IC: Slow down

[martha-thomae]

Miguel (Research Assistant in Musicology for the ECHOES Project) has noticed a slow down in the IC when clicking on grouping and splitting, and when clicking on a neume it also took longer to get the right place in the page and highlight it. As this was not the case when we started from scratch, he thinks that his might be due to the number of glyphs we have on the training data (around 7,000 glyphs). But I don't know why the training data would affect the speed of grouping and splitting actions. It is true that when splitting glyphs, these moved from a particular class to UNCLASSIFIED; and I think that, when grouping glyphs, they should share a class first. But, I really don't think that the number of examples in training data should affect the grouping and splitting actions.

I don't know if this issue is a slowdown related to an "IC issue" or a "Rodan issue."

@JoyfulGen, @kyrieb-ekat, and @annamorphism have you noticed this behaviour before?

I am also tagging @fujinaga here to keep him up to date and @elsinhadl.

[JoyfulGen]

@martha-thomae, the IC has certainly slowed down before, but I don't remember that being related to training data size. I just gave it a try and I think I got the same result you describe: when I gave the IC no training data, grouping and splitting was very fast, and when I gave it a big training data set grouping and splitting was slower, but not by much. However, my biggest data set is for Einsie and it only contains 2400 glyphs, which is far from 7000. Out of curiosity, do you know why their training data contains so many glyphs? That seems like a lot!

I'm sorry though, other than confirming the issue I don't have any more insight. @homework36 might know more?

[kyrieb-ekat]

Apologies for the delayed response, I'm back now from visiting family. 😁

I haven't encountered this specific issue before, but the IC definitely can be a little slow. I'll echo Gen here and say I haven't seen this as related to training data size (usually it's a general Rodan issue, or if the image is very big and has a looot of detected glyphs on the page).

Would you mind attaching or emailing the training file, image, and other inputs for the IC job, so I can test it a bit?

I'm also curious about the 7,000 glyph training file (wow!!); have y'all tried producing a file with it as it is, or with earlier versions of training files with fewer glyphs? Does Miguel happen to recall around when the slowdown became detectable (glyph size wise) or day? Ah, Rodan!