romanjuranek / librectify Goto Github PK

postprocess_line_segments crashes

Currently user must call set_num_threads to update internal state for parallelism which is for some reason unacceptable.

Instead it goes at full speed

Add at least win32/win64/linux 64 building. If anyone knows how to do that, please tell me

So far everything is documented in public header file. Add doc directory with documentation of API and internal stuff (how it works inside).

compute_rectification_transform returns incorrect coordinates of vps that does not correspond to the transformation

rng in ransac.cpp is a global object so all threads use the same one

Interface the library to python (ctypes/cffi/cython)

Coords passed to compute_transform must be relative to image center not in image space

So far, RANSAC is used. It has fixed number of iterators - i.e. model evaluations on full set of lines which is slow. It can be replaced with PROSAC which is more universal and faster. And additionally, the result would be more stable (so no need of deterministic random generator)
Prosac needs:

• A weighting mechanism for selection of single model. Most likely CHT-based.
• PROSAC sampler

On some images, every run results in different solution which may confuse end users.