Hello. Thank you for organizing this interesting competition.
\nThere is one thing I would like to check before participating.
\nIs hand labeling and/or editing of training data allowed?
\n\nWe can see that the majority of recordings does not have an annotation of background species. Yet, it is highly likely that most of them actually contain one or more additional species.
\n
I'm not sure if this will actually be necessary, but I asked the question because I found the above statement in notebook(BirdCLEF2021: Exploring the data).
", "messages": [ { "id": "1261332", "postDate": "04/03/2021 00:43:43", "content": "Hello. Thank you for organizing this interesting competition.
\nThere is one thing I would like to check before participating.
\nIs hand labeling and/or editing of training data allowed?
\n\nWe can see that the majority of recordings does not have an annotation of background species. Yet, it is highly likely that most of them actually contain one or more additional species.
\n
I'm not sure if this will actually be necessary, but I asked the question because I found the above statement in notebook(BirdCLEF2021: Exploring the data).
", "rawMarkdown": "Hello. Thank you for organizing this interesting competition.\nThere is one thing I would like to check before participating.\nIs hand labeling and/or editing of training data allowed?\n\n> We can see that the majority of recordings does not have an annotation of background species. Yet, **it is highly likely that most of them actually contain one or more additional species.**\n\nI'm not sure if this will actually be necessary, but I asked the question because I found the above statement in notebook(BirdCLEF2021: Exploring the data).", "votes": null }, { "id": "1261677", "postDate": "04/03/2021 10:42:30", "content": "Usually with the training data you can do whatever you like to leverage as much information is possible - and yes hand/autom labelling is also allowed. You can confirm this by looking to past similar competitions (RFCX, eBird etc.)
", "rawMarkdown": "Usually with the training data you can do whatever you like to leverage as much information is possible - and yes hand/autom labelling is also allowed. You can confirm this by looking to past similar competitions (RFCX, eBird etc.)", "votes": null }, { "id": "1261726", "postDate": "04/03/2021 11:38:46", "content": "Yes, it is allowed. That being said, creating strong labels is extremely time consuming. Our work often focuses on many species with a large vocal repertoire. Because of time and personnel constraints, we frequently work with weakly labeled data.
", "rawMarkdown": "Yes, it is allowed. That being said, creating strong labels is extremely time consuming. Our work often focuses on many species with a large vocal repertoire. Because of time and personnel constraints, we frequently work with weakly labeled data.", "votes": null }, { "id": "1262250", "postDate": "04/04/2021 02:39:44", "content": "Thank you for the quick response and clarifying the rules.
", "rawMarkdown": "Thank you for the quick response and clarifying the rules.", "votes": null }, { "id": "1265281", "postDate": "04/06/2021 18:57:16", "content": "Just to add, if you do go to the trouble of hand-segmenting the training data, please consider releasing your labels at the end of the competition! (With bonus brownie points for including a few notes on your methodology.) The segmented data will be helpful for future researchers.
", "rawMarkdown": "Just to add, if you *do* go to the trouble of hand-segmenting the training data, please consider releasing your labels at the end of the competition! (With bonus brownie points for including a few notes on your methodology.) The segmented data will be helpful for future researchers.", "votes": null }, { "id": "1269271", "postDate": "04/10/2021 11:28:59", "content": "Thank you for comment.
\nOK. I don't know if I'll do hand-segmenting the training data yet, but when I do, I'll release it!
Usually with the training data you can do whatever you like to leverage as much information is possible - and yes hand/autom labelling is also allowed. You can confirm this by looking to past similar competitions (RFCX, eBird etc.)
", "votes": null, "replies": [] }, { "id": 1261726, "author_name": "holgerklinck", "author_url": "", "post_date": "04/03/2021 11:38:46", "content": "Yes, it is allowed. That being said, creating strong labels is extremely time consuming. Our work often focuses on many species with a large vocal repertoire. Because of time and personnel constraints, we frequently work with weakly labeled data.
", "votes": null, "replies": [ { "id": 1262250, "author_name": "naoism", "author_url": "", "post_date": "04/04/2021 02:39:44", "content": "Thank you for the quick response and clarifying the rules.
", "votes": null, "replies": [] }, { "id": 1265281, "author_name": "tomdenton", "author_url": "", "post_date": "04/06/2021 18:57:16", "content": "Just to add, if you do go to the trouble of hand-segmenting the training data, please consider releasing your labels at the end of the competition! (With bonus brownie points for including a few notes on your methodology.) The segmented data will be helpful for future researchers.
", "votes": null, "replies": [] }, { "id": 1269271, "author_name": "naoism", "author_url": "", "post_date": "04/10/2021 11:28:59", "content": "Thank you for comment.
\nOK. I don't know if I'll do hand-segmenting the training data yet, but when I do, I'll release it!