Hi again,
Makes sense, good point. I’ll see what we can do. In a short term let me know if you notice this failing for smaller uploads too.
Kind regards,
Balys
After breaking up the two devoured sequences and uploading them again two days ago, I can see that some of the ca. 15k sub‑sequences have already finished processing. Which is great! 
The other sequence broke naturally into larger ca. 24k image sub‑sequences. So, I am not yet sure about their prospects for finishing. Anyhow, I also seem to sense some significant speedup 
in processing lately (maybe due to the hardware upgrade?). 
Feels great!
All previously failed sequences have returned from the dead and are visible in the feed. They continue processing so far, however I am still not quite sure if they are going to finish. Though, the previously undead sequences have begun to feature point clouds and triangulated traffic signs, which is a good sign. I can finally see progress! Thank you @balys!
2 Likes
Just FYI @nikola, undead sequences are indeed now displayed in the contributor’s feed, however they continue to be absent from the organization’s feed.
There is another disparity between the feeds. The organization feed displays ingestion messages, like Processing 1 capture from Aug 4, at the top of the feed, even when logged out. However, logged out users do not see ingestion messages at the top of contributor feeds.
2 Likes
@balys I have another sequence stuck in processing. This time just 287 images and despite atomic uploads! 
Sure, I can delete it and upload again but I do not think this is how it should work.
Stuck Sequences
I am still fighting with yellow sequences stuck in “Map Data Processing”. Unfortunately, I have not figured out any true cause or effective workaround for this issue. What I have observed though is the fact that it is almost certain for a sequence to get stuck in processing when it has an optical or GPS track loop. Limiting sequence image count does not help evidently either. So far, the only mitigation against this issue have been repeated uploads. However, this strategy is cumbersome, time consuming, and a gamble. Not to mention that it is costly for both contributors and Mapillary. I really wish and hope this issue to finally get resolved for good.
Dynamic Reconstruction Bundle Sizes
@paulinus Although I do not know whether these subjects are connected, I have observed that all sequences are split into exactly 75 (or was it 50?) image count bundles for reconstruction. Have you thought about per sequence dynamic reconstruction bundle sizes based on image resolution and average GPS spacing (initial baseline)? Or, are dynamic bundle sizes infeasible at Mapillary’s scale of reconstruction (or maybe I am totally confused)? Thank you.
Why Arbitrary Sequence Grouping is Important
Some contributors want to manage a more or less curated dataset. The current feed, filters, and bbox based queries work just okay for now. Contributors want to find, access, and organize their imagery efficiently. As the platform and user contributions grow, the needs for clear data structures and organizing tools grow too. See also A timeline approach for Mapillary road imagery.