![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F295090%2F009a4ec0b7d10d986093a13846e15473%2FWX20190802-1809352x.png?generation=1564740638252590&alt=media\")
\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F295090%2Fbd62c554470d4e6a5eba9681086ab8ca%2FWX20190802-1810072x.png?generation=1564740688044742&alt=media\")
For example: in some pictures man are labeled as \"man\" while in some other pictures man are labeled as \"person\". Why are the labeling standards not uniform？\n\n
For example: in some pictures man are labeled as \"man\" while in some other pictures man are labeled as \"person\". Why are the labeling standards not uniform？\n\n
First of all, take a moment to have a look at the semantic class hierarchy, mentioned for instance in the Open Images dataset description page. According to this hierarchy, a Man is a Person as much as a Woman, a Boy and a Girl are.\nExhaustively labelling all images is a very intense task, so our friends at Google tried their best but some things are bound to slip through their net, and maybe some people whose age/sex was difficult to determine were just annotated as Person. There is also the possibility that the precise label chosen is just an artefact of the visualisation tool.\nAnyways, this should not count against you in the evaluation, due to the rather sophisticated metrics used in the challenge that are described here. Long story short: if things are set up properly (which might be tricky, still have to figure out the details) you are going to receive 'points' for identifying a Person as a Man with no penalties unless it has been explicitly annotated as e.g. a Woman, and you are not going to receive partial recognition for detecting a Man as a Person without further distinguishing his age/sex.
Thank you so much, really helpful👍
", "rawMarkdown": "Thank you so much, really helpful👍", "votes": null } ], "comments": [ { "id": 606332, "author_name": "lntsmn", "author_url": "", "post_date": "08/23/2019 12:57:28", "content": "First of all, take a moment to have a look at the semantic class hierarchy, mentioned for instance in the Open Images dataset description page. According to this hierarchy, a Man is a Person as much as a Woman, a Boy and a Girl are.\nExhaustively labelling all images is a very intense task, so our friends at Google tried their best but some things are bound to slip through their net, and maybe some people whose age/sex was difficult to determine were just annotated as Person. There is also the possibility that the precise label chosen is just an artefact of the visualisation tool.\nAnyways, this should not count against you in the evaluation, due to the rather sophisticated metrics used in the challenge that are described here. Long story short: if things are set up properly (which might be tricky, still have to figure out the details) you are going to receive 'points' for identifying a Person as a Man with no penalties unless it has been explicitly annotated as e.g. a Woman, and you are not going to receive partial recognition for detecting a Man as a Person without further distinguishing his age/sex.
Thank you so much, really helpful👍
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "590542": "For example: in some pictures man are labeled as \"man\" while in some other pictures man are labeled as \"person\". Why are the labeling standards not uniform？\n\n", "606332": "First of all, take a moment to have a look at the semantic class hierarchy, mentioned for instance in the [Open Images dataset description page](https://storage.googleapis.com/openimages/web/factsfigures.html). According to this hierarchy, a `Man` is a `Person` as much as a `Woman`, a `Boy` and a `Girl` are.\nExhaustively labelling all images is a very intense task, so our friends at Google tried their best but some things are bound to slip through their net, and maybe some people whose age/sex was difficult to determine were just annotated as `Person`. There is also the possibility that the precise label chosen is just an artefact of the visualisation tool.\nAnyways, this should not count against you in the evaluation, due to the rather sophisticated metrics used in the challenge that are described [here](https://storage.googleapis.com/openimages/web/evaluation.html). Long story short: if things are set up properly (which might be tricky, still have to figure out the details) you are going to receive 'points' for identifying a `Person` as a `Man` with no penalties unless it has been _explicitly_ annotated as e.g. a `Woman`, and you are not going to receive partial recognition for detecting a `Man` as a `Person` without further distinguishing his age/sex.", "630568": "Thank you so much, really helpful👍" }, "source": "meta" }