dGB Earth Sciences


As of version 6.6.8, OpendTect’s library of pre-trained models boasts 4 spectaculair seismic enhancement models developed by Lundin’s GeoLab R&D group:

  • Desmile: This model removes steeply dipping noise and smile artefacts
  • SimpleHmult: Attenuates horizontal multiples
  • AJAX: Enhances the visual quality and interpretability
  • SimpleDeNoise: Removes “salt & pepper / jitter” noise.

Sharing Trained Machine Learning Models is Redefining our Modus Operandi

The models are released in OpendTect Pro’s Machine Learning plugin under the “Creative Commons Attribution-ShareAlike License Agreement version 4.0” . This means that anyone with an OpendTect Pro ML license can apply these models to their own projects. All models, except for SimpleHmult can be applied out-of-the-box. Processing time is extremely fast, especially if run on a GPU. For optimal results, SimpleHmult may require flattening on the multiple generating event before applying the model. To facilitate the application of SimpleHmult, OpendTect’s flattening / unflattening workflows have been improved.

We encourage other R&D groups to follow Lundin’s lead. If you have any pre-trained model (seismic, wells, or wells-to-seismic) with potential for re-use on unseen datasets, please contact us. Sharing pre-trained models is the fastest way from research to operational deployment. You do not need to share with the entire world. If you want to share your models only inside your company then that is perfectly fine.

The slider shows an application of SimpleHmult, courtesy Lundin GeoLab.

The next post in this series of posts is available.