{ "id": 403113, "title": "I found an error in ink_label", "url": "/competitions/vesuvius-challenge-ink-detection/discussion/403113", "author_name": "JavaZero", "post_date": "2023-04-21T08:42:15.878000", "votes": 25, "comment_count": 6, "views": 0, "content": "

\"\"

\n

It seems that the label is incorrect👀. I have uploaded a fixed label image. 💯

\n

Thanks to @mecndo for the discovery. 😍

", "messages": [ { "id": 2229308, "postDate": "2023-04-21T08:42:15.880Z", "content": "

\"\"

\n

It seems that the label is incorrect👀. I have uploaded a fixed label image. 💯

\n

Thanks to @mecndo for the discovery. 😍

", "rawMarkdown": "![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F8314415%2F883b9725a79c079e916cde646d9ff0e8%2Fink_label_error.gif?generation=1682066308329795&alt=media)\n\nIt seems that the label is incorrect👀. I have uploaded a fixed label image. 💯\n\nThanks to @mecndo for the discovery. 😍", "votes": 24 }, { "id": 2232081, "postDate": "2023-04-24T02:05:06.297Z", "content": "

Interesting find. Morhpological operations should be able remove any small target pixels in the truth masks. Maybe it's worth trying to see how it affects training?

", "rawMarkdown": "Interesting find. [Morhpological](https://docs.opencv.org/4.x/d9/d61/tutorial_py_morphological_ops.html) operations should be able remove any small target pixels in the truth masks. Maybe it's worth trying to see how it affects training?", "votes": 3, "replies": [ { "id": 2232084, "postDate": "2023-04-24T02:13:46.110Z", "content": "

thx a lot. I will finish the experiment. (But I dont think it will affect a lot😂)

", "rawMarkdown": "thx a lot. I will finish the experiment. (But I dont think it will affect a lot😂)", "votes": 1 } ] }, { "id": 2229315, "postDate": "2023-04-21T08:53:38.813Z", "content": "

yeah, i saw that too, another one that im not sure about is this one on fragment 3 which is way larger:
\nf3
\nf3

", "rawMarkdown": "yeah, i saw that too, another one that im not sure about is this one on fragment 3 which is way larger: \n[f3](https://i.postimg.cc/rFzDMJNY/Screenshot-from-2023-04-21-11-05-35.png)\n[f3](https://i.postimg.cc/KvPdKrzm/000.png)", "votes": 4, "replies": [ { "id": 2229320, "postDate": "2023-04-21T09:08:47.443Z", "content": "

Thx for the extra info😁

", "rawMarkdown": "Thx for the extra info😁", "votes": 1 } ] }, { "id": 2234836, "postDate": "2023-04-25T14:16:00.793Z", "content": "

Good catches, thank you! These are clear mistakes. I've updated the Frag1 and Frag3 label images on my end and another organizer will likely post them soon.

", "rawMarkdown": "Good catches, thank you! These are clear mistakes. I've updated the Frag1 and Frag3 label images on my end and another organizer will likely post them soon.", "votes": 2 }, { "id": 2231037, "postDate": "2023-04-23T03:32:13.227Z", "rawMarkdown": "", "votes": -1, "isDeleted": true } ], "comments": [ { "id": 2232081, "author_name": "Isaac B", "author_url": "", "post_date": "2023-04-24T02:05:06.297000", "content": "

Interesting find. Morhpological operations should be able remove any small target pixels in the truth masks. Maybe it's worth trying to see how it affects training?

", "votes": 3, "replies": [ { "id": 2232084, "author_name": "JavaZero", "author_url": "", "post_date": "2023-04-24T02:13:46.110000", "content": "

thx a lot. I will finish the experiment. (But I dont think it will affect a lot😂)

", "votes": 1, "replies": [] } ] }, { "id": 2229315, "author_name": "Séraphin Lampion", "author_url": "", "post_date": "2023-04-21T08:53:38.813000", "content": "

yeah, i saw that too, another one that im not sure about is this one on fragment 3 which is way larger:
\nf3
\nf3

", "votes": 4, "replies": [ { "id": 2229320, "author_name": "JavaZero", "author_url": "", "post_date": "2023-04-21T09:08:47.443000", "content": "

Thx for the extra info😁

", "votes": 1, "replies": [] } ] }, { "id": 2234836, "author_name": "Stephen Parsons", "author_url": "", "post_date": "2023-04-25T14:16:00.793000", "content": "

Good catches, thank you! These are clear mistakes. I've updated the Frag1 and Frag3 label images on my end and another organizer will likely post them soon.

", "votes": 2, "replies": [] }, { "id": 2231037, "author_name": "", "author_url": "", "post_date": "2023-04-23T03:32:13.227000", "content": "", "votes": -1, "replies": [] } ], "raw_markdown_by_id": { "2229308": "![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F8314415%2F883b9725a79c079e916cde646d9ff0e8%2Fink_label_error.gif?generation=1682066308329795&alt=media)\n\nIt seems that the label is incorrect👀. I have uploaded a fixed label image. 💯\n\nThanks to @mecndo for the discovery. 😍", "2232081": "Interesting find. [Morhpological](https://docs.opencv.org/4.x/d9/d61/tutorial_py_morphological_ops.html) operations should be able remove any small target pixels in the truth masks. Maybe it's worth trying to see how it affects training?", "2229315": "yeah, i saw that too, another one that im not sure about is this one on fragment 3 which is way larger: \n[f3](https://i.postimg.cc/rFzDMJNY/Screenshot-from-2023-04-21-11-05-35.png)\n[f3](https://i.postimg.cc/KvPdKrzm/000.png)", "2234836": "Good catches, thank you! These are clear mistakes. I've updated the Frag1 and Frag3 label images on my end and another organizer will likely post them soon.", "2231037": "" } }