I have a question for some of the more experienced contributors.
I am aware of other similar posts about this topic (e.g : Long or short sequences, which is better? ) but they seem pretty old and even though they mention that any sequence length is okay, I’ve noticed that my larger sequences often take a very long time to be processed. Some of them contain 10,000–20,000+ images, and I still have uploads from weeks (or even months) ago that remain partially processed and stuck in the yellow state.
Looking at the profiles of some of the most active contributors (@thewizard, @allen, @ThailandandCambodia, @doriel, and many others), I noticed that their uploads seem to consist of much shorter sequences—typically a few hundred images, and rarely more than about 1,000 images per sequence.
This made me wonder whether this is intentional and considered good practice.
As I’m becoming more active on Mapillary, should I start splitting my longer recordings into smaller sequences before uploading? Does this improve processing time, reliability, or the chances that a sequence gets fully processed? Or does it make no real difference and my observations are just a coincidence?
I’d be very interested to hear how other heavy contributors organize their uploads and whether you’ve found an optimal sequence size.
I’m using mapillary_tools which cuts off the image sequences at 1000. I’m uploading primarily video (dashcam via Mapillary desktop uploader, the rest via CLI mapillary_tools). Video sequences are always cut off, images not, that’s the difference.
There should not be any issue witth biggger sequences, it will take a bit longer due to the bigger size.
I don’t see any reliability difference, but don’t ever go past the default 1000 sequence size. In most cases though the quantity breaks to a smaller number for other reasons. (like duplicate/cutoff parameters and video length) I do see a slightly longer delay processing videos on the server vs exif populated images, but I’d expect that.
The official policy is “the more the merrier” and “all image count limits per sequence have been removed”, which is good and makes sense. However, the truth is that until a few years ago there used to be a more or less hard limit of 1000 images per sequence, mainly due to RAM constraints on the backend. Of course, RAM constraints continue to exist but sequences are handled differently by the backend now than before. Unfortunately, the solution is not perfect yet and thus you can observe yellow sequences stuck in processing. In fact, sequences can get stuck in yellow processing with any number of images. So, it is not a matter of image count per sequence at all and there is no optimal sequence size either. To this very day, contributors have no idea why sequences get stuck or how to effectively work around this issue because nobody from Mapillary deemed it important enough to explain it. So far, the only mitigation for contributors around this issue can be deleting and uploading a stuck sequence again. But, it is just a gamble.
Generally speaking, stuck sequences are not lost or anything like that. They are just processed partially and become an annoyance. You are not doing anything wrong if your sequences happen to get stuck in processing. Occasionally, Mapillary initiates a global reprocessing run which may actually finalize stuck sequences.
So far, the only mitigation for contributors around this issue can be deleting and uploading a stuck sequence again. But, it is just a gamble.
