dvid-point-cloud 0.1.0

Library for creating point clouds for sparse volumes within DVID

dvid-point-cloud

Library for creating point clouds for sparse volumes within DVID, with support for multi-scale sampling and vectorized operations.

Installation

pip install dvid-point-cloud

Or install from source:

git clone https://github.com/username/dvid-point-cloud.git
cd dvid-point-cloud
pip install -e .

For development, install with extra dependencies:

pip install -e ".[dev]"

Usage

Basic Point Cloud Generation

Generate a uniform point cloud from a DVID label:

import dvid_point_cloud as dpc

# Define parameters
server = "http://my-dvid-server.janelia.org"
uuid = "bc9a0f"  # Hexadecimal string identifying the version
label_id = 189310  # The neuron/segment ID
density = 0.01  # Sample 1% of the voxels

# Generate the point cloud
point_cloud = dpc.uniform_sample(server, uuid, label_id, density)

# point_cloud is a numpy array with shape (N, 3) 
# where each row is an XYZ coordinate
print(f"Generated a point cloud with {len(point_cloud)} points")

Multi-Scale Sampling

Sample points at different resolution scales, where the downsampling factor is 2**scale:

# Sample at scale 0 (original resolution)
points_s0 = dpc.uniform_sample(server, uuid, label_id, density, scale=0)

# Sample at scale 2 (downsampled by factor of 4)
points_s2 = dpc.uniform_sample(server, uuid, label_id, density, scale=2)

# Sample at scale 3 (downsampled by factor of 8)
points_s3 = dpc.uniform_sample(server, uuid, label_id, density, scale=3)

Fixed Count Sampling

Sample a specific number of points instead of a density:

# Sample 1000 points
points = dpc.uniform_sample(server, uuid, label_id, 1000)

DataFrame Output

Get results as a pandas DataFrame:

# Get results as a DataFrame with x, y, z columns
df_points = dpc.uniform_sample(
    server, uuid, label_id, density, 
    output_format="dataframe"
)
print(df_points.head())

Supervoxel Sampling

Sample from supervoxels instead of agglomerated bodies:

# Sample from supervoxels
sv_points = dpc.uniform_sample(
    server, uuid, label_id, density,
    supervoxels=True
)

Multiple Bodies

Generate point clouds for multiple bodies:

body_ids = [189310, 189311, 189312]
# sample 10% of voxels
body_points = dpc.sample_for_bodies(
    server, uuid, "segmentation", body_ids, 0.1, scale=0
)

# body_points is a dictionary mapping body IDs to point clouds
for body_id, points in body_points.items():
    print(f"Body {body_id}: {len(points)} points")

Guarantee Sampling With No Duplicates

The default sampling function does not guarantee results have no duplicates (although it is unlikely). You can use a sampling function with no duplicates guaranteed:

# Sample 1000 points with no duplicates
points = dpc.uniform_sample(server, uuid, label_id, 1000, 
                        sample_from_rles_func=accurate_sample_rles)

There is a performance hit for very large sparse volumes as shown in benchmarks folder.

Neuroglancer Visualization

Generate Neuroglancer-compatible JSON for point cloud visualization:

import json

# First generate the point cloud
points = dpc.uniform_sample(server, uuid, label_id, 1000, scale=0)

# Convert it to Neuroglancer JSON
layer_json = dpc.point_cloud_to_neuroglancer_json(
    points,
    name="my-point-cloud", 
    color="#00ff00",  # Green color
    point_size=2.0    # Slightly larger points
)

# Print or save the JSON
print(json.dumps(layer_json))

# You can then copy this JSON into a Neuroglancer annotation layer
# or use it programmatically with the Neuroglancer Python API

Requirements

• Python 3.7+
• numpy
• pandas
• requests

Performance

The library uses vectorized operations for efficient point cloud generation. Sampling performance improves at higher scale levels, making it practical to generate large point clouds quickly.

The scale parameter determines the downsampling factor (2**scale), so:

• Higher scales = faster processing time
• Lower scales = more detailed point clouds

Development

Running Tests

# Run all tests
pytest

# Run tests with coverage report
pytest --cov=dvid_point_cloud

# Run a specific test file
pytest tests/test_sampling.py

# Run a specific test
pytest tests/test_sampling.py::test_uniform_sample_integration

Linting and Type Checking

# Run linter
flake8 dvid_point_cloud

# Run type checker
mypy dvid_point_cloud