can anyone can share the starter code to build databunch in the fastai? \nI'm able to create SegmentationsItemsList but not successful in labelling with the values in 'labels' column for each image correct to make databunch.
\n\nThanks in advance.
", "messages": [ { "id": "595492", "postDate": "08/09/2019 09:33:10", "content": "can anyone can share the starter code to build databunch in the fastai? \nI'm able to create SegmentationsItemsList but not successful in labelling with the values in 'labels' column for each image correct to make databunch.
\n\nThanks in advance.
", "rawMarkdown": "can anyone can share the starter code to build databunch in the fastai? \nI'm able to create SegmentationsItemsList but not successful in labelling with the values in 'labels' column for each image correct to make databunch. \n\nThanks in advance.", "votes": null }, { "id": "596170", "postDate": "08/10/2019 08:43:56", "content": "I'm not sure if it helps, but you can see here how I've tried to create fastai ImageDataBunch of MNIST.
", "rawMarkdown": "I'm not sure if it helps, but you can see [here](https://www.kaggle.com/vochicong/fast-ai-mnist?scriptVersionId=17945512#Data-augmentation) how I've tried to create fastai ImageDataBunch of MNIST.", "votes": null }, { "id": "597110", "postDate": "08/11/2019 20:56:34", "content": "In this challenge, you first have to localize the characters and then you have to classify them. Unfortunately, you can't do finetuned localization in Fastai. Therefore you have to choose another approach with another package
", "rawMarkdown": "In this challenge, you first have to localize the characters and then you have to classify them. Unfortunately, you can't do finetuned localization in Fastai. Therefore you have to choose another approach with another package", "votes": null }, { "id": "628720", "postDate": "09/17/2019 19:13:47", "content": "The way I approached it was to write code to crop each test image using the bounding box coordinates to create a dataset. I'd recommend using the VIPS library + cython, to speed up the task.
\n\nWord of warning, just over 20% of the classes have only a single image which fast.ai does not play well with when creating test/Val sets so you will have to manually separate your data into training/test sets, I simply moved all classes with only one image into the test set.
", "rawMarkdown": "The way I approached it was to write code to crop each test image using the bounding box coordinates to create a dataset. I'd recommend using the VIPS library + cython, to speed up the task.\n\nWord of warning, just over 20% of the classes have only a single image which fast.ai does not play well with when creating test/Val sets so you will have to manually separate your data into training/test sets, I simply moved all classes with only one image into the test set.", "votes": null } ], "comments": [ { "id": 596170, "author_name": "vochicong", "author_url": "", "post_date": "08/10/2019 08:43:56", "content": "I'm not sure if it helps, but you can see here how I've tried to create fastai ImageDataBunch of MNIST.
", "votes": null, "replies": [] }, { "id": 597110, "author_name": "christianwallenwein", "author_url": "", "post_date": "08/11/2019 20:56:34", "content": "In this challenge, you first have to localize the characters and then you have to classify them. Unfortunately, you can't do finetuned localization in Fastai. Therefore you have to choose another approach with another package
", "votes": null, "replies": [] }, { "id": 628720, "author_name": "tripleaaa", "author_url": "", "post_date": "09/17/2019 19:13:47", "content": "The way I approached it was to write code to crop each test image using the bounding box coordinates to create a dataset. I'd recommend using the VIPS library + cython, to speed up the task.
\n\nWord of warning, just over 20% of the classes have only a single image which fast.ai does not play well with when creating test/Val sets so you will have to manually separate your data into training/test sets, I simply moved all classes with only one image into the test set.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "595492": "can anyone can share the starter code to build databunch in the fastai? \nI'm able to create SegmentationsItemsList but not successful in labelling with the values in 'labels' column for each image correct to make databunch. \n\nThanks in advance.", "596170": "I'm not sure if it helps, but you can see [here](https://www.kaggle.com/vochicong/fast-ai-mnist?scriptVersionId=17945512#Data-augmentation) how I've tried to create fastai ImageDataBunch of MNIST.", "597110": "In this challenge, you first have to localize the characters and then you have to classify them. Unfortunately, you can't do finetuned localization in Fastai. Therefore you have to choose another approach with another package", "628720": "The way I approached it was to write code to crop each test image using the bounding box coordinates to create a dataset. I'd recommend using the VIPS library + cython, to speed up the task.\n\nWord of warning, just over 20% of the classes have only a single image which fast.ai does not play well with when creating test/Val sets so you will have to manually separate your data into training/test sets, I simply moved all classes with only one image into the test set." }, "source": "meta" }