Adam Heisserer

Computational design, data, and architecture.

Drawing with Photogrammetry - 2026 Update

This is an update to a photogrammetry drawing process I've had for a few years.

Summary: I take photos of a scene, then use Colmap and OpenMVS to build a dense point cloud. I use a python viewer to style the line art, altering settings for points connect with their nearest neighbors. This replaces my previous workflow where the whole cloud went into Rhino/Grasshopper for styling. Raw line art goes into Photoshop for blending and coloring adjustments. Full workflow is below.

Digital drawing of a garden’s textured foliage using photogrammetry point cloud data, stylized with glowing line art.
Digital drawing of a sycamore tree and house using photogrammetry point cloud data.
Digital drawing of a classical temple facade using photogrammetry point cloud data, stylized with wireframe overlays.
Digital drawing of a building and tree using photogrammetry point cloud data, stylized with dense line art.
Digital drawing of a tall building with "Frost" signage, rendered with photogrammetry point cloud data and wireframe overlays.
Digital drawing of a multi-story building using photogrammetry point cloud data, stylized with dense, textured line art.
Digital drawing of a sycamore tree and adjacent building using photogrammetry point cloud data, stylized with glowing line art.
Digital drawing of a classical building with columns, rendered using photogrammetry point cloud data and stylized line art.
Digital drawing of two weathered houses using photogrammetry point cloud data, stylized with textured line art.
Digital drawing of a weathered building facade using photogrammetry point cloud data, stylized with fragmented line art.
Digital drawing of a yellow car using photogrammetry point cloud data, stylized with fragmented line art.
Digital drawing of a fragmented building facade using photogrammetry point cloud data, stylized with textured line art.

Full Workflow

  • Take about 50-300 photos of a scene, or take a video and split into frames. Avoid blurring the photos, look for dynamic lighting conditions (sun and shadow).
  • Download COLMAP 4.x, use the cuda version if you have a GPU. I'm using a command line interface in VSCode. COLMAP finds the camera positions.
  • OpenMVS builds the dense point cloud.
  • The output is a text file of 3D points saved as XYZ,RGB values.
  • I'm using Pyside in Python to make a desktop window for editing. Pyvsita draws the points and line connections in the window editor.
  • I export a few hi-res renderings (5,000px wide) from the editor, and combine them in Photoshop. This is typically the same view with different parameters for how the points are connected. I blend them with different opacities, and sometimes filter out all but a few color families.
  • Each point identifies n number of of their nearest neighbors, within a certain minimum and maximum distance. The connecting lines are the same color as one of their endpoints. For most scenes, I target about 200,000 points.