The original dataset is accepting contributions and annotation at https://mekabytes.com/dataset/info/billboards-signs-and-branding :)
The goal of this dataset is to be able to recognize billboards and popular corporate logos so they can be hidden in photos, and in the future so that they can be hidden using augmented reality.
We are settling on a maximalist approach where we would like to block all signage. This includes bus stop ads, store signs, those banners they have on street lights, etc.
Categories
🚧 Billboard - includes advertisements on bus benches and shelters, and the posters on building construction (think with scaffolding).
🏪 Signage - store names, signs on buildings, lists of businesses at a strip mall, also includes any small standalone advertisements like those campaign signs in people's yards or papers on telephone poles.
📦 Branding - logos and names on products, like a coffee cup or scooter, includes car badges.
Seeking Photos on https://mekabytes.com
Right now the images have been mostly collected in Los Angeles, CA. We would love some geographical variety!
If you have any questions about labeling, don't hesitate to leave a comment and check the checkbox to notify the mods.
We are light on branding photos, so pictures of products with logos and brands on them are greatly appreciated!
Version Info
Version: 2022-08-11T18:53:22Z
Type: bounding box
Images: 103
Annotations: 1351
Size (bytes): 315483844
- Downloads last month
- 0