I found the degradation happened as soon as anything is plugged into the USB port. I tried various power sources over several tests, and found that it didn’t depend on the power at all. Even a cable with nothing on the end caused it.
Presumably it’s acting as an antenna and picking up noise (even when smothered with ferrite beads).
Worst case (long data cable plugged into power), either there’s no GPS or Desktop Uploader says “GPS too noisy”. Best case (breakout adapter to shortest possible power-only cable, ferrite beads, power off) I get reasonable GPS, but it drifts around and is nowhere near as accurate.
Here’s a portion of my test. Highlighted track is with nothing connected; others are with a cable attached. Some don’t appear as they’ve been rejected as missing or noisy GPS; the ones you see are the “good” ones:
I’m really impressed at how clear the images are despite the sun being low in the sky.
As you can see in this sequence, the GPS was quite unreliable despite being in a flat and open location. A series of images would be wrong, and then it would correct itself, then drift off again. I did wait for a GPS lock before starting. So not sure how concerned I should be.
Thank you for sharing your extensive testing and experiences. Since you have also tested a simple USB cable and a power bank that should output 5 V DC flat, then I am puzzled too. Maybe it is a firmware bug? Have you installed the GoPro Labs firmware?
You may also want to go with the Mapillary recommended settings, especially 25 fps for many reasons, one among which is that 24 fps are not true 24 fps out of the box but actually 23.976023976 fps.
Yes, avoid everything but 25 fps is my experience . The drift will be visible in the pictures/map.
I use always a USB connected to the Max 2 and a powerbank, never had any GPS issues. I use the passthrough, the device is maybe more shielded than remove the side cover. Also the QA of GoPro could be an issue, I already had 2 broken devices of the 6 devices I had.
Thanks. Maybe I’ve just got an overly-sensitive one
Regarding the video settings, I’ve clocked up several hundred miles at 24(×1000/1001)fps and haven’t seen any drift. I’ve just gone back and checked and images of motorway bridges etc match up exactly with the map, even after a long recording stretch. Have I been lucky, or is this not a problem any more? But I’ll switch to 25fps to be on the safe side…
When cycling, I use TW 5x, when walking, TW 2x. To prevent drift – which mainly occurs whilst walking – I start a recording of up to 10 seconds whilst walking. I then start the recording I want to use. I have Horizon Lock enabled.
As far as the sharpness of the images is concerned, I’ve set the bitrate to the maximum.
Having uploaded my imagery I find it best to have a quick check of the result to make sure I’ve got it right or to see if I could improve things (I now use a different warning flag on my trike as the original obscured too much.
My fingers are too fat! No cure for this I’m afraid. As well as the uploads you will see credited to me, there were several miles of footage I discarded because I’d somehow managed to change the settings, or on one memorable occassion left the lens caps on! I now check my progress after a fairly short distance before continuing.
Initially I couldn’t locate my upload score for London on the leaderboards - I’ve added a discussion topic on the forum at #CompletetheMap London - Greater London boundary - it may be worth checking which areas you are actually uploading for.
I wasn’t expecting to actually receive the camera and initially instructed DHL to leave the package it in my porch (I thought it was the low value gps cover that I was expecting). Luckily my neighbour took it into safe custody for me. Difficult one, because according to the messages I’ve received in the last month I’ve been a winner in various Tesco competitions, have won at least two phones, and am a finalist in various other unlikely competitions. If you get a vague message from DHL about a parcel you’re not expecting, be careful, as it might be something really good!
Many thanks for renewing my interest in uploads, which I also find very useful when updating OpenStreetMap.
@borisI recommend leaving this function off for MAX(2) cameras due to this function’s significant image quality penalty, and I suggest adding this note to the Mapillary Help article. 3D reconstruction already provides Mapillary with computed camera poses, which can be used by MapillaryJS to display all images (not just 360° ones) level with the horizon. Hence, there is no need for source digital horizon leveling rasterization; MapillaryJS just needs an update to make use of computed camera poses.
@GITNE, @Lowiekse - have you folks done comparisons with Horizon Lock on vs off to see the details of the “image quality penalty” - I’d love to take a look at the examples if you have them. That being said, @GITNE I like your idea of applying MJS horizon leveling - let me look into that!
I don’t really see any significant difference in quality. What I find more important is that the photos are level with the horizon. That’s better than photos that are completely crooked.
Sorry for creating confusion here. I have delved a bit deeper into this matter. You are correct @boris. The reorientation step in MAX(2) cameras happens at the coordinate level before sensor pixels are sampled and adjacent pixels are interpolated. In other words, reorientation happens always (even with horizon leveling off) before choosing which sensor pixels to sample for any particular target projection. It is just that when horizon leveling is off, the identity rotation matrix is applied. This is very smart of GoPro indeed, which makes MAX(2) cameras immune to this image quality degradation effect.
I have confused this for the rasterized equirectangular to equirectangular projection rotation image quality degradation effect, which some other cameras are indeed prone to. Afaik especially some older Insta 360 cameras.
Nevertheless, horizon tilted images could at least be displayed level with the horizon without the need for equirectangular to equirectangular resampling. So, Mapillary Help could give like a more general recommendation that horizon leveling should generally be left off because images are horizon leveled at display time anyway.
Haha. I’ve had older people asking what on earth is on top of my car. Younger adults sometimes pull a pose. And kids jump up and down and wave, some even waving me to stop so they can ask “Are we going to be on Google Maps?”.
I’m interested in the idea of magnetic Mapillary logos that I can attach to my car while I’m capturing imagery so that anyone curious knows that I’m doing mapping-related work and they can look up Mapillary to see what it is.
I’ve had one unpleasant experience where a car full of spoilt rich kids tailgated me for five minutes, laughing and pulling obscene gestures.
It attracts plenty of attention. Thankfully it’s mostly been positive attention so far.
Magnetic logos are really smart, I wonder if you could get Mapillary to grant them with some of the 360 bundles. It would share the platform and help people know we are doing this for good, and not harm.
I like the idea of magnetic logos! One thing we do have available now are Mapillary bumper stickers which you can order from the store (worldwide delivery)
Yes. I’d prefer to be able to easily remove it when I’m not mapping. Am I allowed to get the Mapillary logo printed on a magnetic sign at a local signage shop?
On that note, I have been a long time proponent of having a Mapillary with words like “SURVEYOR”, “LAND SURVEY”, “VERMESSUNG”, or “GÉOMÈTRE” (depending on language and/or country) etc. written on it. We already had this discussion. Such vests can help protect Mapillary contributors from traffic accidents as well as from unfriendly encounters. Furthermore, it signals to curious or overly anxious people no danger (from wo/man with long stick in their hand) from afar.