In San Francisco, several roads are marked as having sidewalks, which is normal and good. But I notice some (like Van Ness Avenue here) are marked with foot=yes as well. Even though you clearly shouldn’t be walking around in the actual busy street.

This seems pretty wrong? Being marked as having a sidewalk should be enough for routers. As a bonus, the sidewalk is double-mapped, and Van Ness should be marked sidewalk=separate, I think.

I see a lot of this in SF, which is why I’ve been hesitant to fix it. But I can fix this, right?

See my previous entry for installation instructions.

So far, mapillary_tools has worked very well for me! Images are sent to Mapillary MOSTLY facing the correct direction, with duplicates (from stopping at intersections, and such) being removed. The work I must do afterwards has been cut dramatically. This makes wandering to new places to capture images a fun adventure for me, and makes editing OSM with captured images + GPX a lot smoother of a process.

I capture images at 1 per second in a vehicle (any faster, and my GPS gets wonky), or 3 or 5 per second while hiking. I use pretty harsh duplicate detection, since my phone’s compass is terrible, and my GPS wanders anywhere in a 1m range while I’m stopped, given enough time. My favorite command to run is:

mapillary_tools process –import_path storage/dcim/Trip1 –cutoff_time 30 –duplicate_distance 1.5 –duplicate_angle 360 –interpolate_directions –offset_angle 0 –overwrite_EXIF_direction_tag –user_name jesseakaraccoon –advanced

Results look good!

The part “–interpolate directions” is like using “normalize sequence” while editing. It’s not perfect, but it’s great! I’ve reported a minor flaw, which causes some images at stops to point in odd directions.

You can install mapillary_tools on an Android device without rooting it. Here was my process:

Start by downloading Termux, a good command line program for Android.

Then, install mapillary_tools according to Linux instructions on the README. However, because Python 2 is no longer supported, the “pip” command seen in the readme will not work.

Use the command…

pkg install python2

…to install the necessary version of Python. Then follow instructions as-is, but replace instances of “pip install” with “python2 -m pip install”

Finally, on Termux, you must run the command “termux-setup-storage” to access the device’s internal storage.

As a non-programmer, this took me a while to figure out. I hope this helps someone!

What I WANT to do is take Mapillary photos with my phone’s wide angle lens. If I could find a way to easily do this, Mapillary would be a productive way to have fun and adventure! Maybe even a good excuse to buy a GoPro.

However, I have an Android phone and Android tablet. I don’t have a traditional desktop computer. I’ve found that Mapillary’s support for users without an Apple nor Windows device has been…completely minimal. And it’s frustrating me to the point of burnout.

In terms of editing OSM, images taken with the standard lens on the Mapillary Android app can be nearly useless sometimes. I’ve had plenty of instances where I’ve been unable to verify street names, other signage, etc. Gathering house addresses, names of businesses, and other features not DIRECTLY near the road is often impossible. By the time something is close enough to see well, it’s out of frame or blocked by the vehicle ahead of me.

Plus, aesthetically, having imagery that only looks straight down roads is super-unappealing. It squanders Mapillary’s ability to stitch nearby photos together.

The last few days, I’ve instead been taking pictures with my wide angle lens on OpenCamera. The results have been REALLY useful for OSM editing, and creating pleasing Mapillary image collections.

The problem is, the only option for uploading is the Web uploader – I can’t use the Desktop uploader.

So EVERY sequence I submit, I need to personally, MANUALLY hunt down duplicate images (from standing still at intersections) AND normalize the sequence. And make sure no sequence is too big. AND the web uploader behaves frustratingly when you’re using a touch screen instead of the mouse it expects you to use.

Anyone got a straightforward fix for this?

Aside from imports, about 99% of the features mapped in Fortuna were mapped by me. It’s an interesting feeling.

It feels like the local map is a personal creative project. Like building a model train…that, uh, helps other people navigate and travel.

The most difficult part is deciding what’s important – what counts as “progress.” One of the big problems I’ve fixed is mapping errors that make Osmand spaz out when you’re traveling along the 101 or Main Street.

I’m gravitating toward focusing on having correct lanes, turn lanes, nearby buildings, and nearby driveways along the most important roads. Then slowly focusing on less important roads as I make progress.

And osmose error hunting. That’s always fun!

I’m excited about the idea of getting Fortuna’s roads and trails entirely covered with Mapillary. It makes editing OSM much easier for me, being able to look back on any bit of road I’ve driven on. I hope that the machine learning Mapillary does can one day make OSM mapping much, much quicker.

My main frustration with Mapillary (Android app) is that it has a poor sense for where the camera is pointing. I tell it my camera points forward, but 1/5 of the pictures are like, “are you sure you weren’t pointing 150 degrees backward.”

No. Normalize sequence.