Hi there, after continue my contribution to Mapillary after years, I see huge steps forwards. But also I found that a lot of features were added, but the overall feeling like a unified workflow? As this affects different projects and it’s not 1:1 suggestion to implementations, I think it’s better to post it here, than splitting it up into different github issues. But I will split it into several posts, so you show your support, or give threaded comments
All in all, this ideas focus on a improved user for different user groups
regular viewers
regular contributors
powerusers viewing
new users
And of course, this focus all the global entities
photos
sequence
tags
object detections
users
@mapillary it would be great, to get any feedback, if this ideas can be realized and if we should discuss more details
Currently the editing features seem to be spread on very different places and have very different behaviour, which looks not straight forward and hold users from (learing to) contributing
I used the existing tool to work on a dozen tracks (existing and my own ones). While the basic features are ok, I noticed that speed is very crucial:
slow when moving nodes (~ 3 secs)
slow when switching active node and photo upates ( 1 sec)
Better general design needed (offline? desktop app?) when editing hundreds of nodes for QA (if needed, I can post more on my personal experiences and workflow)
grabbing nodes doesn’t work 100% accurate (looks like race condition between focus and mouse move?)
selecting multiple nodes to move nodes / normalize camera along track / … doesn’t exist
normalize camera is only for full sequence, but not selected nodes only
MRU for last used tags, camera degrees, …
Voice control? to avoid context switches and browsing long lists / entering text using keyboard / … (e.g. selecting tags texts, rotating photo camera, … ). I did some research and prototype for JOSM voice control.
At work, I’m part of the city government GIS team and see good chances to contribute with more people / vehicles / … But I miss some aspects to organise a team with a joint mission:
Workflow for teams → pool of tasks and photos and managers
plan track records on map and assign to / let be picked by field teams
n users / teams which submit photos by app, dashcams, …
n users / teams which do privacy review / QA / classification before publication
show recorded tracks AFAP (to avoid dublicated tracks)
app could show planned recording track via argumented reality within video-stream (check what is recorded, while keeping focused on where to drive)
keep material more easily also as local copy (photos, QA improvements, ) → avoiding vendor lockin, if mapillary might go down (hopefully not!) and make sure we spend resources sustainably
how can organisations help to improve object detection and see how their contributions improve the detection?
app upload nodes + thumbnails first → faster to continue on editing / filtering the contributed tracks before the heavy photo bulk upload finishes (think of poor ADSL uplinks!)
The detection of city furnitures and amenities is one of the important features esp. for governmental endusers. So improving and fixing wrong detection seems to be very important, but is currently impossible?
mix of tag and object detection view, to let users report missing / wrong classification
How to flag wrong object classification? (pick right class, merge with surrounding objects, draw better outline, …)
show quality of classifier → gamification
tags show on all sequences, which show this location → verify this occurence and focus on non detected objects
Mapillary is based on community to take photos and classify content. It would be nice, if there are better features to help to keep this team together and improve the quality of contributions:
see local contributors within area
directmessages to users than commenting single photos
monitor local contributions (soft review) → Dashboard / notifications for new tracks and tracks that have been reviewed (e.g. with sequence tag for geometry, photo, tagging review?)
While I like the universally mind of tags, I think there is more power to come by extending this feature
Let users suggest new tags via UI
tags with explaination (like stackoverflow, link to wikidata like OSM tag wiki and example Photo ). Desribing what it it is not → homogenize global usage
describe how a tag should be marked - biggest possible outline or minimal? What if multiple objects in a row e.g. bollards? …
which tags are extracted by OSM and don’t need to be entered manually?
which tags are you currently looking for?
better campaigning and stats on tags / tagging community see voice.mozilla.org
gamification like maproulette challenges → e.g. lets classify bollards → tag them in your area or do a review of already flagged photos
Currently the node density on the maps is overwhelming and hard to pick good choice. I think you can improve it:
360° Images / panoramas as single stills, not as (pretty sparse and more messing than useful) track e.g. here in Rostock
cluster and color tracks on well covered major roads (e.g. by vehicle, by lane, by age…)
tracks as linestrings only, without thick nodes (just highlight a node as mouse hover)
global map is just green → not useful to pick a photo anymore. How about to switch to a heatmap → show coverage or curated pick of famous / unknown / community choosen locations like bestofosm → beam there
I will be honest to say I haven’t read it all. It’s obvious you put in a lot of time, but reading it globally I wonder… could you pick a few highlights for starters? What would you love mostly to be implemented first?