Closed hjwdzh closed 4 years ago
If we are lucky enough to beat existing methods with current brute-force optimization, we can go further to explore deep deformation functions for better convergence.
Seem to be solved. Will test for all benchmark shapes.
I think it is worth devoting an ablation study section to compare the various normalizing flow bijector methods: (NeuralODE, MAF, RealNVM, etc...)
If we are lucky enough to beat existing methods with current brute-force optimization, we can go further to explore deep deformation functions for better convergence.