- Label cost: L1/L2 norm of parameter vector?
- Degree 4 polynomial surfaces: is there a reason why even degree polynomials are better? Is our failure to fit some points with degree 3 a lack of degrees of freedom or a failure in fitting?
- Finding the boundary for open surfaces: graph cut with scribbles
- Link all scribbled “in” nodes to the source node, same for “out” nodes
- Better final result visualization
- Project strokes on result
- Heatmap of stroke/mesh distances
- Final result sharp features misalign with strokes
- This is pretty bad in some cases (eg shoe) –> quantify deviation
- Can we be better at fitting the surfaces?
- Re-fitting the surface models with weights (higher weights at strokes corresponding to seams)
- Artists compensation for their sketches
- Find some kind of official platform, maybe Gumroad?
- Upwork: we could also pay them to do some new sketches
- Best outcome would be to have the right to distribute the sketches
- Ask Seok-Hyung Bae for their sketch data
Meeting
Notes