![\"\"](\"https://github.com/root4kaido/kaggle-discussion-image/blob/main/SETI%20Breakthrough%20Listen/oof_pred_3.png?raw=true\")
Hi.
\nOf the oof predictions made by my model, I extracted those with target=1 and pred=0.
\nHere are some examples．(0, 2, and 4 signals only)
![\"\"](\"https://github.com/root4kaido/kaggle-discussion-image/blob/main/SETI%20Breakthrough%20Listen/oof_pred_4.png?raw=true\")
![\"\"](\"https://github.com/root4kaido/kaggle-discussion-image/blob/main/SETI%20Breakthrough%20Listen/oof_pred_5.png?raw=true\")
![\"\"](\"https://github.com/root4kaido/kaggle-discussion-image/blob/main/SETI%20Breakthrough%20Listen/oof_pred_6.png?raw=true\")
Where are the signs from the aliens…?
\nThe images do not look much different from the image with target=0.
Is there any way to enhance the sign with some image processing?
\n(For example, gamma correction, contrast adjustment, etc.)
\nIf you can find it, that might be a useful preprocessing step.
(Frist of all, there really are signs on these images, aren't there…?)
\nThanks.
Update: img size
", "messages": [ { "id": 1349839, "postDate": "2021-06-15T05:53:07.540Z", "content": "Hi.
\nOf the oof predictions made by my model, I extracted those with target=1 and pred=0.
\nHere are some examples．(0, 2, and 4 signals only)
Where are the signs from the aliens…?
\nThe images do not look much different from the image with target=0.
Is there any way to enhance the sign with some image processing?
\n(For example, gamma correction, contrast adjustment, etc.)
\nIf you can find it, that might be a useful preprocessing step.
(Frist of all, there really are signs on these images, aren't there…?)
\nThanks.
Update: img size
", "rawMarkdown": "Hi.\n\nOf the oof predictions made by my model, I extracted those with target=1 and pred=0.\nHere are some examples．(0, 2, and 4 signals only)\n\n\n\n\n\n\n\n\n\nWhere are the signs from the aliens...?\nThe images do not look much different from the image with target=0.\n\nIs there any way to enhance the sign with some image processing?\n(For example, gamma correction, contrast adjustment, etc.)\nIf you can find it, that might be a useful preprocessing step.\n\n(Frist of all, there really are signs on these images, aren't there...?)\nThanks.\n\nUpdate: img size", "votes": 10 }, { "id": 1349992, "postDate": "2021-06-15T07:56:52.843Z", "content": "Some signals become visible for human eye with normalization technique introduced in here. I previously used following function to detect easter eggs.
\ndef normalize_ft(x):\n x = (x - np.mean(x, axis=1, keepdims=True)) / np.std(x, axis=1, keepdims=True)\n\n x = (x - np.mean(x, axis=2, keepdims=True)) / np.std(x, axis=2, keepdims=True)\n return x\nFor ccdf663b52e8.npy, I could see a very small signal in upper right part in img[2].
\n![\"\"](\"https://storage.googleapis.com/kagglesdsdata/datasets/1235322/2335204/seti_ccdf663b52e8_2.png?X-Goog-Algorithm=GOOG4-RSA-SHA256&X-Goog-Credential=databundle-worker-v2%40kaggle-161607.iam.gserviceaccount.com%2F20210627%2Fauto%2Fstorage%2Fgoog4_request&X-Goog-Date=20210627T072542Z&X-Goog-Expires=345599&X-Goog-SignedHeaders=host&X-Goog-Signature=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\")
I am not sure it is useful for DL models.
", "rawMarkdown": "Some signals become visible for human eye with normalization technique introduced in [here](https://www.kaggle.com/agentauers/seti21-normalization-of-data/comments). I previously used following function [to detect easter eggs](https://www.kaggle.com/tomooinubushi/easter-egg-detection-with-template-matching).\n```\ndef normalize_ft(x):\n x = (x - np.mean(x, axis=1, keepdims=True)) / np.std(x, axis=1, keepdims=True)\n\n x = (x - np.mean(x, axis=2, keepdims=True)) / np.std(x, axis=2, keepdims=True)\n return x\n```\n\nFor ccdf663b52e8.npy, I could see a very small signal in upper right part in img[2]. \n\n\n\nI am not sure it is useful for DL models.\n", "votes": 7, "replies": [ { "id": 1350131, "postDate": "2021-06-15T09:41:17.377Z", "content": "You're right! Some of the signals are coming to light!
\nThanks for introducing a nice approach.
\nAnd I have a feeling that we should also address the problem of very small signals like this one…
For ccdf663b52e8.npy, is this the signal you were talking about?
\nI noticed that in the non-normalized version of ccdf663b52e8.npy, there are some horizontal streaks that appear intermittently and randomly during the \"ON\" snippets. But I guess these horizontal streaks don't count as techno-signatures. Rather, we are only looking for (possibly faint) signals that look like pencil drawings and only appear in the \"ON\" snippets.
", "rawMarkdown": "For ccdf663b52e8.npy, is this [the signal](https://imgur.com/gallery/F3oOGTs) you were talking about?\n\nI noticed that in the non-normalized version of ccdf663b52e8.npy, there are some horizontal streaks that appear intermittently and randomly during the \"ON\" snippets. But I guess these horizontal streaks don't count as techno-signatures. Rather, we are only looking for (possibly faint) signals that look like pencil drawings and only appear in the \"ON\" snippets.\n" } ] } ], "comments": [ { "id": 1349992, "author_name": "tomoo inubushi", "author_url": "", "post_date": "2021-06-15T07:56:52.843000", "content": "Some signals become visible for human eye with normalization technique introduced in here. I previously used following function to detect easter eggs.
\ndef normalize_ft(x):\n x = (x - np.mean(x, axis=1, keepdims=True)) / np.std(x, axis=1, keepdims=True)\n\n x = (x - np.mean(x, axis=2, keepdims=True)) / np.std(x, axis=2, keepdims=True)\n return x\nFor ccdf663b52e8.npy, I could see a very small signal in upper right part in img[2].
\n
I am not sure it is useful for DL models.
", "votes": 7, "replies": [ { "id": 1350131, "author_name": "sqrt4kaido", "author_url": "", "post_date": "2021-06-15T09:41:17.377000", "content": "You're right! Some of the signals are coming to light!
\nThanks for introducing a nice approach.
\nAnd I have a feeling that we should also address the problem of very small signals like this one…
For ccdf663b52e8.npy, is this the signal you were talking about?
\nI noticed that in the non-normalized version of ccdf663b52e8.npy, there are some horizontal streaks that appear intermittently and randomly during the \"ON\" snippets. But I guess these horizontal streaks don't count as techno-signatures. Rather, we are only looking for (possibly faint) signals that look like pencil drawings and only appear in the \"ON\" snippets.
", "votes": 0, "replies": [] } ] } ], "raw_markdown_by_id": { "1349839": "Hi.\n\nOf the oof predictions made by my model, I extracted those with target=1 and pred=0.\nHere are some examples．(0, 2, and 4 signals only)\n\n\n\n\n\n\n\n\n\nWhere are the signs from the aliens...?\nThe images do not look much different from the image with target=0.\n\nIs there any way to enhance the sign with some image processing?\n(For example, gamma correction, contrast adjustment, etc.)\nIf you can find it, that might be a useful preprocessing step.\n\n(Frist of all, there really are signs on these images, aren't there...?)\nThanks.\n\nUpdate: img size", "1349992": "Some signals become visible for human eye with normalization technique introduced in [here](https://www.kaggle.com/agentauers/seti21-normalization-of-data/comments). I previously used following function [to detect easter eggs](https://www.kaggle.com/tomooinubushi/easter-egg-detection-with-template-matching).\n```\ndef normalize_ft(x):\n x = (x - np.mean(x, axis=1, keepdims=True)) / np.std(x, axis=1, keepdims=True)\n\n x = (x - np.mean(x, axis=2, keepdims=True)) / np.std(x, axis=2, keepdims=True)\n return x\n```\n\nFor ccdf663b52e8.npy, I could see a very small signal in upper right part in img[2]. \n\n\n\nI am not sure it is useful for DL models.\n" } }