Migrating from MeshLab / PyMeshLab

MeshLab's editing filters map onto triwarp.remesh, triwarp.smoothing, triwarp.holes, and triwarp.repair. The structural difference: a PyMeshLab MeshSet is a mutable, stateful container where each filter mutates current_mesh() in place; triwarp filters take (vertices, faces) and return a new pair, so nothing needs a fresh MeshSet per call.

import pymeshlab as ml    # before
import triwarp as tw      # after

Remeshing and decimation

PyMeshLab	triwarp
`meshing_isotropic_explicit_remeshing(targetlen=...)`	`remesh.isotropic_remesh(target_length=...)`
`meshing_decimation_quadric_edge_collapse(targetfacenum=...)`	`remesh.quadric_decimate`
`meshing_decimation_clustering(threshold=...)`	`remesh.cluster_decimate`
`meshing_surface_subdivision_midpoint`	`remesh.subdivide`
`meshing_surface_subdivision_loop`	`remesh.subdivide_loop`
`meshing_repair_non_manifold_edges`	`repair.split_non_manifold_vertices` / `remove_non_manifold_faces`

Hole filling and cleanup

PyMeshLab	triwarp
`meshing_close_holes(maxholesize=...)`	`holes.fill_small(max_edges=...)` (a boundary-edge-count threshold, matching pymeshfix — see the function's own docstring for the exact off-by-one convention against MeshLab's `maxholesize`), or `holes.fill_min_weight` for every boundary regardless of size
`meshing_remove_duplicate_faces`	`repair.resolve_duplicated_faces`
`meshing_remove_unreferenced_vertices`	`repair.remove_unreferenced_vertices`
`meshing_remove_connected_component_by_diameter` / `_by_face_number`	`repair.remove_small_components`
`meshing_snap_mismatched_borders`	`holes.stitch` / `holes.stitch_min_weight`
No direct equivalent (a `PyTMesh` / pymeshfix operation)	`repair.collapse_small_triangles`, `repair.fix_self_intersections`

Smoothing and curvature

PyMeshLab	triwarp
`apply_coord_laplacian_smoothing`	`smoothing.filter_laplacian`
`apply_coord_taubin_smoothing`	`smoothing.filter_taubin`
`apply_coord_hc_laplacian_smoothing`	`smoothing.filter_humphrey`
`compute_curvature_principal_directions_per_vertex`	`curvature.principal_curvature`
`compute_scalar_by_shape_diameter_function_per_vertex`	`visibility.shape_diameter`
`compute_scalar_ambient_occlusion`	`visibility.ambient_occlusion`

Reconstruction and resampling

PyMeshLab	triwarp
`generate_surface_reconstruction_screened_poisson`	`reconstruction.screened_poisson`
`generate_surface_reconstruction_ball_pivoting`	`reconstruction.ball_pivoting`
`generate_resampled_uniform_mesh`	`reconstruction.resample_uniform`

What's different, not just renamed

No MeshSet, no current_mesh(). Every triwarp filter is a pure function over (vertices, faces); there's no session object to build, select a layer in, or read a result back off of.
Length parameters are always the same unit float the rest of the API uses, never a PercentageValue/PureValue wrapper — pass an absolute length (or a fraction you compute yourself against the mesh's own bounding-box diagonal, via bounds.aabb / Trimesh.extents, if that's what a MeshLab default was doing under the hood).
meshing_close_holes's size threshold is a perimeter (an edge-length sum); triwarp's holes.fill_small follows pymeshfix's convention instead (a boundary edge count) — see that function's docstring for why, and for the exact conversion when porting a MeshLab-tuned threshold.