A Neural Net First Break Picker From XtremeGeo. Inhuman Results, Inhuman Speed.
Features:
- Easy to train new models using either Flatirons or Phoenix Data
- Use models provided by XtremeGeo or create proprietary custom models
- May be used to pick the following type of data:
- Flatirons project data- no need to import into Phoenix
- Phoenix project data
- SEG-Y data
- More options on the way
- It’s fast, even on low end GPU cards
- Very high-quality picks. May be as good or better than human picks on difficult data
Technical:
- Built using LibTorch, the C++ foundation of industry-leading PyTorch
- GPU is required for the current release, future releases will not require this
Analytical comparison results

Errors are in milliseconds
- PickX picks were not modified in any way
- PickX picks are made using the Amplitude Onset approach
- Comparison made using “Average Pick Error” of delay time and VNS tomography solutions
- User picks were given an advantage- they were usually refined using the delay time stack approach
- “User” picks are in general a combination of auto picks, manual picks, etc. and were made in Flatirons