The training images for imc2023_hapier scenes for bike and chairs look like they are taken from the same deck. Some of the bike images have table and chairs in the background. Some of the chairs images have one or more bikes in the background. It does seem that the bikes and chairs are arranged differently in the bike images and chairs images. It might be that all of the bike images were taken at essentially the same time. And maybe all of the chairs images were taken at some different time.
But, looking at the pt_stpeters_stpauls images it seems that the stpauls images were taken at different times (different sky conditions, different people in image, etc.).
It seems to me what groups into a \"scene\" is applied differently in these instances. Using the stpauls images it seems the criteria of a scene are something like \"it's the same place\". But, for the bike and chairs images it seems like the criteria are more along the lines of \"everything in the scene is posed the same way\" (which would make many of the stpauls images be different scenes).
Is there any definite criteria for what makes a \"scene\"?
", "messages": [ { "id": "3171384", "postDate": "04/05/2025 16:28:35", "content": "The training images for imc2023_hapier scenes for bike and chairs look like they are taken from the same deck. Some of the bike images have table and chairs in the background. Some of the chairs images have one or more bikes in the background. It does seem that the bikes and chairs are arranged differently in the bike images and chairs images. It might be that all of the bike images were taken at essentially the same time. And maybe all of the chairs images were taken at some different time.
But, looking at the pt_stpeters_stpauls images it seems that the stpauls images were taken at different times (different sky conditions, different people in image, etc.).
It seems to me what groups into a \"scene\" is applied differently in these instances. Using the stpauls images it seems the criteria of a scene are something like \"it's the same place\". But, for the bike and chairs images it seems like the criteria are more along the lines of \"everything in the scene is posed the same way\" (which would make many of the stpauls images be different scenes).
Is there any definite criteria for what makes a \"scene\"?
", "rawMarkdown": "The training images for `imc2023_hapier` scenes for `bike` and `chairs` look like they are taken from the same deck. Some of the `bike` images have table and chairs in the background. Some of the `chairs` images have one or more bikes in the background. It does seem that the bikes and chairs are arranged differently in the `bike` images and `chairs` images. It might be that all of the `bike` images were taken at essentially the same time. And maybe all of the `chairs` images were taken at some different time.\n\nBut, looking at the `pt_stpeters_stpauls` images it seems that the `stpauls` images were taken at different times (different sky conditions, different people in image, etc.).\n\nIt seems to me what groups into a \"scene\" is applied differently in these instances. Using the `stpauls` images it seems the criteria of a scene are something like \"it's the same place\". But, for the `bike` and `chairs` images it seems like the criteria are more along the lines of \"everything in the scene is posed the same way\" (which would make many of the `stpauls` images be different scenes).\n\nIs there any definite criteria for what makes a \"scene\"?", "votes": null }, { "id": "3172786", "postDate": "04/07/2025 07:27:19", "content": "You're correct, that one's tricky. I was simply aggregating datasets from previous years together and I didn't realize that these two overlap somewhat (which was irrelevant in previous editions of the challenge). It's tricky to define bikes do move, whereas the table and chairs are stationary. I'll see if it's possible to remove one of them. In principle: feel free to remove this scene.
", "rawMarkdown": "You're correct, that one's tricky. I was simply aggregating datasets from previous years together and I didn't realize that these two overlap somewhat (which was irrelevant in previous editions of the challenge). It's tricky to define bikes do move, whereas the table and chairs are stationary. I'll see if it's possible to remove one of them. In principle: feel free to remove this scene.", "votes": null }, { "id": "3173319", "postDate": "04/07/2025 18:56:00", "content": "We kept the data as-is, but the documentation has been amended to reflect this fact.
", "rawMarkdown": "We kept the data as-is, but the documentation has been amended to reflect this fact.", "votes": null } ], "comments": [ { "id": 3172786, "author_name": "eduardtrulls", "author_url": "", "post_date": "04/07/2025 07:27:19", "content": "You're correct, that one's tricky. I was simply aggregating datasets from previous years together and I didn't realize that these two overlap somewhat (which was irrelevant in previous editions of the challenge). It's tricky to define bikes do move, whereas the table and chairs are stationary. I'll see if it's possible to remove one of them. In principle: feel free to remove this scene.
", "votes": null, "replies": [ { "id": 3173319, "author_name": "eduardtrulls", "author_url": "", "post_date": "04/07/2025 18:56:00", "content": "We kept the data as-is, but the documentation has been amended to reflect this fact.
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "3171384": "The training images for `imc2023_hapier` scenes for `bike` and `chairs` look like they are taken from the same deck. Some of the `bike` images have table and chairs in the background. Some of the `chairs` images have one or more bikes in the background. It does seem that the bikes and chairs are arranged differently in the `bike` images and `chairs` images. It might be that all of the `bike` images were taken at essentially the same time. And maybe all of the `chairs` images were taken at some different time.\n\nBut, looking at the `pt_stpeters_stpauls` images it seems that the `stpauls` images were taken at different times (different sky conditions, different people in image, etc.).\n\nIt seems to me what groups into a \"scene\" is applied differently in these instances. Using the `stpauls` images it seems the criteria of a scene are something like \"it's the same place\". But, for the `bike` and `chairs` images it seems like the criteria are more along the lines of \"everything in the scene is posed the same way\" (which would make many of the `stpauls` images be different scenes).\n\nIs there any definite criteria for what makes a \"scene\"?", "3172786": "You're correct, that one's tricky. I was simply aggregating datasets from previous years together and I didn't realize that these two overlap somewhat (which was irrelevant in previous editions of the challenge). It's tricky to define bikes do move, whereas the table and chairs are stationary. I'll see if it's possible to remove one of them. In principle: feel free to remove this scene.", "3173319": "We kept the data as-is, but the documentation has been amended to reflect this fact." }, "source": "meta" }