The progress bar is currently turned off during training because the total number of queue pops is unknown (since a pop can be valid but not result in a merge), and exceeding 100% of merges results in a crash.
Possible options:
find an upper bound for the total number of examined edges and use that as the total. Downside is that this might be much higher than necessary, resulting in over-inflated ETAs and never reaching 100%.
find an average case for the total number of examined cases, and modify the progress bar behaviour to prevent crashes upon reaching 100%.
find a way to update the total during execution.
find a way to update the current number of merges during execution.
Note: the total number of actual merges during learning is equal to the number of superpixels minus 1 minus the number of segments in the ground truth.
The progress bar is currently turned off during training because the total number of queue pops is unknown (since a pop can be valid but not result in a merge), and exceeding 100% of merges results in a crash.
Possible options: