microviewer 1.20.0

Visualize 3D numpy arrays in the browser.

microviewer

Multiplatform 3D numpy image browser based viewer.

from microviewer import view, hyperview, objects

view(numpy_image) # for gray and color images
view(numpy_image, seg=True) # for segmentation labels
view(numpy_image, seg=True, port=8082)

hyperview(image, labels) # interactive overlay

# uses vtk to visualize meshes, bounding boxes, 
# skeletons, and point clouds
objects([ mesh, skeleton, bbox, point_cloud ])

uview image.npy.gz # view as image
uview labels.npy.gz --seg # view as segmentation
uview image.npy.gz labels.npy.gz # view image w/ overlay
uview image.npy.gz --paint # view image w/ blank overlay

Visualize 3D numpy arrays in your browser without difficult installation procedures or reformatting your data. The code is CPU based and the image is uncompressed in memory. You're limited to images that are at most 2^31 bytes large (~2.1 GB) due to browser limitations.

Features

• 3 axis visualization of 3D images.
• Grayscale images with segmentation overlay.
• Segmentation selection with brush tools.
• Direct voxel painting.
• Save segmentation as .npy or .ckl, an advanced compresssion format.
• Undo/Redo

3D Image Support

• 8-bit grayscale 3D images
• color images (including 3 channel 3D images)
• floating point images
• boolean images
• segmentation labels
• .npy, .ckl, .nrrd, or .nii format

For .ckl, .nrrd, and .nii formats, you must install crackle-codec, pynrrd, and nibabel respectively. All are installed by the "all_formats" optional installation argument.

3D Object Support

• osteoid Skeletons
  • Color By: Connected Component, Radius, Cross Sectional Area
• zmesh and CloudVolume Meshes
• Point Clouds
• Bounding Boxes

Example

import numpy as np
import zmesh
import kimimaro
import crackle
import microviewer
from osteoid.lib import Bbox

labels = crackle.load("connectomics.npy.ckl.gz", label=62347522)
resolution = np.array([16,16,40])

skel = kimimaro.skeletonize(
  labels, 
  teasar_params={
    'scale': 3,
    'pdrf_exponent': 8,
  },
  anisotropy=resolution,
)[1]

mesher = zmesh.Mesher(resolution)
mesher.mesh(labels)
mesh = mesher.get(1)

bbox = Bbox([0,0,0], np.array(labels.shape) * resolution)

microviewer.objects([ bbox, mesh, skel ], skeleton_color_by='radius')


# x = cross sectional area (if precomputed)
# r = radius (if precomputed, but usually is)
# c = connected components
microviewer.objects([ bbox, mesh, skel ], skeleton_color_by='x', skeleton_heatmap={
  "min": 100, # automatically determined if unspecified
  "max": 100000, # automatically determined if unspecified
  "scale": "log", # linear (default) or log
})

Installation

pip install microviewer

By default, only .npy files are supported. Additional formats are supported via optional downloads.

Possible optional arguments:

all_formats, nii, nrrd, ckl, jxl,
objects,
all

Note: jxl and objects have relatively heavy downloads.

pip install "microviewer[all]"

History

This microviewer package has been a part of CloudVolume since 2018, but is now broken out into its own package for more flexible wider use. Microviewer uses a modified version of https://github.com/seung-lab/data-cube-x/ (2016) to represent the array in Javascript, which was originally developed for eyewire.org.