{ "id": 292807, "title": "Is This False Positive?", "url": "/competitions/tensorflow-great-barrier-reef/discussion/292807", "author_name": "", "post_date": "2021-12-03T13:37:54.493169100Z", "votes": 23, "comment_count": 6, "views": 0, "content": "

This is oof of video_1/9088.jpg
\n\"2021-12-03-22-13-56\"

\n

and this looks like a starfish to me. Is this false positive or actual starfish?
\n\"2021-12-03-22-14-21\"

\n

[EDIT]
\ndifferent frame looks more like a starfish
\n\"2021-12-03-22-47-57\"

", "messages": [ { "id": "1604540", "postDate": "12/03/2021 13:37:54", "content": "

This is oof of video_1/9088.jpg
\n\"2021-12-03-22-13-56\"

\n

and this looks like a starfish to me. Is this false positive or actual starfish?
\n\"2021-12-03-22-14-21\"

\n

[EDIT]
\ndifferent frame looks more like a starfish
\n\"2021-12-03-22-47-57\"

", "rawMarkdown": "This is oof of video_1/9088.jpg\n\"2021-12-03-22-13-56\"\n\nand this looks like a starfish to me. Is this false positive or actual starfish?\n\"2021-12-03-22-14-21\"\n\n[EDIT]\ndifferent frame looks more like a starfish\n\"2021-12-03-22-47-57\"", "votes": null }, { "id": "1604543", "postDate": "12/03/2021 13:41:39", "content": "

As for me, it looks like a COST. But we never will be sure.
\nIt's likely that annotators miss some COSTS.
\nYour model detects COSTS much better than me as a human 😄

", "rawMarkdown": "As for me, it looks like a COST. But we never will be sure.\nIt's likely that annotators miss some COSTS.\nYour model detects COSTS much better than me as a human 😄", "votes": null }, { "id": "1604621", "postDate": "12/03/2021 14:58:27", "content": "

Yes, I noticed the same. In some frames you can find something that looks like starfish.

", "rawMarkdown": "Yes, I noticed the same. In some frames you can find something that looks like starfish.", "votes": null }, { "id": "1604862", "postDate": "12/03/2021 19:34:22", "content": "

I dunno i see COTS everywhere lately, sunflowers, hedgehogs so yeah there is a COTS :D

", "rawMarkdown": "I dunno i see COTS everywhere lately, sunflowers, hedgehogs so yeah there is a COTS :D", "votes": null }, { "id": "1656871", "postDate": "01/19/2022 18:00:07", "content": "

\"\"
\nthis feels to me as True Positive
\nlet's look very close
\n\"\"

\n

In most cases, my models can detect COTS before the GT finds it in future frames (it detect a bit early). is there anything I can do to get rid of it

", "rawMarkdown": "![](https://i.ibb.co/LRs8jcy/image.png)\nthis feels to me as True Positive\nlet's look very close\n![](https://i.ibb.co/wgM6wKc/image.png)\n\nIn most cases, my models can detect COTS before the GT finds it in future frames (it detect a bit early). is there anything I can do to get rid of it", "votes": null }, { "id": "1657099", "postDate": "01/19/2022 22:48:24", "content": "

Rough idea is:

\n
    \n
  1. Manually label them
    2. \n
  2. Automatically label them
    3. \n
", "rawMarkdown": "Rough idea is:\n\n1. Manually label them\n2. Automatically label them\n - Tracking COTS and pseudo label or discard first N frame (if discarding, we lose GT sample, and pseudo labeling may produce false GT labels)\n - pseudo label with your detector (may also produce false GT labels)", "votes": null }, { "id": "1657104", "postDate": "01/19/2022 22:55:50", "content": "

False GT labels might not affect the score seriously in this competition, because evaluation is based on F2 (x4 weights FN than FP).

", "rawMarkdown": "False GT labels might not affect the score seriously in this competition, because evaluation is based on F2 (x4 weights FN than FP).", "votes": null } ], "comments": [ { "id": 1604543, "author_name": "meowmeowmeowmeowmeow", "author_url": "", "post_date": "12/03/2021 13:41:39", "content": "

As for me, it looks like a COST. But we never will be sure.
\nIt's likely that annotators miss some COSTS.
\nYour model detects COSTS much better than me as a human 😄

", "votes": null, "replies": [] }, { "id": 1604621, "author_name": "remekkinas", "author_url": "", "post_date": "12/03/2021 14:58:27", "content": "

Yes, I noticed the same. In some frames you can find something that looks like starfish.

", "votes": null, "replies": [] }, { "id": 1604862, "author_name": "lukaszborecki", "author_url": "", "post_date": "12/03/2021 19:34:22", "content": "

I dunno i see COTS everywhere lately, sunflowers, hedgehogs so yeah there is a COTS :D

", "votes": null, "replies": [] }, { "id": 1656871, "author_name": "morizin", "author_url": "", "post_date": "01/19/2022 18:00:07", "content": "

\"\"
\nthis feels to me as True Positive
\nlet's look very close
\n\"\"

\n

In most cases, my models can detect COTS before the GT finds it in future frames (it detect a bit early). is there anything I can do to get rid of it

", "votes": null, "replies": [ { "id": 1657099, "author_name": "tatamikenn", "author_url": "", "post_date": "01/19/2022 22:48:24", "content": "

Rough idea is:

\n
    \n
  1. Manually label them
    2. \n
  2. Automatically label them
    3. \n
", "votes": null, "replies": [] }, { "id": 1657104, "author_name": "tatamikenn", "author_url": "", "post_date": "01/19/2022 22:55:50", "content": "

False GT labels might not affect the score seriously in this competition, because evaluation is based on F2 (x4 weights FN than FP).

", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "1604540": "This is oof of video_1/9088.jpg\n\"2021-12-03-22-13-56\"\n\nand this looks like a starfish to me. Is this false positive or actual starfish?\n\"2021-12-03-22-14-21\"\n\n[EDIT]\ndifferent frame looks more like a starfish\n\"2021-12-03-22-47-57\"", "1604543": "As for me, it looks like a COST. But we never will be sure.\nIt's likely that annotators miss some COSTS.\nYour model detects COSTS much better than me as a human 😄", "1604621": "Yes, I noticed the same. In some frames you can find something that looks like starfish.", "1604862": "I dunno i see COTS everywhere lately, sunflowers, hedgehogs so yeah there is a COTS :D", "1656871": "![](https://i.ibb.co/LRs8jcy/image.png)\nthis feels to me as True Positive\nlet's look very close\n![](https://i.ibb.co/wgM6wKc/image.png)\n\nIn most cases, my models can detect COTS before the GT finds it in future frames (it detect a bit early). is there anything I can do to get rid of it", "1657099": "Rough idea is:\n\n1. Manually label them\n2. Automatically label them\n - Tracking COTS and pseudo label or discard first N frame (if discarding, we lose GT sample, and pseudo labeling may produce false GT labels)\n - pseudo label with your detector (may also produce false GT labels)", "1657104": "False GT labels might not affect the score seriously in this competition, because evaluation is based on F2 (x4 weights FN than FP)." }, "source": "meta" }