masadcv / FastGeodis

Fast Implementation of Generalised Geodesic Distance Transform for CPU (OpenMP) and GPU (CUDA)
https://fastgeodis.readthedocs.io
BSD 3-Clause "New" or "Revised" License
90 stars 14 forks source link

[FEATURE] Provide optimised version of signed geodesic and euclidean distance transform #27

Closed masadcv closed 2 years ago

masadcv commented 2 years ago

Is your feature request related to a problem? Please describe. Currently the library support signed distance transforms by taking two distance transforms and subtracting them. This could be optimised further for example as discussed here https://github.com/masadcv/FastGeodis/issues/9

Describe the solution you'd like Implement signed distance transforms with better optimisation

Describe alternatives you've considered Using the current signed distance transform methods provided in FastGeodis. These work but have room for further optimisation

Additional context This could improve interactive segmentation methods execution time, especially implementation for method called Geos from:

Criminisi, Antonio, Toby Sharp, and Andrew Blake. "Geos: Geodesic image segmentation." European Conference on Computer Vision. Springer, Berlin, Heidelberg, 2008.