Home

Awesome

pydelatin

A Python wrapper of hmm (of which Delatin is a port) for fast terrain mesh generation.

A screenshot of Glacier National Park taken from the demo. The mesh is created using pydelatin, encoded using quantized-mesh-encoder, served on-demand using dem-tiler, and rendered with deck.gl.

Install

With pip:

pip install pydelatin

or with Conda:

conda install -c conda-forge pydelatin

On Windows, installing via Conda is strongly recommended.

If installing with pip on Windows, glm is a prerequisite for building from source. Open an issue if you'd like to help package binary wheels for Windows.

Using

Example

from pydelatin import Delatin

tin = Delatin(terrain, width, height)
# Mesh vertices
tin.vertices
# Mesh triangles
tin.triangles

API

The API is similar to that of hmm.

Additionally I include a helper function: decode_ele, to decode a Mapbox Terrain RGB or Terrarium PNG array to elevations.

Delatin

Arguments
Attributes

util.rescale_positions

A helper function to rescale the vertices output to a new bounding box. Returns an ndarray of shape (-1, 3) with positions rescaled. Each row represents a single 3D point.

Arguments

Saving to mesh formats

Quantized Mesh

A common mesh format for the web is the Quantized Mesh format, which is supported in Cesium and deck.gl (via loaders.gl). You can use quantized-mesh-encoder to save in this format:

import quantized_mesh_encoder
from pydelatin import Delatin
from pydelatin.util import rescale_positions

tin = Delatin(terrain, max_error=30)
vertices, triangles = tin.vertices, tin.triangles

# Rescale vertices linearly from pixel units to world coordinates
rescaled_vertices = rescale_positions(vertices, bounds)

with open('output.terrain', 'wb') as f:
    quantized_mesh_encoder.encode(f, rescaled_vertices, triangles)

Meshio

Alternatively, you can save to a variety of mesh formats using meshio:

from pydelatin import Delatin
import meshio

tin = Delatin(terrain, max_error=30)
vertices, triangles = tin.vertices, tin.triangles

cells = [("triangle", triangles)]
mesh = meshio.Mesh(vertices, cells)
# Example output format
# Refer to meshio documentation
mesh.write('foo.vtk')

Martini or Delatin?

Two popular algorithms for terrain mesh generation are the "Martini" algorithm, found in the JavaScript martini library and the Python pymartini library, and the "Delatin" algorithm, found in the C++ hmm library, this Python pydelatin library, and the JavaScript delatin library.

Which to use?

For most purposes, use pydelatin over pymartini. A good breakdown from a Martini issue:

Martini:

Delatin:

Benchmark

The following uses the same dataset as the pymartini benchmarks, a 512x512 pixel heightmap of Mt. Fuji.

For the 30-meter mesh, pydelatin is 25% slower than pymartini, but the mesh is much more efficient: it has 40% fewer vertices and triangles.

pydelatin is 4-5x faster than the JavaScript delatin package.

Python

git clone https://github.com/kylebarron/pydelatin
cd pydelatin
pip install '.[test]'
python bench.py
mesh (max_error=30m): 27.322ms
vertices: 5668, triangles: 11140

mesh (max_error=1m): 282.946ms
mesh (max_error=2m): 215.839ms
mesh (max_error=3m): 163.424ms
mesh (max_error=4m): 127.203ms
mesh (max_error=5m): 106.596ms
mesh (max_error=6m): 91.868ms
mesh (max_error=7m): 82.572ms
mesh (max_error=8m): 74.335ms
mesh (max_error=9m): 65.893ms
mesh (max_error=10m): 60.999ms
mesh (max_error=11m): 55.213ms
mesh (max_error=12m): 54.475ms
mesh (max_error=13m): 48.662ms
mesh (max_error=14m): 47.029ms
mesh (max_error=15m): 44.517ms
mesh (max_error=16m): 42.059ms
mesh (max_error=17m): 39.699ms
mesh (max_error=18m): 37.657ms
mesh (max_error=19m): 36.333ms
mesh (max_error=20m): 34.131ms

JS (Node)

This benchmarks against the delatin JavaScript module.

git clone https://github.com/kylebarron/pydelatin
cd test/bench_js/
yarn
wget https://raw.githubusercontent.com/mapbox/delatin/master/index.js
node -r esm bench.js
mesh (max_error=30m): 143.038ms
vertices: 5668
triangles: 11140

mesh (max_error=0m): 1169.226ms
mesh (max_error=1m): 917.290ms
mesh (max_error=2m): 629.776ms
mesh (max_error=3m): 476.958ms
mesh (max_error=4m): 352.907ms
mesh (max_error=5m): 290.946ms
mesh (max_error=6m): 240.556ms
mesh (max_error=7m): 234.181ms
mesh (max_error=8m): 188.273ms
mesh (max_error=9m): 162.743ms
mesh (max_error=10m): 145.734ms
mesh (max_error=11m): 130.119ms
mesh (max_error=12m): 119.865ms
mesh (max_error=13m): 114.645ms
mesh (max_error=14m): 101.390ms
mesh (max_error=15m): 100.065ms
mesh (max_error=16m): 96.247ms
mesh (max_error=17m): 89.508ms
mesh (max_error=18m): 85.754ms
mesh (max_error=19m): 79.838ms
mesh (max_error=20m): 75.607ms

License

This package wraps @fogleman's hmm, a C++ library that is also MIT-licensed.