search

GFZ / arosics

AROSICS - Automated and Robust Open-Source Image Co-Registration Software
https://git.gfz-potsdam.de/danschef/arosics
Apache License 2.0
137 stars 26 forks source link

reproducibility of processing with RANSAC #18

Closed nth29 closed 1 year ago

nth29 commented 1 year ago

Description

When using Arosics with RANSAC GCP outliers filtering, which is a not deterministic process, for similar input parameters, the final set of GCPs found by COREG_LOCAL is not always the same.

Because tuning Arosics parameters is an iterative process, we are working on a reduced 3 band dataset of an hyperspectral datacubes for this "tunning" step and, once we will have found the good parameters, we would like to apply exactly the same shift to the full hyperspectral cube.

For that, due to RANSAC, keeping the same input parameters is not enough, thus, is there a way to save COREG_LOCAL GCPs results and parameters, and apply them afterward using DESHIFTER for example ?

danschef commented 1 year ago

arosics version 1.9.0 now uses a fixed random_state by default when calling RANSAC (implemented here). This ensures reproducable results from the RANSAC outlier filtering and should therefore fix this issue.