File IO#

This tutorial shows how basic data structures are read and written by Open3D.

Point cloud#

The code below reads and writes a point cloud.

[2]:

print("Testing IO for point cloud ...")
sample_pcd_data = o3d.data.PCDPointCloud()
pcd = o3d.io.read_point_cloud(sample_pcd_data.path)
print(pcd)
o3d.io.write_point_cloud("copy_of_fragment.pcd", pcd)

Testing IO for point cloud ...
PointCloud with 113662 points.

[2]:

True

By default, Open3D tries to infer the file type by the filename extension. The following point cloud file types are supported:

Format	Description
`xyz`	Each line contains `[x, y, z]`, where `x`, `y`, `z` are the 3D coordinates
`xyzn`	Each line contains `[x, y, z, nx, ny, nz]`, where `nx`, `ny`, `nz` are the normals
`xyzrgb`	Each line contains `[x, y, z, r, g, b]`, where `r`, `g`, `b` are in floats of range `[0, 1]`
`pts`	The first line is an integer representing the number of points. The subsequent lines follow one of these formats: `[x, y, z, i, r, g, b]`, `[x, y, z, r, g, b]`, `[x, y, z, i]` or `[x, y, z]`, where `x`, `y`, `z`, `i` are of type `double` and `r`, `g`, `b` are of type `uint8`
`ply`	See Polygon File Format, the ply file can contain both point cloud and mesh data
`pcd`	See Point Cloud Data

It’s also possible to specify the file type explicitly. In this case, the file extension will be ignored.

Mesh#

The code below reads and writes a mesh.

[3]:

print("Testing IO for meshes ...")
knot_data = o3d.data.KnotMesh()
mesh = o3d.io.read_triangle_mesh(knot_data.path)
print(mesh)
o3d.io.write_triangle_mesh("copy_of_knot.ply", mesh)

Testing IO for meshes ...
[Open3D INFO] Downloading https://github.com/isl-org/open3d_downloads/releases/download/20220201-data/KnotMesh.ply
[Open3D INFO] Downloaded to /root/open3d_data/download/KnotMesh/KnotMesh.ply
TriangleMesh with 1440 points and 2880 triangles.

[3]:

True

Compared to the point cloud data structure, a mesh has triangles that define the 3D surface.

By default, Open3D tries to infer the file type by the filename extension. The following mesh file types are supported:

Format	Description
`ply`	See Polygon File Format, the ply file can contain both point cloud and mesh data
`stl`	See StereoLithography
`obj`	See Object Files
`off`	See Object File Format
`gltf`/`glb`	See GL Transmission Format
`fbx`	Autodesk FBX (via Assimp)

Multi-material models (including USD usd/usda/usdc/usdz) should be loaded with read_triangle_model rather than read_triangle_mesh. USD import is experimental (meshes and PBR materials only; see section below).

Triangle mesh model (multi-material)#

For assets with multiple meshes and PBR materials (typical for glTF/GLB, FBX, and USD), use read_triangle_model. It returns a :class:open3d.visualization.rendering.TriangleMeshModel suitable for the Filament-based visualizer.

USD (.usd, .usda, .usdc, .usdz) import is experimental: only mesh geometry and PBR material read support is provided (base color, normal, metallic/roughness, occlusion, and related factors). There is no USD export, and scene features such as animation, lights, and cameras are not imported.

A complete example that downloads Apple’s sample biplane USDZ and visualizes it lives at examples/python/io/model_io.py:

import open3d as o3d
from pathlib import Path
import urllib.request

url = (
    "https://developer.apple.com/augmented-reality/quick-look/models/"
    "biplane/toy_biplane_realistic.usdz")
dest = Path("toy_biplane_realistic.usdz")
if not dest.is_file():
    urllib.request.urlretrieve(url, dest)

model = o3d.io.read_triangle_model(str(dest))
o3d.visualization.draw([{"name": "biplane", "geometry": model}])

Or run: python examples/python/io/model_io.py

[4]:

# Optional: run only if the USDZ was downloaded (see examples/python/io/model_io.py)
import os
from pathlib import Path

usdz = Path("toy_biplane_realistic.usdz")
if usdz.is_file():
    model = o3d.io.read_triangle_model(str(usdz))
    print(f"USD model: {len(model.meshes)} meshes, {len(model.materials)} materials")
else:
    print("Skip USD demo: download toy_biplane_realistic.usdz or run model_io.py")

Skip USD demo: download toy_biplane_realistic.usdz or run model_io.py

Image#

The code below reads and writes an image.

[5]:

print("Testing IO for images ...")
image_data = o3d.data.JuneauImage()
img = o3d.io.read_image(image_data.path)
print(img)
o3d.io.write_image("copy_of_Juneau.jpg", img)

Testing IO for images ...
[Open3D INFO] Downloading https://github.com/isl-org/open3d_downloads/releases/download/20220201-data/JuneauImage.jpg
[Open3D INFO] Downloaded to /root/open3d_data/download/JuneauImage/JuneauImage.jpg
Image of size 800x489, with 3 channels.
Use numpy.asarray to access buffer data.

[5]:

True

The size of the image is readily displayed using print(img).

Both jpg and png image files are supported.