[Imports] AI-generated building footprints import in Uganda

[kiggudde deogratias]

Dear list,

Earlier this year, the Uganda team of Humanitarian OpenStreetMap Team (HOT)
began the process of incorporating AI-generated building footprints of
three districts in Uganda - Kapelebyong, Butambala, and Amuria - into
OpenStreetMap. At the start of this exercise, our team used the RapID
platform and the MapwithAI plug-in in JOSM to upload this data. Later in
the process, our team transitioned to a different workflow, which enabled
mappers to process and clean the buildings dataset offline and then upload
the data through JOSM. Since RapID is used as an editor in OSM and the
buildings were directly available via RapID, our team mistakenly assumed
that no additional documentation would be needed to kick-start the process.
During the implementation of the second workflow, the DWG flagged the
exercise and informed our team that adding AI-generated building footprints
to OSM is considered an import and, because no discussions were facilitated
about the import through the mailing list, this exercise was in violation
of the import guidelines. As soon as our team was notified of the
violation, all mapping activities were halted and we began working to
develop the necessary project-specific import guidelines to be aligned with
the broader import guidelines.

The purpose of this message is to acknowledge our mistake and the genuine
oversight our team made in not considering this activity an import,
consequently violating the import guidelines process. Our intention is to
provide a clear and honest overview of when the import was carried out,
what it consisted of, how it was carried out, and what we still intend to
complete after approval to ensure there is transparency around the initial
import process and the steps we have taken since the violation was flagged
to become aligned with the import guidelines process.

Timeline

The process of incorporating building footprints for Uganda (using RapID)
began on 13 February 2020 and continued through 30 April 2020. When this
was flagged by the DWG on 30 April 2020 as an import, all activities were
halted. On 1 May 2020, a handful of quality checks were carried out to
ensure data was completely validated. From 1 May onwards, our team has been
working to develop and disseminate the import workflow according to the
import guidelines.

Purpose of the import

In early 2020, the Uganda Bureau of Statistics (UBOS) reached out to HOT
requesting an updated building footprint and roads dataset for the
districts of Kapelebyong, Amuria, and Butambala to support the institution
in planning and implementing a pilot household census exercise. These three
districts, in particular, will be part of the pilot household census
exercise planned to take place in early to mid-2021. UBOS plans to use the
building footprint dataset to estimate the amount of time and resources
(enumerators, tablets, etc.) needed to execute the pilot household census
exercise. The AI-generated building footprints dataset itself was created
by Microsoft in 2018 and given their continued interest in supporting the
thriving OSM ecosystem, HOT is facilitating the import of a subset of
buildings in a specified AOI that will directly benefit local data
end-users, including the UBOS. More information on how these AI-generated
buildings were created can be found at [1] & [2]. This dataset is licensed
under Microsoft under the Open Data Commons Open Database License (ODbL)

Process

In terms of the actual process, as described earlier in this note, our team
started the import process using RapID and the MapwithAI plug-in in JOSM,
but later switched to an offline process where data would be cleaned and
then uploaded to OSM through JOSM as this method was more easily adopted by
the team. Wiki documentation for this import was developed and published
[3], detailing the metadata and workflows executed to import the building
footprints into OSM.

On 23 June 2020, these import guidelines were shared with the talk-ug
mailing list to gather feedback and/or resistance to the activity from the
local community. As of today (7 July) we have not received any objections
or resistance to the import.

Next Steps

With the import halted and having shared the import guidelines with the
local OSM mailing list in Uganda, we are now sharing the import guidelines
with the global import mailing list. Once again, our team genuinely
apologizes for violating the import guidelines that are in place. We
welcome comments/feedback from this community about the steps we have since
taken thus far to ensure we’re being compliant with the formal process and
are keen to forge the way forward that’s accepted by its members.


My best,

Deogratius Kiggudde
OSM username: kiggudde

[1]
https://blogs.bing.com/maps/2019-09/microsoft-releases-18M-building-footprints-in-uganda-and-tanzania-to-enable-ai-assisted-mapping

[2] https://github.com/microsoft/Uganda-Tanzania-Building-Footprints

[3]https://wiki.openstreetmap.org/wiki/WikiProject_Uganda
Microsoft_building_imports_for_UBOS_pilot_census_preparation
<https://wiki.openstreetmap.org/w/index.php?title=WikiProject_Uganda/Microsoft_building_imports_for_UBOS_pilot_census_preparation&redirect=no>

--
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstreetmap.org/pipermail/imports/attachments/20200709/6b294a44/attachment.htm>