-
Notifications
You must be signed in to change notification settings - Fork 13
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Help needed: layers sometimes completely empty #1207
Comments
Ok! I have a more concrete example. I started by making layer models with this score: I categorized my layers the usual way:
Those models produce almost impeccable layers 1 and 2, but they put all of the words in layer 1 as well, which means that layer 3 is completely empty. You can see this on this Pixel screenshot. The clefs and notes are blue and the staff lines are red, which is correct. However, the words should be yellow, but they're blue, and there's no yellow anywhere: I then iterated those models, to try to improve the results. The new models I ended up with do indeed put text in layer 3, but they also do much worse on both staff lines and notes. You can see that there's a lot of blue on the staff lines, and quite a bit on the words as well, and there's a lot of yellow in the notes: @fujinaga, is this expected behaviour? |
I was able to replicate this issue, where the third layer, which ought to have been entirely text, appeared empty. However on closer examination it appears that only a pixel or two makes it through. See attached images (layers 3 are the problem layers)- there are a couple small specks present so they're not wholly empty. The groupings of the images below are the layers 0 (background), layers 1, layers 2, and layers 3. Interestingly the very edge of the image being included has been an observed behavior before, where the edges eventually never get pulled through for any interactive layer (i.e., layers 1-3 and always remain in L0/background). |
Amazing! @fujinaga we have results that can be reproduced. The following layer separation models produce an (almost) empty layer 3 every time: Paco Test 13 Background Model.hdf5.zip This is the workflow: Fast.json |
There are two things that happen:
Like so:
Like so (you can see that the staff lines and text were extracted, but not the glyphs at all):
This phenomenon has happened at the beginning of iteration or later down the line, with cropped images or full images, with iteration or without. I'm making this issue because I'm not sure where to go from here and I want to make sure to have a record of it.
These models produce all empty layers (so no layer separation at all):
Background Model MTWS 023 round 3.hdf5.zip
Model 1 MTWS 023 round 3.hdf5.zip
Model 2 MTWS 023 round 3.hdf5.zip
Model 3 MTWS 023 round 3.hdf5.zip
These models produce an empty layer 1:
Beatrice 6+7 Background Model.hdf5.zip
Beatrice 6+7 Model 1.hdf5.zip
Beatrice 6+7 Model 2.hdf5.zip
Beatrice 6+7 Model 3.hdf5.zip
And these models produce empty layers 1 and 3 (so only layer 2 gets separated):
Beatrice 7 part 2 Background Model.hdf5.zip
Beatrice 7 part 2 Model 1.hdf5.zip
Beatrice 7 part 2 Model 2.hdf5.zip
Beatrice 7 part 2 Model 3.hdf5.zip
The text was updated successfully, but these errors were encountered: