![\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F6004998%2Fb8ebc6d7a88071ec20bf1cfd0e08bc48%2F1.PNG?generation=1684857839791262&alt=media\")
\n
![\"the](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F6004998%2F40d192528cf0e70088365f97bcf9fe83%2F2.PNG?generation=1684857889493002&alt=media\")
this is my RLE, and I can not see why this is happening, so please help!
\ndef rle(output):
\n flat_img = output.flatten()
\n flat_img = np.where(output > 0.5, 1, 0).astype(np.uint8)
\n starts = np.array((flat_img[:-1] == 0) & (flat_img[1:] == 1))
\n ends = np.array((flat_img[:-1] == 1) & (flat_img[1:] == 0))
\n starts_ix = np.where(starts)[0] + 2
\n ends_ix = np.where(ends)[0] + 2
\n lengths = ends_ix - starts_ix
\n return \" \".join(map(str, sum(zip(starts_ix, lengths), ())))
another qu, is this overfitting?
\n
this is my RLE, and I can not see why this is happening, so please help!
\ndef rle(output):
\n flat_img = output.flatten()
\n flat_img = np.where(output > 0.5, 1, 0).astype(np.uint8)
\n starts = np.array((flat_img[:-1] == 0) & (flat_img[1:] == 1))
\n ends = np.array((flat_img[:-1] == 1) & (flat_img[1:] == 0))
\n starts_ix = np.where(starts)[0] + 2
\n ends_ix = np.where(ends)[0] + 2
\n lengths = ends_ix - starts_ix
\n return \" \".join(map(str, sum(zip(starts_ix, lengths), ())))
another qu, is this overfitting?
\n
Are you training and validating/evaluating on the same data? Your output looks almost identical to the mask.
", "rawMarkdown": "Are you training and validating/evaluating on the same data? Your output looks almost identical to the mask.", "votes": null }, { "id": "2272645", "postDate": "05/24/2023 16:11:24", "content": "Sharing my working RLE code here. To test if it works for you, try submitting the mask as your prediction for each fragment and you should get a score of 0.11:
\n\nTHRESHOLD = \n\n ():\n pixels = np.where(output.flatten() > THRESHOLD, , ).astype(np.uint8)\n pixels[] = \n pixels[-] = \n runs = np.where(pixels[:] != pixels[:-])[] + \n runs[::] = runs[::] - runs[:-:]\n .join((x) x runs)\n\n\nrle_output1 = rle(FIRST_PREDICTION)\nrle_output2 = rle(SECOND_PREDICTION)\n\n( + rle_output1 + + rle_output2, file=(, ))\nCould you please elaborate about this\"try submitting the mask as your prediction for each fragment and you should get a score of 0.11:\"
", "rawMarkdown": "Could you please elaborate about this\"try submitting the mask as your prediction for each fragment and you should get a score of 0.11:\"", "votes": null }, { "id": "2272662", "postDate": "05/24/2023 16:22:21", "content": "It's an overfitting problem, actually the model architecture is very good I think, but the problem is with the noise of the data and the lack of augmentation
", "rawMarkdown": "It's an overfitting problem, actually the model architecture is very good I think, but the problem is with the noise of the data and the lack of augmentation", "votes": null }, { "id": "2274389", "postDate": "05/26/2023 00:10:09", "content": "If you create a submission notebook that runs RLE on the masks for the test fragments (no model involved) you will earn a score of 0.11 on the leaderboard. That's one way to ensure your RLE code is working.
", "rawMarkdown": "If you create a submission notebook that runs RLE on the masks for the test fragments (no model involved) you will earn a score of 0.11 on the leaderboard. That's one way to ensure your RLE code is working.", "votes": null } ], "comments": [ { "id": 2272604, "author_name": "stevenhewitt", "author_url": "", "post_date": "05/24/2023 15:51:24", "content": "Are you training and validating/evaluating on the same data? Your output looks almost identical to the mask.
", "votes": null, "replies": [ { "id": 2272662, "author_name": "salaheldinelnabarawy", "author_url": "", "post_date": "05/24/2023 16:22:21", "content": "It's an overfitting problem, actually the model architecture is very good I think, but the problem is with the noise of the data and the lack of augmentation
", "votes": null, "replies": [] } ] }, { "id": 2272645, "author_name": "stevenhewitt", "author_url": "", "post_date": "05/24/2023 16:11:24", "content": "Sharing my working RLE code here. To test if it works for you, try submitting the mask as your prediction for each fragment and you should get a score of 0.11:
\n\nTHRESHOLD = \n\n ():\n pixels = np.where(output.flatten() > THRESHOLD, , ).astype(np.uint8)\n pixels[] = \n pixels[-] = \n runs = np.where(pixels[:] != pixels[:-])[] + \n runs[::] = runs[::] - runs[:-:]\n .join((x) x runs)\n\n\nrle_output1 = rle(FIRST_PREDICTION)\nrle_output2 = rle(SECOND_PREDICTION)\n\n( + rle_output1 + + rle_output2, file=(, ))\nCould you please elaborate about this\"try submitting the mask as your prediction for each fragment and you should get a score of 0.11:\"
", "votes": null, "replies": [ { "id": 2274389, "author_name": "stevenhewitt", "author_url": "", "post_date": "05/26/2023 00:10:09", "content": "If you create a submission notebook that runs RLE on the masks for the test fragments (no model involved) you will earn a score of 0.11 on the leaderboard. That's one way to ensure your RLE code is working.
", "votes": null, "replies": [] } ] } ] } ], "raw_markdown_by_id": { "2271091": "this is my RLE, and I can not see why this is happening, so please help!\n\ndef rle(output):\n flat_img = output.flatten()\n flat_img = np.where(output > 0.5, 1, 0).astype(np.uint8)\n starts = np.array((flat_img[:-1] == 0) & (flat_img[1:] == 1))\n ends = np.array((flat_img[:-1] == 1) & (flat_img[1:] == 0))\n starts_ix = np.where(starts)[0] + 2\n ends_ix = np.where(ends)[0] + 2\n lengths = ends_ix - starts_ix\n return \" \".join(map(str, sum(zip(starts_ix, lengths), ())))\n\n\nanother qu, is this overfitting? \n", "2272604": "Are you training and validating/evaluating on the same data? Your output looks almost identical to the mask.", "2272645": "Sharing my working RLE code here. To test if it works for you, try submitting the mask as your prediction for each fragment and you should get a score of 0.11:\n```python\n# Set this to your prediction binarization threshold. I use 0.5 for my models.\nTHRESHOLD = 0.5\n\ndef rle(output):\n pixels = np.where(output.flatten() > THRESHOLD, 1, 0).astype(np.uint8)\n pixels[0] = 0\n pixels[-1] = 0\n runs = np.where(pixels[1:] != pixels[:-1])[0] + 2\n runs[1::2] = runs[1::2] - runs[:-1:2]\n return ' '.join(str(x) for x in runs)\n\n# Substitute below with your predictions as numpy arrays.\nrle_output1 = rle(FIRST_PREDICTION)\nrle_output2 = rle(SECOND_PREDICTION)\n\nprint(\"Id,Predicted\\na,\" + rle_output1 + \"\\nb,\" + rle_output2, file=open('submission.csv', 'w'))\n```", "2272657": "Could you please elaborate about this\"try submitting the mask as your prediction for each fragment and you should get a score of 0.11:\"", "2272662": "It's an overfitting problem, actually the model architecture is very good I think, but the problem is with the noise of the data and the lack of augmentation", "2274389": "If you create a submission notebook that runs RLE on the masks for the test fragments (no model involved) you will earn a score of 0.11 on the leaderboard. That's one way to ensure your RLE code is working." }, "source": "meta" }