Hi everyone, I'm curious how you guys encode temporal information, i.e., the ordering of strokes, into images, for now I'm using Beluga's method:
\n\npython\ndef draw_cv2(raw_strokes, size=256, lw=6, time_color=True):\n img = np.zeros((BASE_SIZE, BASE_SIZE), np.uint8)\n for t, stroke in enumerate(raw_strokes):\n for i in range(len(stroke[0]) - 1):\n color = 255 - min(t, 10) * 13 if time_color else 255\n _ = cv2.line(img, (stroke[0][i], stroke[1][i]),\n (stroke[0][i + 1], stroke[1][i + 1]), color, lw)\n if size != BASE_SIZE:\n return cv2.resize(img, (size, size))\n else:\n return img\n\nThe method above is quite similar to this paper, Sketch-a-Net that Beats Humans
And today I saw two newly published methods:
\n\nThere is another paper, Sketch-R2CNN - An Attentive Network for Vector Sketch Recognition, it doesn't use hard coded colors, instead its colors are learnt by a RNN model, which looks very promising. However the author didn't publish the code so I don't have a chance to try it. Lukasz Grad has published his implementation here, https://www.kaggle.com/c/quickdraw-doodle-recognition/discussion/73699 , I'll give it a try later.
\n\nThe four methods above are what I know so far.
\n\nCould you share your code in this thread to show how you generate images? Thanks a lot!
", "messages": [ { "id": 433698, "postDate": "2018-12-05T11:14:43.950Z", "content": "Hi everyone, I'm curious how you guys encode temporal information, i.e., the ordering of strokes, into images, for now I'm using Beluga's method:
\n\npython\ndef draw_cv2(raw_strokes, size=256, lw=6, time_color=True):\n img = np.zeros((BASE_SIZE, BASE_SIZE), np.uint8)\n for t, stroke in enumerate(raw_strokes):\n for i in range(len(stroke[0]) - 1):\n color = 255 - min(t, 10) * 13 if time_color else 255\n _ = cv2.line(img, (stroke[0][i], stroke[1][i]),\n (stroke[0][i + 1], stroke[1][i + 1]), color, lw)\n if size != BASE_SIZE:\n return cv2.resize(img, (size, size))\n else:\n return img\n\nThe method above is quite similar to this paper, Sketch-a-Net that Beats Humans
And today I saw two newly published methods:
\n\nThere is another paper, Sketch-R2CNN - An Attentive Network for Vector Sketch Recognition, it doesn't use hard coded colors, instead its colors are learnt by a RNN model, which looks very promising. However the author didn't publish the code so I don't have a chance to try it. Lukasz Grad has published his implementation here, https://www.kaggle.com/c/quickdraw-doodle-recognition/discussion/73699 , I'll give it a try later.
\n\nThe four methods above are what I know so far.
\n\nCould you share your code in this thread to show how you generate images? Thanks a lot!
", "rawMarkdown": "Hi everyone, I'm curious how you guys encode temporal information, i.e., the ordering of strokes, into images, for now I'm using Beluga's method:\n\n```python\ndef draw_cv2(raw_strokes, size=256, lw=6, time_color=True):\n img = np.zeros((BASE_SIZE, BASE_SIZE), np.uint8)\n for t, stroke in enumerate(raw_strokes):\n for i in range(len(stroke[0]) - 1):\n color = 255 - min(t, 10) * 13 if time_color else 255\n _ = cv2.line(img, (stroke[0][i], stroke[1][i]),\n (stroke[0][i + 1], stroke[1][i + 1]), color, lw)\n if size != BASE_SIZE:\n return cv2.resize(img, (size, size))\n else:\n return img\n```\nThe method above is quite similar to this paper, [Sketch-a-Net that Beats Humans][1]\n\nAnd today I saw two newly published methods:\n\n2. Siyuan Dang's kenel, [How I colored strokes](https://www.kaggle.com/sheboke93/how-i-colored-strokes)\n3. Gunther's kenel, [Fastai conform dataset with resnet18](https://www.kaggle.com/guntherthepenguin/fastai-resnet18-color-coded-and-focalloss) , \n\n\n\nThere is another paper, [Sketch-R2CNN - An Attentive Network for Vector Sketch Recognition][2], it doesn't use hard coded colors, instead its colors are learnt by a RNN model, which looks very promising. However the author didn't publish the code so I don't have a chance to try it. Lukasz Grad has published his implementation here, https://www.kaggle.com/c/quickdraw-doodle-recognition/discussion/73699 , I'll give it a try later.\n\nThe four methods above are what I know so far.\n\nCould you share your code in this thread to show how you generate images? Thanks a lot!\n\n\n [1]: https://arxiv.org/pdf/1501.07873.pdf\n [2]: https://arxiv.org/pdf/1811.08170.pdf" } ], "comments": [], "raw_markdown_by_id": { "433698": "Hi everyone, I'm curious how you guys encode temporal information, i.e., the ordering of strokes, into images, for now I'm using Beluga's method:\n\n```python\ndef draw_cv2(raw_strokes, size=256, lw=6, time_color=True):\n img = np.zeros((BASE_SIZE, BASE_SIZE), np.uint8)\n for t, stroke in enumerate(raw_strokes):\n for i in range(len(stroke[0]) - 1):\n color = 255 - min(t, 10) * 13 if time_color else 255\n _ = cv2.line(img, (stroke[0][i], stroke[1][i]),\n (stroke[0][i + 1], stroke[1][i + 1]), color, lw)\n if size != BASE_SIZE:\n return cv2.resize(img, (size, size))\n else:\n return img\n```\nThe method above is quite similar to this paper, [Sketch-a-Net that Beats Humans][1]\n\nAnd today I saw two newly published methods:\n\n2. Siyuan Dang's kenel, [How I colored strokes](https://www.kaggle.com/sheboke93/how-i-colored-strokes)\n3. Gunther's kenel, [Fastai conform dataset with resnet18](https://www.kaggle.com/guntherthepenguin/fastai-resnet18-color-coded-and-focalloss) , \n\n\n\nThere is another paper, [Sketch-R2CNN - An Attentive Network for Vector Sketch Recognition][2], it doesn't use hard coded colors, instead its colors are learnt by a RNN model, which looks very promising. However the author didn't publish the code so I don't have a chance to try it. Lukasz Grad has published his implementation here, https://www.kaggle.com/c/quickdraw-doodle-recognition/discussion/73699 , I'll give it a try later.\n\nThe four methods above are what I know so far.\n\nCould you share your code in this thread to show how you generate images? Thanks a lot!\n\n\n [1]: https://arxiv.org/pdf/1501.07873.pdf\n [2]: https://arxiv.org/pdf/1811.08170.pdf" } }