oslandia/geo3dfeatures

Migrated to: https://gitlab.com/Oslandia/geo3dfeatures

geo3dfeatures's Issues

Manage `NaN` values before clustering

In the current state of the code, NaN values make the program fail. We could improve the code robustness by explicitly managing this aspect. Two actions to take:

• message the user
• drop the points with NaN features

Improve the normalization scheme

For now, we normalize the features before running prediction models in a very simple way. Maybe we could improve that scheme by specifying the features to normalize (and the normalization function) in a configuration file.

Improve the post-processing performances

For now, the post-processing (when required) is run in a "brute-force" mode. We could certainly be smarter, for example by parallelizing the treatment (as we do for feature extraction).

Some benchmarking may be done from the PCL library: http://pointclouds.org/documentation/tutorials/resampling.php#moving-least-squares.

Introduce the weak supervision in the CLI

We have some raw material related to weak supervision (see the relevant notebook in the examples folder). We could do a significant step towards the method usage in industrial contexts by introducing it in the project CLI, as a specific program.

Generalize the label glossary

For now, we are dealing with hard-coded labels (see geo3dfeatures/tools/__init__.py). A better practice could be to pass the dataset glossary as an argument for each program (e.g. -l/--label-glossary).

We could have a configuration file for each dataset of interest, like:

• config/dataset_1.json
• ...
• config/dataset_n.json