Brainstorming ideas for the future of MapSwipe

danbjoseph · 2022-09-27T12:17:49.566Z

We’re looking to gather as many ideas as possible for changes, improvements, or additional functionalities people might like to see in MapSwipe! How could any part of the MapSwipe ecosystem be better, or what is missing? We’d love to hear your thoughts and ideas, please submit ideas using this: 1 question survey form

MapSwipe is a mobile app for Android and iOS for crowdsourcing map data. Originally it was used for a first pass in rural, mostly unmapped areas to identify the locations of buildings and roads. Those locations were then used to focus projects on the Tasking Manager so that mappers wouldn’t have to spend time panning across large swathes of forest and desert.

Right now there are three mission types:

Build area - users are shown a grid of six satellite imagery squares per screen and asked to tap the ones that contain a certain feature
Footprint validation - users are shown the outline of a feature from OSM over satellite imagery and asked to confirm the outline is correct and the feature exists
Change detection - users are shown a tile from 2 different imagery sources and asked to determine if some change occurred

mcliquid · 2022-09-27T17:11:40.063Z

I went to a lightning talk today on AI and machine learning. Therefore, I got the idea there whether MapSwipe could be the basis for machine learning “results”. So that the users in the app by answering in the app could also expand a sample set of image recognition databases at the same time.

Are there already thoughts about this?

danbjoseph · 2022-09-27T18:41:02.827Z

It’s an idea that has come up but we don’t have anything detailed out yet. A talk probably didn’t get too deep into the details, but I would be curious if you (or anyone else) has…

seen any existing examples or guidance on those types of workflows validating ML results?
particular types of ML results that are of most interest?
seen any write ups on the best (technical) ways to version and share an image recognition database?
seen any write ups on licensing and any ethical concerns related to image recognition databases?

Pedja · 2022-09-28T06:04:49.306Z

Creating Windows version would allow people to working in comfort of large screen, keyboard and mouse.

mcliquid · 2022-09-28T11:45:13.888Z

I guess a Web-App (Website) would be good enough for this usecase?

mcliquid · 2022-09-28T11:51:57.005Z

All in all I’m not very familiar with this!

danbjoseph:

seen any existing examples or guidance on those types of workflows validating ML results?

As far as I know, Google ReCaptcha responses are used for this. (Mark all bridges, bikes, hydrants, …). But of course this is not OpenSource and thus not valuable for this.