In previous competitions I've used @paulorzp 's amazing notebook here for encoding and decoding RLE strings.
\nUnfortunately, it seems that due to the size of the masks in this competition, I ran into memory problems (it kept crashing).
\nSo I made a more efficient implementation of his mask2rle function below:
\ndef mask2rle(img):\n '''\n Efficient implementation of mask2rle, from @paulorzp\n https://www.kaggle.com/xhlulu/efficient-mask2rle\n https://www.kaggle.com/paulorzp/rle-functions-run-lenght-encode-decode\n --\n img: numpy array, 1 - mask, 0 - background\n Returns run length as string formated\n '''\n pixels = img.T.flatten()\n pixels = np.pad(pixels, ((1, 1), ))\n runs = np.where(pixels[1:] != pixels[:-1])[0] + 1\n runs[1::2] -= runs[::2]\n return ' '.join(str(x) for x in runs)\nYou can find my notebook here if you want to test it.
", "messages": [ { "id": 1085214, "postDate": "2020-11-20T18:56:17.517Z", "content": "In previous competitions I've used @paulorzp 's amazing notebook here for encoding and decoding RLE strings.
\nUnfortunately, it seems that due to the size of the masks in this competition, I ran into memory problems (it kept crashing).
\nSo I made a more efficient implementation of his mask2rle function below:
\ndef mask2rle(img):\n '''\n Efficient implementation of mask2rle, from @paulorzp\n https://www.kaggle.com/xhlulu/efficient-mask2rle\n https://www.kaggle.com/paulorzp/rle-functions-run-lenght-encode-decode\n --\n img: numpy array, 1 - mask, 0 - background\n Returns run length as string formated\n '''\n pixels = img.T.flatten()\n pixels = np.pad(pixels, ((1, 1), ))\n runs = np.where(pixels[1:] != pixels[:-1])[0] + 1\n runs[1::2] -= runs[::2]\n return ' '.join(str(x) for x in runs)\nYou can find my notebook here if you want to test it.
", "rawMarkdown": "In previous competitions I've used @paulorzp 's amazing [notebook here](https://www.kaggle.com/paulorzp/rle-functions-run-lenght-encode-decode) for encoding and decoding RLE strings. \n\nUnfortunately, it seems that due to the size of the masks in this competition, I ran into memory problems (it kept crashing).\n\nSo I made a more efficient implementation of his mask2rle function below:\n```python\ndef mask2rle(img):\n '''\n Efficient implementation of mask2rle, from @paulorzp\n https://www.kaggle.com/xhlulu/efficient-mask2rle\n https://www.kaggle.com/paulorzp/rle-functions-run-lenght-encode-decode\n --\n img: numpy array, 1 - mask, 0 - background\n Returns run length as string formated\n '''\n pixels = img.T.flatten()\n pixels = np.pad(pixels, ((1, 1), ))\n runs = np.where(pixels[1:] != pixels[:-1])[0] + 1\n runs[1::2] -= runs[::2]\n return ' '.join(str(x) for x in runs)\n```\n\nYou can find my [notebook](https://www.kaggle.com/xhlulu/efficient-mask2rle) here if you want to test it.", "votes": 25 }, { "id": 1086254, "postDate": "2020-11-21T14:18:51.780Z", "content": "I think the difference is that np.concatenate is changed to np.pad? Am I missing something?
\nI took another path in this notebook: this operation is only to encode first and last pixel. If you can ignore these pixels, you don't need to pad the input with zeros.
", "rawMarkdown": "I think the difference is that np.concatenate is changed to np.pad? Am I missing something?\n\nI took another path in [this notebook](https://www.kaggle.com/bguberfain/memory-aware-rle-encoding): this operation is only to encode first and last pixel. If you can ignore these pixels, you don't need to pad the input with zeros." }, { "id": 1085990, "postDate": "2020-11-21T10:12:29.107Z", "rawMarkdown": "", "isDeleted": true }, { "id": 1085329, "postDate": "2020-11-20T20:50:34.400Z", "content": "Awesome, thanks for sharing
", "rawMarkdown": "Awesome, thanks for sharing" } ], "comments": [ { "id": 1086254, "author_name": "Bruno G. do Amaral", "author_url": "", "post_date": "2020-11-21T14:18:51.780000", "content": "I think the difference is that np.concatenate is changed to np.pad? Am I missing something?
\nI took another path in this notebook: this operation is only to encode first and last pixel. If you can ignore these pixels, you don't need to pad the input with zeros.
", "votes": 0, "replies": [] }, { "id": 1085990, "author_name": "", "author_url": "", "post_date": "2020-11-21T10:12:29.107000", "content": "", "votes": 0, "replies": [] }, { "id": 1085329, "author_name": "Matt", "author_url": "", "post_date": "2020-11-20T20:50:34.400000", "content": "Awesome, thanks for sharing
", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "1085214": "In previous competitions I've used @paulorzp 's amazing [notebook here](https://www.kaggle.com/paulorzp/rle-functions-run-lenght-encode-decode) for encoding and decoding RLE strings. \n\nUnfortunately, it seems that due to the size of the masks in this competition, I ran into memory problems (it kept crashing).\n\nSo I made a more efficient implementation of his mask2rle function below:\n```python\ndef mask2rle(img):\n '''\n Efficient implementation of mask2rle, from @paulorzp\n https://www.kaggle.com/xhlulu/efficient-mask2rle\n https://www.kaggle.com/paulorzp/rle-functions-run-lenght-encode-decode\n --\n img: numpy array, 1 - mask, 0 - background\n Returns run length as string formated\n '''\n pixels = img.T.flatten()\n pixels = np.pad(pixels, ((1, 1), ))\n runs = np.where(pixels[1:] != pixels[:-1])[0] + 1\n runs[1::2] -= runs[::2]\n return ' '.join(str(x) for x in runs)\n```\n\nYou can find my [notebook](https://www.kaggle.com/xhlulu/efficient-mask2rle) here if you want to test it.", "1086254": "I think the difference is that np.concatenate is changed to np.pad? Am I missing something?\n\nI took another path in [this notebook](https://www.kaggle.com/bguberfain/memory-aware-rle-encoding): this operation is only to encode first and last pixel. If you can ignore these pixels, you don't need to pad the input with zeros.", "1085990": "", "1085329": "Awesome, thanks for sharing" } }