Pandora

Fixed

- Correct refinement bug when using subpixel. [223]
- Corrections to bug after new xarray version. [224]

Added

- Memory consumption estimation. [222]
- Add extra requirements making mccnn plugin available.

Added

- Binder for Pandora's notebooks. [215]
- Version handling with setuptools_scm. [212]
- Set dataset's transform to None if it is the identity matrix. [211]
- Handling of np.inf values on input images. [210]
- Remove copy from check_dataset and doesn't return dataset anymore. [219]
- Doc for confidence and piecewise optimization.
- Do not force omp layer on numba and silence its warning.

Fixed

- Correction on multi scale processing to allocate the confidence measure. [217]
- Correct non-masked nan values on validity mask. [218]
- Margin computation without optimisation step. [220]

Added

- Implementation of the ambiguity confidence measure. [162]
- Implementation of the bilateral filter to suppress OpenCV dependency. [148]
- Checker to check if an image dataset is valid. [196]
- Creation of Jupyter notebooks for Pandora.[153]

Fixed

- Correct Nan propagation of the bilateral filter. [159]

BREAKING CHANGE

- GeoRef is now supported and both `crs` & `transform` ought to be set as attributes of images dataset provided to Pandora [158]

Fixed

- Call to interpolation functions. [175]
- Check right disparities. [177]
- Disparity refinement after disparity map filtering step. [154]

Added

- Implementation of multi scale processing for the disparity map computation. [80]
- A `./data_sample/` directory with images and configuration files for user's first steps with Pandora [168]
- Add extra requirements making sgm plugin available [183]

Changed

- Semantic change: stereo becomes matching_cost (transition name, module, files and class name). [170]
- Merge image on input configuration section. [169]
- Enable the use of GraphMachine if graphviz avalaible to generate machine states diagram. [149]
- Move find_valid_neighbors function to img_tools. [184]
- Renamed json_cheker to check_json. [182]

Fixed

- Confidence measure that computes the distance LR / RL in cross checking method. [155]

Added

- Classification and segmentation maps as input. [126]

Changed

- Non truncation of the cost volume and the disparity map, which have the input image's size. [152]
- Input masks convention and creation of masks dataset.