FWIW I voted that its ok to delete + not redraw immediately, but I think for me it also depends on how far “out” they are. If it’s just a detached house or a block of flats that’s a bit skewhiff, then fine i’d adjust it accordingly because it’s not taking up time. However, if it’s something like a row of terraces or something so far out to the point where it’s easier to delete and redraw then I may as well delete the thing (and bear in mind i primarily use iD and I don’t plan on changing to JOSM).
This was my main reason for voting for deleting immediately without replacing - especially for buildings IMO we’ve got to have quality over quantity so that these slop buildings aren’t sitting there for years waiting for somebody to come and sort them out. I’m not going to repeat specific points already raised in this thread because I have nothing new to add.
Also, we have the cadastral parcels overlay at our disposal. It is available on Rapid. There is no excuse. If people are adding hundreds and thousands of AI slop terraces and semis (as per the original post) and seemingly not making an attempt at splitting or aligning to cadastral, surely this is verges on a case of not following the import guidelines as mappers should be taking the time to review and map them properly?
I voted to delete without redrawing as splitting buildings will likely be very off putting to any new mappers. I’m relatively new to mapping and have tried splitting buildings in iD. It’s not intuitive.
I can’t imagine deciding to start mapping my local area with iD and/or streetcomplete and finding that most buildings need splitting. That feels more demotivating to me than starting from scratch.
While JOSM may have tools to make splitting buildings fairly trivial I think it’s worth considering if JOSM users want to go around fixing the majority of these merged buildings.
I am actually on the fence. Straight up deleting poorly drawn buildings (within reason) can sometimes harm the map data, especially if the buildings have other useful data.
But I think this makes sense for extreme cases. I will never not share this image (despite not being AI, but part of a high volume = better mapping task group) as I can’t imagine the presence of these buildings helps the map, they likely damage it from poor quality and people using it seeing nothing but these, assuming its broken:
As a counterpoint, these buildings were added over 10 years ago, and not touched since, so I wonder if there would actually have been anything better added in the intervening period.
I think these are still able to be incrementally improved.
This is the result of a couple of minutes in JOSM. It was split at the inflection points manually, the corners aligned, orthogonalised, then split using a custom terracer plugin that handles wonky shapes. The geometry was actually quite close, so has generally split buildings where you would expect.
It’s not perfect and the original building didn’t have the crenelations to the terrace at one point (so the split doesn’t either), but would be a good starting point for someone to add house numbers using Street Complete, for example.
I’m waiting for someone to tidy up this shape. I’ve tried to represent the building appropriately. I wonder what AI would make of the aerial imagery. Way: Hux Shard (1391774013) | OpenStreetMap
I fully agree. It has been a decade, but because too many people use the mantra of “some data is better than no data” it has laid dormant all that time.
Thanks to your work fixing it though, it has made far better and more stable data for those who will come to modify the data after you
though, I can now no longer share it as an example :(
It’s a plugin I’ve been pottering on for the last couple of weeks - not quite ready to be published yet, but will hopefully be soon.
I have seen some AI buildings that are supposed to be rectangular look not dissimilar to that, where there is a reflection in the imagery. The architect was clearly enjoying themselves.
Having low quality data is a double edged sword - I agree.
I know Rapid already has some guard rails, but maybe it needs a way of directing users to more accurate aligning and splitting of things they’re adding (or of existing data, though this probably isn’t best suited to rapid/Id).
Possibly a bit late to the party but would it be possible to import the Ordnance Survey GIS buildings into OSM as the basic property details. I realise they are a simplified form of properties but the lines follow some aspect of the facades and anchor the shapes in the real world and may provide a more reliable base pending further updates by splitting individual properties ? They would also be compatible with other maps.
I would agree that if AI imports are based on google there will be many inaccuracies, both in location and number and apart from the indication of a built up area have little merit in terms of detail. The OS GIS based property details are available for public use unlike the building forms in Mastermap which “should” be accurate but are subject to copyright and require payment(s) to OS.
Whilst accurately mapping buildings is the preferred scenario an indication of built up areas and rural buildings is helpful so long as subsequent accurate mapping of the buildings does not involve additional work. Possibly some people may value their own vanities but if wholesale deletion improves accuracy this should not be resisted and a more reliable update should not provoke a negative response.
In some areas with mostly pre-war housing stock and out of copyright OS 1:2500/1:5000 maps, automated tracing of building outlines could be a good start. They’d need to be checked carefully against current aerial imagery to ensure that they’re still there and the outlines haven’t changed substantially.
I’d be wary of that. I have traced quite a few buildings on OSM in my mapper life to date, and in places with predominantly “standard” square buildings I can easily do 2000 buildings(*) while watching a film in the evening. If a few more people were to do that occasionally, you’d have a better quality building data set than if you auto-trace buildings. I don’t know why it doesn’t happen in the UK, I have a feeling everyone is waiting for an import that will eventually come but never does. And for me personally, while tracing buildings is a bit repetetive, I do occasionally enjoy it - but I would certainly not enjoy cleaning up after some auto-tracing software. (Or manually breaking up OS-sourced building “blocks”.) So - why not just roll up one’s sleeves and get going.
(*) Using the JOSM building plugin and, where necessary, extrusion mode. Building plugin lets you draw a building with 2 clicks when it has the same angle as current reference building, or three clicks and one press of the shift key otherwise.
By OS GIS data, do you mean OS OpenMaps? If so, please have a look here Way: 1367948873 | OpenStreetMap . I don’t like this imported data, but I’m not going spend time adding detail now. I was mapping from aerial imagery when this was imported. To me, the difference is significant when compared to some nearby manually mapped buildings.
IMO these are worse than the Bing ones. The only benefit to the OS outlines is that they won’t be offset, but I think that is more than offset by them being purposely generalised to the point of uselessness.
I think they might even impede a ground survey of e.g. addresses. With GPS accuracy being what it is in urban areas I often rely on the shapes and relative positions of the building being roughly correct to determine which building on the map actually corresponds to the address I’m looking at. If there is no building at all I at least have the confidence that dropping a node in roughly the right place is improving things. If the building is there, but made deliberately terrible I don’t know which building to label and am more likely to skip it as I don’t want to put the number on the wrong house as that can cause headaches for future mappers.
I can’t speak for anyone else, but in my own experience the imagery in the UK seems particularly muddy. I started tracing buildings in Gloucester a few years ago and found it a far less enjoyable experience than doing the same thing in my other areas of interest.
Of course in the back of my mind there was also the knowledge that the data definitely already exists and could be brought in if the national mapping agency wasn’t set up to combine the worst aspect of private companies and government departments, but I think it was mostly the fact that I was squinting at blurry low contrast imagery.
Also decently mapping terraces can be really annoying with the current tools.
Yes, the GIS I refer to is OS OpenMap Local – a screenshot of the latest file is attached. I havnt done an overlay with your referenced shot but it looks much the same with a bit of tidying and omission of some small buildings – possibly sheds or garages.
I have also included a screenshot of another “online source” which is similar to MasterMap and I now appreciate the problem. I used the GIS OpenMap in an established area (not related to OSM) and the experience was reliable and the GIS provided a good background to develop the detail.
