Hello everyone,
I’m wondering: Is anything other than 360° photography still worthwhile for Mapillary these days? Image processing is lagging behind again—many uploads sit in the queue for long. If you look at the current top lists, they’re often dominated by one-dimensional images of the same street over and over. Meanwhile, 360° cameras have become affordable and capture far more detail than conventional cameras, even when mounted inside a car. A single pass covers the entire surroundings, including side streets—this saves time and significantly boosts coverage per kilometer.
What do you think?
Are you still planning to use classic cameras, or have you fully switched to 360°?
Are there scenarios where regular photos still make sense (e.g., pedestrian perspectives, narrow alleys)?
How do you deal with processing wait times—do you prioritize certain areas or just wait it out?
Should 360-degree photos have their own category in Mapillary’s ranking?
I agree that 360 images are the most valuable for the reasons you mentioned - however not everyone has access to a 360 camera and classic images are definitely preferable to no images
For processing wait times, this is something we’re actively tracking and improving, and it shouldn’t ideally be a consideration for the community. We need to make sure every image is processed quickly, regardless of where and what type.
In terms of ranking 360 images differently, yes, this is a good point, let me think about this for our backlog.
Certainly, 360° imagery does not solve everything, neither does planar imagery.
It is like with everything else; low hanging fruit is easy to pick. If you really want to make a positive impact then you always have to make some minimal investment and leave your comfort zone. Go off familiar roads. That’s life.
Pure numbers never tell the full story. Numbers can be indicators at best but they always need context in order to paint the picture (no pun intended).
Oh, I beg to differ. Although it is true that 360° cameras have become more affordable and better, they are still far from being equivalent to traditional cameras in terms of image quality. And yes, the MAX2 is objectively better than the MAX. Same goes for Insta360 iterations. However, you have to keep in mind that resolution requirements for 360° imagery increase cubed (not even just squared!) compared to planar imagery. Increases in resolution also lead to higher signal processing throughput requirements, which also increase exponentially. Energy demands and die area grow too. So, 360° cameras still have a long way to go. By the way, Google Maps has already gone through all this trouble over a decade ago, made the same experiences, and came to the same conclusions. Just have a look at the history of iterations of their custom built 360° hardware.
Besides, all extremely wide fixed aperture and fixed focus cameras, like 360°, action, and smartphone cameras have a limited effective scanning range of about 5 m to 10 m (16’ to 33’). Beyond that, you get really blurry edges which work against good reconstruction and computer vision. And, you cannot compensate for these blurry edges even with the most effective adaptive sharpening algorithms. Hence, traditional cameras will not become obsolete for mapping for a long time.
Let me put it like this; If I could swap the MAX for a HERO, I would do it instantly. Why? Because the MAX does neither do 360°, nor planar imagery well. The MAX has a great GPS receiver and extraordinary long battery life but that’s about it. Whereas a HERO has been basically made perfect for mapping by the form factor and price since the HERO8.
Generally always because of resolution and sharpness in the distance.
I have recently started to do atomic uploads for areas where I have captured multiple sequences. Though, I am not entirely sure that it really helps to avoid sequences getting stuck in processing. This strategy may also kind of grant or schedule additional time for reconstruction to complete in an area because every uploaded sequence apparently triggers a separate amount of reconstruction time, whereas a lump of sequences seems to share only one time slice. Though again, I do not know for sure. I am just testing and poking because the whole process is perhaps a bit (too) opaque (for me).
Maybe? I will not object. On the other hand, imho 360° imagery is not inherently better or worse than planar imagery. But, separate 360° figures would certainly require more UI clutter to work through in the web app. Hence, I am not sure it is worth the effort.
360 capture remains painful. If anything, it became more of a hassle with higher resolution sensors - you can’t just run your LG 360 off the mapillary app anymore.
At the same time, the level of detail in most 360 images remains low(er) especially when compared with decent non-360 cameras.
Hence, I am not sure it is worth the effort.