Help us improve our ML workflow for fault predictions.
Currently, the best results are obtained when OpendTect’s pre-trained ML Fault Predictor (a 3D Unet) is applied to an input volume that is pre-processed with an Edge-Preserving Smoothing filter.
This works extremely well in settings with clean-cut faults that exhibit distinct fault throws. To accommodate situations, in which the user is not satisfied with these results, we would like to develop a solution that gives the user more control over the final outcome.
We envision a workflow in which the user interprets faults on a small number of lines. From these interpretations a new set of training examples is created, which are used either to improve the pre-trained Fault Predictor, or to train a new model from scratch.
We challenge the OpendTect Machine Learning Development Community (OMLDev) community to improve our current best results on the Delft data set Delft features popup structures associated with complex faults exhibiting strike-slip movements. More information can be found on the OMLDev Discord server.
To join our community and win this prize, please click the OMLDev Discord invite link