What’s the deal with this filter? Travel Method “On foot” doesn’t find anything at all. Even though I know of a few tracks that are “On foot”.
It is effectively a speed filter, not really a travel or capture mode filter.
@Mapillary Generally speaking, most people would prefer things to be called by or named for what they are, not by wishful thinking. Aside from that, I think it would be better to focus time and energy on implementing far more usable and easier to implement filters first, like by resolution or camera make and model, than on something that is more difficult to implement, incomplete in the end, and confusing to users.
1 Like
My sequences on bicycle are the best.
1 Like
Cars with feet instead of wheels? What is their maximum speed on foot?
Hi folks - yes agreed, the filter is not working as intended quite yet - we are fixing/retraining and will have a more accurate version for you soon (cc: @caglarpmeta) . It’s not just based on speed, there is also a computer vision component - this will become a lot more accurate shortly. Thanks for the patience!
3 Likes
A dedicated “on foot” filter would actually be really useful for checking trails, pedestrian paths, parks, and old town areas. I’ve had a few cases where car imagery completely missed the details I was looking for.
1 Like
The filter works well for my sequences. When I apply the filter, I can see all the images I’ve taken myself, as well as those taken by others.
1 Like
I’ve only done a few sequences actually fully on foot, and this feature did correctly identify them. Not that that’s particularly impressive to calculate an average speed.
But I also have a lot of sections in the middle of bike trips where I walked for a while, sometimes for minutes. Some of the longer ones got identified correctly, but only when the vast majority of it was on foot. In other words, this obviously classifies the whole sequence and not the actual travel sections. So depending on your sequence length and image size, it will identify sequences differently. And it will just seemingly guess and assume if you happen to switch from walking to biking, for example.
I also have some mixed sequences where I biked through a lot of obstacles stopping, walking, pausing, etc. Some of these also got classified as walking even though none of the other sequences in the same trip were walking. A couple of them even were purely biking only with pauses (but no images during pauses).
So this is obviously just naively using average speed (and whatever vision there may be is obviously not able to tell it apart any better), so I imagine someone stuck in traffic will also get classified as walking.
Also, I don’t really see what the point is of offering one arbitrary “on foot” label when you could just offer an average speed limit filter. Or coloring by speed. Not unless the classification is actually good. Though how would it ever tell apart someone jogging versus someone slowly cycling, for example. It seems like a pointless exercise when you don’t offer any way to manually tag the sequences to correct it.
2 Likes