Now I’m uploading thousands 360 panoramas of scenic outdoors locations and got stuck in a little problem. I cannot upload my latest panoramas taken with drone DJI Mavic 3 as their resolution is 103 megapixels, but Mapillary has the limit of 100 megapixels.
This means I need to shrink images, but doing so will drastically reduce the details of distant objects which is very important thing in aerial panoramas, when travelers or hunters are able to zoom it in and carefully examine distant terrain to plan their offroad navigation in wilderness areas.
I’d suggest Mapillary developers either to increase image resolution limit to suit modern cameras, or, as it was implemented in google maps, to set file size limit instead of megapixels limit (googlemaps accept any image up to 75mb file size without resolution limit).
Thanks for reporting. Could you share a Google Drive link to an original file? Looking online it seems to suggest that the resolution of a DJI Mavic 3 should be 20MP?
@coronaviking - can you give it a go now, we just released an update for this so that it should be working. One note though is that Mapillary is predominantly intended for street level imagery, and less so for aerial panoramas.
I have been begging for this for a long time now. Although I am happy that the pixel limit has been raised, it is not exactly what we might have been expecting. 108,000,000 MP continues to give really wired unnatural width and height limits for 2:1 spherical projection images of 14,696×7,348. In my understanding natural limits should be based on powers of 2, like 2^27 = 134,217,728 pixels (basically 128 MP). @boris Why is there a pixel limit in the first place since you scale down all images for reconstruction (and presumably also segmentation) anyway? Are “image bombs” really a problem? Why not go with JPEG’s design limit of 65,535?
I have a few 22,010×11,005 = 242,220,050 images sitting on the shelf for upload but a handful of trial runs of scaling down to those wired widths and heights have revealed some ugly looking artefacts. Sure, one could go down to the next lower power of 2 below the limit but then the purpose of all these details in the images are rendered mute.
@coronaviking - thank you for the confirmation that it is now working for you!
@GITNE - you have a good point - there is a limit because of processing which is actually done at full resolution for accuracy. We can think about scaling this down to support higher resolutions going forward, but for now we’re not aware of cameras that are outputting higher than 108 MP, so this is lower down in the priority list.
we’re not aware of cameras that are outputting higher than 108 MP
Panoramic images may be created from many individual photos combined into a single image using a stitching software (such as PTgui or Hugin). Resolution of such panoramas is effectively unlimited (except the bounds set by file formats themselves).
Such panoramas, obviously, require much more manual work for capturing and stitching than simply taking 360° videos/photos using built-in functionality of dedicated cameras, and uploading them. Don’t get me wrong – I’m not trying to diminish the value of that quickly-captured imagery: for the vast majority of the places covered, it is great, and it wouldn’t be possible to have that coverage without the speed and ease-of-use of these dedicated cameras. However, the mere fact that the manually-created panoramas require much more work means that they usually depict something special enough to justify the effort spent – and this is their value.
I believe the storage and processing power required for these manually-created, potentially-very-high-resolution panoramas would be negligible compared to that needed for the sheer amount of “regular” images, captured at every few meters of millions of streets. So, what about trying to find some way to relax the resolution limitations and allow them? To protect from abuse, other kinds of limits may be set – for example, Mapillary may limit how many images with “large” resolution (e.g. >108 Mpx) each user can upload in a given time frame.
Thanks @DmitTrix - that makes sense. For now I think I would recommend for folks who are manually creating panoramas to downscale to 108 MP. I know this probably isn’t the answer you’re looking for, but right now I would say 108 PM is a very high quality image which should express the necessary detail even for “special” locations. Increasing this limit further is lower down on our priority list (also because Mapillary is really optimized for captures across distance (“regular” images as you referred to them), and less so for single panorama images)