Dumb Question Department: Rear-Facing Camera?

If I were to mount a separate GoPro on the roof of my car facing rearwards, would that produce images as useful as if they were captured moving forwards?

Actually, now that I think about it, I already have a GPS-equipped, dual-channel StreetGuardian dashcam that produces good quality video. If I can convert the rear video output to sequential JPEG’s with the required metadata, might that be worth the bother of processing and uploading?

I prefer the external GoPro to the dashcam for the front view because my car has a nose that would do Jimmy Durante proud; and pointing the camera high enough for the nose to not dominate the raster would reduce its usefulness for its primary purpose of documenting idiot drivers and suicidal deer. But the rear camera has no such obstructions.

I’m also curious about how much of a hassle uploading using the GoPro Fusion currently is. Has someone combined the workflow in such a way that I could, say, dump all the images in a folder, let the computer munch on them whilst I sleep, and then automatically upload them when they’re thoroughly munched?

And while we’re asking dumb questions, is there such a thing as a supported 360 camera that’s non-spherical? I’m a pilot, so I like skies as much as the next guy. But using almost half the storage and bandwidth to capture imagery of the sky and the roof of my car doesn’t seem like the most efficient use of resources.



1 Like

There are already sequences that are rear-facing and they seem to work fine as long as you got the position/heading correct - @gpsmapper had some examples in another thread.

There are also a couple of threads on the Fusion and its workflow, and while I still need to do a proper upload of mine, it looks like it’s fairly easy as Export->Stitch in Fusion studio->set offset for the center of the image to be facing the direction of travel, depending on how you mount the camera->normalise sequence for headings->upload. Fusion is one of the better and easier to work with for Mapillary.

I don’t think any of the 360 cameras are “supported” as such, Mapillary don’t have enough resources for that. So if you managed to get a 360 equirect that does not shoot the sky I don’t see how it wouldn’t work with mapillary. Can only think of a normal camera+lens attachment that would do that though.
Or you could try getting multiple action cams and arranging them so they cover 360, getting better image quality, sometimes cheaper than a 360 setup


Okay, thank you. That’s helpful.

For the sake of simplicity, I’m leaning toward just getting another GoPro Hero and mounting it facing aft. My reasoning is that it will prevent me from having to drive each route twice, which is helpful because I tend to combine errands into routes that often wind up being big circles.

Also, most of the roads in question are extremely rural, so the side views aren’t very important. I mean, really, we’re talking trees and farmland. The road itself is what’s most important.

The other consideration is the bandwidth for the uploads. I regularly exceed my cable Internet caps, but my ISP has never made a fuss over it (probably because I also pay my bill on time every month and rarely bother them). But I don’t want to push the envelope too much. I have to do some research on the comparative bandwidth using the GoPro Fusion versus a second Hero.

On the other hand, I also have a grandfathered “unlimited” 4G hotspot account with Verizon that actually works amazingly well considering that I live in the boonies. I use it mainly for backup Internet, so it doesn’t get a huge amount of use. Maybe I should upload the next sequence or two using the hotspot and see whether it ruffles VZW’s feathers.

I also have a beast of a laptop that I rarely use that could be used for this sort of work without tying up my “work” computer. I also administer three Linux servers, so if Linux tools make for better workflow, I can upload the files to one of the servers and process them there. (I could also install Linux on the laptop, for that matter.)

So the Fusion 360 isn’t out of the running yet. If VZW doesn’t vomit on the humongous uploads and I use the laptop or one of the servers for whatever processing is needed, using the Fusion might actually be easier (and probably would be cheaper, since they’re basically giving them away right now) than using a second Hero.

Thanks again. This gives me some more to think about. If nothing else, maybe it will help stave off senility for a few more days.


Loosely related

jpg compression will make large amounts of sky area small in byte size. Put some trees in it and the filesize gets a lot larger. My 4K road images range from 600KB to 1900KB but that is at a (too high) quality.

mp4 to jpg conversion also creates artifacts that dont compress well. It’s a bit rate, frame rate thing that may be worth experimenting with.

The tools will convert video to jpg’s. At the moment only the GoPro and BlackVue is supported for reading video GPS data. Other devices need a gpx or nmea file as well. Your GoPro command line will be similar to my BlackVue one. (Debian Linux) I “process” each days take with a script, then upload unattended/remotely either as a batch or day at a time. It has become quite painless for me to upload 100,000-200,000 images that way. I could just throw all the mp4’s in a folder and just let it run if I wanted to…


@StephaneP built a 4 camera custom rig that creates images towards what you want I think, I saw an image of his here:

On his website he has a complete instruction how to build one:

In my opinion “360” is the way to go here :muscle:
I have bought a “YI360” (cheapest I found was $199 at Amazon) and am in the process of creating an “auto mucher” :wink:
When I’ve got it “all done” and you’re interested I could try and create a tool for you?
(OS? Linux would be ideal, but I think I can get it to work on Windows also…)

1 Like

Thank you.

I think I’m going to test the uploads using the VZW hotspot before I decide what to do next. If it doesn’t vomit or get throttled on huge, sustained uploads, that will solve the bandwidth part of the dilemma.

I’m inclined to think it will be fine because we have very few people feeding off VZW’s tower here. I live in a rural village of < 500 full-time residents, but we have two fairly important state highways that merge here for a few miles. Because of that, we have 4G service with VZW, ATT, and TMO (the three major mobile carriers in the US), which means that none of the towers are very busy.

I’ve never experienced any apparent throttling or deprioritization using the VZW hotspot. But I also have never done sustained uploads of thousands of JPG’s. I have to test that first. Unfortunately, testing the upload means having something to test it with, which seems unlikely for a few days because of the weather.

I have machines running Windows 8.1 and Windows 10 Pro in the office, as well as an older machine running Debian that I use as a file server and testing server. I doubt it could handle anything processor-intensive. I’m also root on three high-end CentOS servers that I access remotely, and I could build, virtualize, or repurpose a more powerful Linux box for the office if I needed to.

But as I said, I do want to test the uploading on the hotspot first because that will factor into my decision. So thank you, but please don’t do any coding just for me just yet. I appreciate your offer, though.


i throw all my photos into lightroom and then export them at 75% quality which nearly halves the filesize (the math doesnt make sense) but even pixel peeping i couldnt really tell the different between the original and the 75% version

another thing i do is upload using vivaldi (chrome) and use the dev tools to throttle the upload speed. i have it limited to 300kbs most of the time and some nights ill switch to 550kbs if i want to get something uploaded quicker. something like that might help if you want to be sure to stay under the radar from your ISP. (the main reason i throttle is so i can still do other things on the internet while photos are uploading)

also, remote desktop software like anyDesk is useful if you want to connect to your laptop from your work computer or phone to add new sequences or change the throttling speed

1 Like

Thank you. What camera are you using?

I really didn’t think about batch processing. I don’t use Lightroom (nor anything else Adobe makes) since they went to a subscription plan, but I have other tools that can do that. And yes, even knocking a very small percentage from the quality of a JPG dramatically reduces the file size with little or no degradation that is visible to the human eye.

I do have RDP set up on Windows and SSH on Linux (along with Web interfaces for some things), so remote access is well-covered. Thanks for the suggestion, though. My preferred device for RDP / SSH / SCP / etc. is actually my Amazon Fire HD 10 with a BT keyboard. I’m too old to be tapping on phones.


mostly a sony xz1 compact and then a gopro and a mi sphere depending on if theyre needed

i must try something else like gimp to reduce the file sizes. i just had lightroom from years ago but its a pain to use sometimes since you have to import photos before you can work on them

ha, sounds like a good setup, much better than trying to control a remote desktop on a tiny screen like the xz1 compact

1 Like

imagemagick does command line quality (and a whole lot other) changes. “mogrify” accepts wildcards. eg for me to change quality on my todays take of 25,000 images;

mogrify -quality 85 /home/user/Mapillary/20190607[ab]BV/images/mapillary_sampled_video_frames/*/*jpg


I ran today’s images through Affinity Photo as a batch file with instructions to reduce the quality but retain and embed the metadata. It seemed to work so far as the upload was concerned: It did map the route correctly. But the 5787 images are still “Pending” hours later. So maybe it didn’t work as well as I thought.

I’m wondering if there’s a time conflict between the time the picture was taken (which was preserved in the metadata) and the time the file was created (which would be whenever it was processed by Affinity Photo). Or maybe there’s just a backlog and the images are fine. I suppose I’ll know by morning.

On a more positive note, I uploaded all the images (about 11 gig worth) using the Verizon 4G Hotspot, which didn’t complain in the least. In fact, it may have been a bit faster than my Cable Internet connection.


1 Like

probably just a backlog. sometimes they seem to get processed quickly and other times not so much, maybe it depends on how much other people are uploading at the same time. i think ive noticed 360 photos taking a bit longer to process, which makes sense. maybe if the more objects in your photos the longer it will take as well?

i have photos i took a week ago that still arent showing up on the map but some photos i took a few days ago are… so i dont know whats going on behind the scenes

1 Like

Yeah, a backlog was all it was. By this morning nearly all of them had been processed.

Affinity Photo does a pretty nice job in terms of preserving quality. It doesn’t reduce file size as much as some more aggressive utilities, but signs and the like remain readable. It also didn’t vomit on a batch of almost 6,000 images. That was on a PC with an i7 and 32GB of RAM. It pegged the CPU at 100 percent for a while, but eventually got through the batch with no hiccups.

The only caveat is that even if the “Strip Metadata” macro isn’t selected, you still have to check “Embed Metadata” in the batch definition. It’s also good to define a separate folder to save to, otherwise it will overwrite the originals without prompting.

Affinity Photo is what I use for almost all photo and image editing now that Adobe has gone subscription-only. Once I got used to it, I found I liked it better than Photoshop and Fireworks. It has a pretty shallow learning curve, too. The batch function was just enhanced in the brand-new release, so this was also a good chance to test that out.

Hi, rear images is useful, not only the navigation on Mapillary website would be better, the point cloud would have more detail.

the workflow is exactly the same as a “regular sequence”. the only thing you have to do, if using the Mapillary Uploader, is set the OFFSET at 180. Or once the imagery is processed on Mapillary website, go to edit and apply a manual 180 OFFSET to the sequence.
Rear sequences is useful on one way streets. If your car has a heating grid on the rear glass, do recommend using a gopro outside the car, or the sequence will look crappy (i tried it, and those lines in the rear glass ruins it).