v0.2.0

• Added new models : LitePT without dist_axes and even without Z0.

v0.1.0

Changes

Models

Release with two models. One based on LitePT and another on PTv3. Both are trained on SegmentForest dataset at 0.05m voxel size with full 3DFin features (dist_axes, z0).PTv3 model remains the default, but LitePT is recommended.

Software

• Add logging (on CLI)
• Add QT GUI as standalone and as a CloudCompare plugin
• Improved integration of builtin torch FlashAttn: It should improve default performance out of the box across all OSes, drop vanilla FlashAttn support.
• Add a device selection ("cuda" or "cpu") mechanism (both in CLI and GUI) when the package is build with CUDA.

v0.0.1

Weight trained on Segmented forest at 0.05m voxel size