![\"enter](\"https://storage.googleapis.com/kaggle-forum-message-attachments/inbox/113660/06662aea40ac0139cd85bff303c2ec55/4A0392E6-BC2F-4283-ADA6-2D00C850BFC3.gif\")
detect arrow as attention foucus
\n\nocr to recognize words
", "messages": [ { "id": 407024, "postDate": "2018-10-20T08:14:38.233Z", "content": "detect arrow as attention foucus
\n\nocr to recognize words
", "rawMarkdown": "1. group drawing by into same country and time\n2. maybe you can estimate the time taken to draw the doodle?\n\n---\n\ndetect arrow as attention foucus\n\n---\nocr to recognize words\n", "votes": 23 }, { "id": 408588, "postDate": "2018-10-23T05:55:34.057Z", "content": "https://medium.com/@enjalot/machine-learning-for-visualization-927a9dff1cab
\n\n
![\"enter](\"https://storage.googleapis.com/kaggle-forum-message-attachments/inbox/113660/7c1a291ca647d640adc18735e651e4df/0B7D1265-252B-449D-8F2A-59458BA4A9E0.png\")
order of the stroke:
\n\nif you play with quick draw before, you will know that the game will \"prompt you until you get it right\". The last stroke may be one that will differentiate the object from confusing one.
\n\nshould we give more attention to it?
", "rawMarkdown": "order of the stroke:\n\nif you play with quick draw before, you will know that the game will \"prompt you until you get it right\". The last stroke may be one that will differentiate the object from confusing one.\n\nshould we give more attention to it?", "votes": 5, "replies": [ { "id": 407783, "postDate": "2018-10-21T18:40:16.653Z", "content": "In my understanding a stroke is more useless if it's been drawn latter. Because wrong predictions often happen on those with a lot strokes which make them a mess. I played it before and I have to keep adding strokes until the last second. So I think last strokes are not that useful.
", "rawMarkdown": "In my understanding a stroke is more useless if it's been drawn latter. Because wrong predictions often happen on those with a lot strokes which make them a mess. I played it before and I have to keep adding strokes until the last second. So I think last strokes are not that useful.", "votes": 1 } ] }, { "id": 407424, "postDate": "2018-10-21T06:13:23.803Z", "content": "For image based cnn, sketch stroke are thin, not so invariant feature. To improve, image gradient field can be used
\n\n![\"enter](\"https://storage.googleapis.com/kaggle-forum-message-attachments/inbox/113660/7fd11390e33b78d88ba89e0c6a6b6a52/E33C3835-3913-4531-B416-3DE4428BB229.png\")
We are interested in user reaction time, how long the user takes to think and draw the object. Drawing time can be estimated from the number of strokes(number of time pen leaves the paper) and number of points per strokes, which is available in data.
\n\nIf the user is unfamiliar with the word, eg non native english speaker, the reaction time is longer
", "rawMarkdown": "We are interested in user reaction time, how long the user takes to think and draw the object. Drawing time can be estimated from the number of strokes(number of time pen leaves the paper) and number of points per strokes, which is available in data.\n\nIf the user is unfamiliar with the word, eg non native english speaker, the reaction time is longer", "votes": 3 }, { "id": 412390, "postDate": "2018-10-30T05:57:13.510Z", "content": "One weird idea:
\n\nTrain an estimator that predict the score gain/ loss if i interchange the first and second place prediction.
\n\nE.g original prediction: baseball, baseball bat, xxx
\n\nInterchanged prediction: baseball bat, baseball, xxx
", "rawMarkdown": "One weird idea:\n\nTrain an estimator that predict the score gain/ loss if i interchange the first and second place prediction.\n\n\nE.g original prediction: baseball, baseball bat, xxx\n\nInterchanged prediction: baseball bat, baseball, xxx", "votes": 1 }, { "id": 408594, "postDate": "2018-10-23T06:04:14.757Z", "content": "\n\n", "rawMarkdown": "http://louistiao.me/posts/notebooks/visualizing-the-latent-space-of-vector-drawings-from-the-google-quickdraw-dataset-with-sketchrnn-pca-and-t-sne/\n\n\nhttps://gweb-cloudblog-publish.appspot.com/products/gcp/drawings-in-the-cloud-introducing-the-quick-draw-dataset/amp/", "votes": 1 }, { "id": 415437, "postDate": "2018-11-05T05:42:35.823Z", "content": "Has anyone successfully incorporated Country or Number of Strokes info? They didn't add anything to my model, made it a little bit worse.
", "rawMarkdown": "Has anyone successfully incorporated Country or Number of Strokes info? They didn't add anything to my model, made it a little bit worse. ", "votes": 2 }, { "id": 1076126, "postDate": "2020-11-12T09:23:51.767Z", "content": "I write a script for a beginner who wants to Learn OCR Step by Step. Just go and watch and support me if you like:-
\nhttps://www.kaggle.com/naim99/ocr-text-recognition-ocr-space-api-tesseract
", "rawMarkdown": "I write a script for a beginner who wants to Learn OCR Step by Step. Just go and watch and support me if you like:-\n\nhttps://www.kaggle.com/naim99/ocr-text-recognition-ocr-space-api-tesseract" }, { "id": 416317, "postDate": "2018-11-06T14:33:52.587Z", "content": "sparse convolution:
\n\nhttps://github.com/facebookresearch/SparseConvNet
", "rawMarkdown": "sparse convolution:\n\n\nhttps://github.com/facebookresearch/SparseConvNet\n\n" }, { "id": 416289, "postDate": "2018-11-06T14:13:42.743Z", "content": "visualisation:\nhttp://vallandingham.me/quickdraw/
", "rawMarkdown": "visualisation:\nhttp://vallandingham.me/quickdraw/" }, { "id": 415583, "postDate": "2018-11-05T10:59:29.623Z", "content": "possible data augmentation:
\n\nhttps://panly099.github.io/skSyn.html
\n\nhttps://github.com/panly099/sketchSynthesis
\n\nFree-hand sketch synthesis with deformable stroke models
\n\nhttps://github.com/val-iisc/sketch-parse
\n\n\"SketchParse: Towards Rich Descriptions For Poorly Drawn Sketches Using Multi-Task Deep Networks\", ACM Multimedia (ACMMM) 2017.
\n\nsketch parsing/ vectorisation : can be used as additional supervision signal in multi-task setting?
\n\nSemantic Segmentation for Line Drawing Vectorization Using\nNeural Networks
\n\nhttp://www.byungsoo.me/project/vectornet/paper.pdf\nhttps://www.microsoft.com/en-us/research/wp-content/uploads/2012/01/16-2012-eccv-sketch-segmenation.pdf
", "rawMarkdown": "possible data augmentation:\n\nhttps://panly099.github.io/skSyn.html\n\nhttps://github.com/panly099/sketchSynthesis\n\nFree-hand sketch synthesis with deformable stroke models\n\n\n---\nhttps://github.com/val-iisc/sketch-parse\n\n \"SketchParse: Towards Rich Descriptions For Poorly Drawn Sketches Using Multi-Task Deep Networks\", ACM Multimedia (ACMMM) 2017.\n\n---\n\nsketch parsing/ vectorisation : can be used as additional supervision signal in multi-task setting?\n\nSemantic Segmentation for Line Drawing Vectorization Using\nNeural Networks\n\n\nhttp://www.byungsoo.me/project/vectornet/paper.pdf\nhttps://www.microsoft.com/en-us/research/wp-content/uploads/2012/01/16-2012-eccv-sketch-segmenation.pdf" }, { "id": 407515, "postDate": "2018-10-21T11:31:09.423Z", "content": "The time to draw the doodle and other timing details can be easily extracted from the raw data, which contains [x, y, t] coordinates of pen/mouse strokes.
", "rawMarkdown": "The time to draw the doodle and other timing details can be easily extracted from the raw data, which contains [x, y, t] coordinates of pen/mouse strokes." }, { "id": 407406, "postDate": "2018-10-21T05:40:07.300Z", "content": "example where ocr to recognize words can be important:
\n\nnaming of file: (top prob after softmax), (key id), (top 3 class)
\n\n![\"enter](\"https://storage.googleapis.com/kaggle-forum-message-attachments/inbox/113660/cbee46ed9e840de5c921360fc30870fb/0.15_9997386404410976_owl_frog_kangaroo.png\")
\n ![\"enter](\"https://storage.googleapis.com/kaggle-forum-message-attachments/inbox/113660/fa3f80258d1e1f53d3ee8df574c3788c/0.25_9722430504684690_sea_turtle_frog_teddy-bear.png\")
In the few examples where drawings have been annotated (words or arrows) it seems obvious that the annotations are added last. It might be possible to extract the strokes used for labels during preprocessing.
", "rawMarkdown": "In the few examples where drawings have been annotated (words or arrows) it seems obvious that the annotations are added last. It might be possible to extract the strokes used for labels during preprocessing." } ] } ], "comments": [ { "id": 408588, "author_name": "hengck23", "author_url": "", "post_date": "2018-10-23T05:55:34.057000", "content": "https://medium.com/@enjalot/machine-learning-for-visualization-927a9dff1cab
\n\n
order of the stroke:
\n\nif you play with quick draw before, you will know that the game will \"prompt you until you get it right\". The last stroke may be one that will differentiate the object from confusing one.
\n\nshould we give more attention to it?
", "votes": 5, "replies": [ { "id": 407783, "author_name": "lulala", "author_url": "", "post_date": "2018-10-21T18:40:16.653000", "content": "In my understanding a stroke is more useless if it's been drawn latter. Because wrong predictions often happen on those with a lot strokes which make them a mess. I played it before and I have to keep adding strokes until the last second. So I think last strokes are not that useful.
", "votes": 1, "replies": [] } ] }, { "id": 407424, "author_name": "hengck23", "author_url": "", "post_date": "2018-10-21T06:13:23.803000", "content": "For image based cnn, sketch stroke are thin, not so invariant feature. To improve, image gradient field can be used
\n\n
We are interested in user reaction time, how long the user takes to think and draw the object. Drawing time can be estimated from the number of strokes(number of time pen leaves the paper) and number of points per strokes, which is available in data.
\n\nIf the user is unfamiliar with the word, eg non native english speaker, the reaction time is longer
", "votes": 3, "replies": [] }, { "id": 412390, "author_name": "hengck23", "author_url": "", "post_date": "2018-10-30T05:57:13.510000", "content": "One weird idea:
\n\nTrain an estimator that predict the score gain/ loss if i interchange the first and second place prediction.
\n\nE.g original prediction: baseball, baseball bat, xxx
\n\nInterchanged prediction: baseball bat, baseball, xxx
", "votes": 1, "replies": [] }, { "id": 408594, "author_name": "hengck23", "author_url": "", "post_date": "2018-10-23T06:04:14.757000", "content": "\n\n", "votes": 1, "replies": [] }, { "id": 415437, "author_name": "HuyenNguyen", "author_url": "", "post_date": "2018-11-05T05:42:35.823000", "content": "Has anyone successfully incorporated Country or Number of Strokes info? They didn't add anything to my model, made it a little bit worse.
", "votes": 2, "replies": [] }, { "id": 1076126, "author_name": "Naim Mhedhbi", "author_url": "", "post_date": "2020-11-12T09:23:51.767000", "content": "I write a script for a beginner who wants to Learn OCR Step by Step. Just go and watch and support me if you like:-
\nhttps://www.kaggle.com/naim99/ocr-text-recognition-ocr-space-api-tesseract
", "votes": 0, "replies": [] }, { "id": 416317, "author_name": "hengck23", "author_url": "", "post_date": "2018-11-06T14:33:52.587000", "content": "sparse convolution:
\n\nhttps://github.com/facebookresearch/SparseConvNet
", "votes": 0, "replies": [] }, { "id": 416289, "author_name": "hengck23", "author_url": "", "post_date": "2018-11-06T14:13:42.743000", "content": "visualisation:\nhttp://vallandingham.me/quickdraw/
", "votes": 0, "replies": [] }, { "id": 415583, "author_name": "hengck23", "author_url": "", "post_date": "2018-11-05T10:59:29.623000", "content": "possible data augmentation:
\n\nhttps://panly099.github.io/skSyn.html
\n\nhttps://github.com/panly099/sketchSynthesis
\n\nFree-hand sketch synthesis with deformable stroke models
\n\nhttps://github.com/val-iisc/sketch-parse
\n\n\"SketchParse: Towards Rich Descriptions For Poorly Drawn Sketches Using Multi-Task Deep Networks\", ACM Multimedia (ACMMM) 2017.
\n\nsketch parsing/ vectorisation : can be used as additional supervision signal in multi-task setting?
\n\nSemantic Segmentation for Line Drawing Vectorization Using\nNeural Networks
\n\nhttp://www.byungsoo.me/project/vectornet/paper.pdf\nhttps://www.microsoft.com/en-us/research/wp-content/uploads/2012/01/16-2012-eccv-sketch-segmenation.pdf
", "votes": 0, "replies": [] }, { "id": 407515, "author_name": "Paul Jurczak", "author_url": "", "post_date": "2018-10-21T11:31:09.423000", "content": "The time to draw the doodle and other timing details can be easily extracted from the raw data, which contains [x, y, t] coordinates of pen/mouse strokes.
", "votes": 0, "replies": [] }, { "id": 407406, "author_name": "hengck23", "author_url": "", "post_date": "2018-10-21T05:40:07.300000", "content": "example where ocr to recognize words can be important:
\n\nnaming of file: (top prob after softmax), (key id), (top 3 class)
\n\n\n
In the few examples where drawings have been annotated (words or arrows) it seems obvious that the annotations are added last. It might be possible to extract the strokes used for labels during preprocessing.
", "votes": 0, "replies": [] } ] } ], "raw_markdown_by_id": { "407024": "1. group drawing by into same country and time\n2. maybe you can estimate the time taken to draw the doodle?\n\n---\n\ndetect arrow as attention foucus\n\n---\nocr to recognize words\n", "408588": "https://medium.com/@enjalot/machine-learning-for-visualization-927a9dff1cab\n\n\n ![enter image description here][1]\n\n ![enter image description here][2]\n\n\n [1]: https://storage.googleapis.com/kaggle-forum-message-attachments/inbox/113660/06662aea40ac0139cd85bff303c2ec55/4A0392E6-BC2F-4283-ADA6-2D00C850BFC3.gif\n [2]: https://storage.googleapis.com/kaggle-forum-message-attachments/inbox/113660/7c1a291ca647d640adc18735e651e4df/0B7D1265-252B-449D-8F2A-59458BA4A9E0.png", "407480": "order of the stroke:\n\nif you play with quick draw before, you will know that the game will \"prompt you until you get it right\". The last stroke may be one that will differentiate the object from confusing one.\n\nshould we give more attention to it?", "407424": "For image based cnn, sketch stroke are thin, not so invariant feature. To improve, image gradient field can be used\n\n\n ![enter image description here][1]\n\n\n [1]: https://storage.googleapis.com/kaggle-forum-message-attachments/inbox/113660/7fd11390e33b78d88ba89e0c6a6b6a52/E33C3835-3913-4531-B416-3DE4428BB229.png", "407419": "We are interested in user reaction time, how long the user takes to think and draw the object. Drawing time can be estimated from the number of strokes(number of time pen leaves the paper) and number of points per strokes, which is available in data.\n\nIf the user is unfamiliar with the word, eg non native english speaker, the reaction time is longer", "412390": "One weird idea:\n\nTrain an estimator that predict the score gain/ loss if i interchange the first and second place prediction.\n\n\nE.g original prediction: baseball, baseball bat, xxx\n\nInterchanged prediction: baseball bat, baseball, xxx", "408594": "http://louistiao.me/posts/notebooks/visualizing-the-latent-space-of-vector-drawings-from-the-google-quickdraw-dataset-with-sketchrnn-pca-and-t-sne/\n\n\nhttps://gweb-cloudblog-publish.appspot.com/products/gcp/drawings-in-the-cloud-introducing-the-quick-draw-dataset/amp/", "415437": "Has anyone successfully incorporated Country or Number of Strokes info? They didn't add anything to my model, made it a little bit worse. ", "1076126": "I write a script for a beginner who wants to Learn OCR Step by Step. Just go and watch and support me if you like:-\n\nhttps://www.kaggle.com/naim99/ocr-text-recognition-ocr-space-api-tesseract", "416317": "sparse convolution:\n\n\nhttps://github.com/facebookresearch/SparseConvNet\n\n", "416289": "visualisation:\nhttp://vallandingham.me/quickdraw/", "415583": "possible data augmentation:\n\nhttps://panly099.github.io/skSyn.html\n\nhttps://github.com/panly099/sketchSynthesis\n\nFree-hand sketch synthesis with deformable stroke models\n\n\n---\nhttps://github.com/val-iisc/sketch-parse\n\n \"SketchParse: Towards Rich Descriptions For Poorly Drawn Sketches Using Multi-Task Deep Networks\", ACM Multimedia (ACMMM) 2017.\n\n---\n\nsketch parsing/ vectorisation : can be used as additional supervision signal in multi-task setting?\n\nSemantic Segmentation for Line Drawing Vectorization Using\nNeural Networks\n\n\nhttp://www.byungsoo.me/project/vectornet/paper.pdf\nhttps://www.microsoft.com/en-us/research/wp-content/uploads/2012/01/16-2012-eccv-sketch-segmenation.pdf", "407515": "The time to draw the doodle and other timing details can be easily extracted from the raw data, which contains [x, y, t] coordinates of pen/mouse strokes.", "407406": "example where ocr to recognize words can be important:\n\nnaming of file: (top prob after softmax), (key id), (top 3 class)\n\n\n ![enter image description here][1]\n ![enter image description here][2]\n\n\n [1]: https://storage.googleapis.com/kaggle-forum-message-attachments/inbox/113660/cbee46ed9e840de5c921360fc30870fb/0.15_9997386404410976_owl_frog_kangaroo.png\n [2]: https://storage.googleapis.com/kaggle-forum-message-attachments/inbox/113660/fa3f80258d1e1f53d3ee8df574c3788c/0.25_9722430504684690_sea_turtle_frog_teddy-bear.png" } }