[FEATURE request]

I really like Mapillary but what bothers me is sometimes the lack of quality in some pictures. Some images or image sequences are useless because they are too dark, too blurry, sometimes the distances between images are like 500m and some images are so tilted that they are kind of useless too.

The suggestion is that while processing the image for the data collection, the images should also be graded with a quality score ranging from 1-5 based on different factors like described above. Later, users can filter images based on the grade just like they can filter based on time. In my opinion, this should enhance the quality of mapillary because people will be motivated to create better images.

There are several reasons why the older pictures tend not to be so good. One of them is that we thought that many bad pictures would be synthesized into one good picture.


So the quality of the newer images are better? From which date did they get better, just for filtering.

There is no particular date. It is just a tendency.
There have been problems with the focus in the Google API that have been solved for example.
More and more people come to see that a smartphone on a bike is usually a bad idea.
And smartphone manufacturers keep on lying when they tell that they can shoot in low light.

But of course there are still mavericks and their defenders.

Why is it a bad idea? If you capture in mid-day when there is enough light, I don’t get the problem. And what is a better way of capturing then?

In mid day without vibrations OK, but even then you won’t be able to read the fine print.
But to please our viewers we prefer to use action cams.

Some signposts are surprisingly hard to capture, even with my top of the bill action camera.
I really should have stopped at every sign. Now someone must return because some pictures are unreadable.

I see. Maybe it is indeed not a good idea to capture with a normal phone. But we can of course not compete with for example Google streetview and Cyclomedia their quality and equipement. Of course, you want the best images for Mapillary but you can’t expect the best quality.

I have to note that also camera’s mounted on cars can produce some low quality image. So stating that using a bicycle is not good is not completely true in my opinion. If you have a stable camera and you can cycle straight then you can also produce good quality images.

it’s a trade off between quality and quantity.
“low quality” is also subjective

Dear Vincent (reply to message 8/9) : using an action cam on the handlebars of a bicycle will produce decent pics, as will any camera with decent optics and a short exposure time. Smartphones don’t necessarily expose at the shortest possible time, thus there is likely to be some blur - EDIT : both motion and because the image stabilisation may be optimised for shaking hands rather than the handlebars vibrating.

Second point to mention is that earlier cameras, and cameras by particular manufacturers, had lenses, sensors etc. which weren’t up to the job of capturing and storing images of the built environment quickly and in great detail - even though these may have produced an “ohhh / Woww” response from the audience viewing someone skiing down a hill.

Requirements for image stabilisation are different when on a ‘kasseistrook’ *- the cobblestone roads in some parts of Belgium; and to properly process all fine detail (including traffic signs) you’d need more processing power than for what’s essentially a white blanket (edited blanked to blanket) with some dark pine tree-shaped forms in the distance in the above skiing down hill experience. * hard to describe, just come on over, ride your bicycle along one at reasonable speed, and you’ll instantly feel what I mean - it’s that ‘thoroughly shaken’ feeling, an irregular shaking of varying intensity as each sett has it’s own character.

Dynamic range is one of the factors in the camera: the range of brightness between the brightest lit, and the darkest object - some cameras will expose for perfectly lit clouds because that’s what the manufacturers expect their audience would want : great clouds over a non-descript landscape (which will be blurry anyhow because of down hill speed), or a beautiful sunset while cycling somewhere.

Just my tuppence <

@Vincent50 : those (dark / blurry / infrequent) pictures can still be of use : I frequently ride a bicycle along back roads which - going by the date filter in Mapillary - haven’t been covered for several years : in that case, any pic showing that there’s still a sign which looks like the one photographed years ago confirms continued existence; similarly, a missing sign* - even if blurred - will show something has changed. * that could be a sign in the current, but not in the previous sequence, or a sign in the earlier but not in the recent sequence.

Then there are roads where the relevant authority never bothered to put up any signs : one picture every so often serves to show that the road was documented, there’s just ‘nothing to see here’ - or I may point the camera sideways to show autumn colours, or point it down to record a fine smooth or an atrocious road surface - as the case may be.

In itself, the camera used (an Olympus Tough TG-5) is capable of taking excellent pictures - on a tripod, but there is an issue with the thing sometimes dithering before deciding which picture mode to use, and the intelligent image stabilisation can’t always cope with ‘jerky’ movement - leading to ‘late’ and blurry pics.

EDIT : the camera produces a shutter sound and shows a blank screen for an instant in a way one would have expected from a 35mm film SLR, but those are indeed simulated, and do not coincide with the actual picture being taken; as confirmed when the shutter sound is set to off in a really quiet area : the screen gets blanked at a set interval after pressing the shutter, but when you listen really carefully there is a faint ‘click’ at a later moment, that interval being random and occasionally showing the camera being put back in the protective pouch when relying on the simulated sound to indicate the picture had been taken : that’s how long it may take for the camera to eventually take the picture, and obviously that will mean a blurred picture, which may still show more than no picture at all would, which is why it’ll be uploaded regardless. / end of edit /

It may still be helpful to grade pics, but then in combination with routing software which enables me to plan a route taking in several ways not yet covered, or covered long ago, or under adverse conditions, or with a camera which we’d now consider inadequate. And obviously allowing to specify walking / cycling / motor vehicle as means of transport. www.brouter.de offers such a routing engine, and it offers a Mapillary overlay, but doesn’t (yet) allow to limit ways shown to those not covered in the past two years.