author_url": "", "post_date": "2022-12-04T00:37:44.420000", "content": "Training should be done separately from infer.
", "votes": 0, "replies": [] }, { "id": 2054243, "author_name": "Dance_with_Data", "author_url": "", "post_date": "2022-12-04T01:05:36.720000", "content": "so you just submit the model object with the notebook and not the actual training part. Am I correct?
", "votes": 0, "replies": [] }, { "id": 2054250, "author_name": "David Roberts", "author_url": "", "post_date": "2022-12-04T01:16:56.620000", "content": "Yep. Train your model locally, on colab or here, then load it into a dataset. Reference your trained model in your inference notebook.
", "votes": 1, "replies": [] } ] } ], "raw_markdown_by_id": { "2054206": "You'll want to process the images during infer similar to the way they are done in your training data. You don't need to create PNG or JPG though, just extract the pixel data and normalize it to the same bit depth of your training images (8 or 16).", "2053804": "Hi, \nI'm new to this type of competitions so my question might be trivial. \nATM, I'm using data that other users have prepared so skipping the data processing step in my notebook. Do we need to have all the processing of data (e.g., dcm to png conversion) in the submitted notebook? \nCheers" } }