Post-deadline summary: Vectorization project

Results

visualise transition function weights
visualise labeling + streamlines
do lots of (automated) tests with different parameters, inputs
export pdf+image
extraction: assignmen to curve segments
improve removal by a LOT
improve the fitting. Global formulation? (splines?)
re-assign the pixels when a curve is removed
improve treatment of Y junctions
figure out how to enable Y branching in the parametrization (Instant Meshes? PGP?)
Bezier merging, otherwise way too many curves. Use the information from the assignment to decide which curves should be merged?
don’t init the sparse triangulation via dense triangulation, use blue noise sampling (BNOT)
clean up the GUIs, clean up the code (simplify!)
optimize the code; merge stuff from parametrization and extraction
more stress tests (sharp angles, sequence of sharp angles, e.g. puppy hair)
comparison to vectorization papers and StrokeAggregator
fix all parameters to reasonable default values
use stroke width wherever possible
paralellize where possible (streamline tracing?)
paper: didactic figures (transition functions etc.)
extraction: insets
extraction: compute grid vertices in all triangles, not just the ones with a pair of isolines – take all nodes in some radius
extraction: when grouping grid vertices, make sure the radius is smaller than feature size! (this should be done already, verify)

we could view the parametrization (+extraction) as assignment (once again) of pixels to param. curves. Then use the param. + position to fit a global network of splines.