{ "id": 231619, "title": "Some doubts about the label", "url": "/competitions/plant-pathology-2021-fgvc8/discussion/231619", "author_name": "", "post_date": "2021-04-09T10:10:39.375055200Z", "votes": 7, "comment_count": 3, "views": 0, "content": "

in the official description it says that:

\n
\n

Unhealthy leaves with too many diseases to classify visually will have the complex class, and may also have a subset of the diseases identified.

\n
\n

So when we look into the data, we find out that there are some pictures labeled with only complex and nothing else, or complex with only one other disease, in which case I guess that the labels are wrong or incomplete. Will this also happen in test set? If so, dose it mean there are some pictures with multiple labels may be labeled wrong?

", "messages": [ { "id": "1268337", "postDate": "04/09/2021 10:10:39", "content": "

in the official description it says that:

\n
\n

Unhealthy leaves with too many diseases to classify visually will have the complex class, and may also have a subset of the diseases identified.

\n
\n

So when we look into the data, we find out that there are some pictures labeled with only complex and nothing else, or complex with only one other disease, in which case I guess that the labels are wrong or incomplete. Will this also happen in test set? If so, dose it mean there are some pictures with multiple labels may be labeled wrong?

", "rawMarkdown": "in the official description it says that: \n\n> Unhealthy leaves with too many diseases to classify visually will have the complex class, and may also have a subset of the diseases identified.\n\nSo when we look into the data, we find out that there are some pictures labeled with only `complex` and nothing else, or `complex` with only one other disease, in which case I guess that the labels are wrong or incomplete. Will this also happen in test set? If so, dose it mean there are some pictures with multiple labels may be labeled wrong?", "votes": null }, { "id": "1272568", "postDate": "04/13/2021 15:17:37", "content": "

@kewang777
\nI agree to your observation, but I think there's a simple workaround. The labels may be wrong, no doubt, but you can simply \"multi-hot\" encode them. That means, an image can have a label as [0,1,0,1,0,1], where first bit is healthy and last is complex. Hope this helps

", "rawMarkdown": "kewang777 \nI agree to your observation, but I think there's a simple workaround. The labels may be wrong, no doubt, but you can simply \"multi-hot\" encode them. That means, an image can have a label as [0,1,0,1,0,1], where first bit is healthy and last is complex. Hope this helps", "votes": null }, { "id": "1272635", "postDate": "04/13/2021 16:23:30", "content": "

I also think that definition of complex is quite vague.

", "rawMarkdown": "I also think that definition of `complex` is quite vague.", "votes": null }, { "id": "1282608", "postDate": "04/24/2021 05:50:57", "content": "

Let me explain complex is used when there are too many classes let say more than 3 diseases and one set of train.csv has 4 diseases classes then you may see 1 class and complex because 3diseases=1complex so <the disease>complex but another example may have only 3 diseases in that case complex . Lets take another example with 5 diseases then <the disease><another disease> complex . Now you may ask what if more than 2 diseases then complex and we have 5 diseases (Just an example)then as per the previous example there would be 3 diseases and a complex class there cant be more than 1 complex class.

", "rawMarkdown": "Let me explain complex is used when there are too many classes let say more than 3 diseases and one set of train.csv has 4 diseases classes then you may see 1 class and complex because 3diseases=1complex so `complex` but another example may have only 3 diseases in that case `complex` . Lets take another example with 5 diseases then ` complex` . Now you may ask what if more than 2 diseases then complex and we have 5 diseases (Just an example)then as per the previous example there would be 3 diseases and a complex class there cant be more than 1 complex class.", "votes": null } ], "comments": [ { "id": 1272568, "author_name": "adityakane", "author_url": "", "post_date": "04/13/2021 15:17:37", "content": "

@kewang777
\nI agree to your observation, but I think there's a simple workaround. The labels may be wrong, no doubt, but you can simply \"multi-hot\" encode them. That means, an image can have a label as [0,1,0,1,0,1], where first bit is healthy and last is complex. Hope this helps

", "votes": null, "replies": [] }, { "id": 1272635, "author_name": "atamazian", "author_url": "", "post_date": "04/13/2021 16:23:30", "content": "

I also think that definition of complex is quite vague.

", "votes": null, "replies": [] }, { "id": 1282608, "author_name": "swaralipibose", "author_url": "", "post_date": "04/24/2021 05:50:57", "content": "

Let me explain complex is used when there are too many classes let say more than 3 diseases and one set of train.csv has 4 diseases classes then you may see 1 class and complex because 3diseases=1complex so <the disease>complex but another example may have only 3 diseases in that case complex . Lets take another example with 5 diseases then <the disease><another disease> complex . Now you may ask what if more than 2 diseases then complex and we have 5 diseases (Just an example)then as per the previous example there would be 3 diseases and a complex class there cant be more than 1 complex class.

", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1268337": "in the official description it says that: \n\n> Unhealthy leaves with too many diseases to classify visually will have the complex class, and may also have a subset of the diseases identified.\n\nSo when we look into the data, we find out that there are some pictures labeled with only `complex` and nothing else, or `complex` with only one other disease, in which case I guess that the labels are wrong or incomplete. Will this also happen in test set? If so, dose it mean there are some pictures with multiple labels may be labeled wrong?", "1272568": "kewang777 \nI agree to your observation, but I think there's a simple workaround. The labels may be wrong, no doubt, but you can simply \"multi-hot\" encode them. That means, an image can have a label as [0,1,0,1,0,1], where first bit is healthy and last is complex. Hope this helps", "1272635": "I also think that definition of `complex` is quite vague.", "1282608": "Let me explain complex is used when there are too many classes let say more than 3 diseases and one set of train.csv has 4 diseases classes then you may see 1 class and complex because 3diseases=1complex so `complex` but another example may have only 3 diseases in that case `complex` . Lets take another example with 5 diseases then ` complex` . Now you may ask what if more than 2 diseases then complex and we have 5 diseases (Just an example)then as per the previous example there would be 3 diseases and a complex class there cant be more than 1 complex class." }, "source": "meta" }