Clara Viz

NVIDIA Clara Viz is a platform for visualization of 2D/3D medical imaging data. It enables building applications that leverage powerful volumetric visualization using CUDA-based ray tracing.

Clara Viz offers a Python Wrapper for rapid experimentation. It also includes a collection of visual widgets for performing interactive medical image visualization in Jupyter Lab notebooks.

Known issues

On Windows, starting with Chrome version 91 (also with Microsoft Edge) the interactive Jupyter widget is not working correctly. There is a delay in the interactive view after starting interaction. This is an issue with the default (D3D11) rendering backend of the browser. To fix this open chrome://flags/#use-angle and switch the backend to OpenGL.

Requirements

• NVIDIA GPU: Pascal or newer, including Pascal, Volta, Turing and Ampere families
• NVIDIA driver: 450.36.06+

Documentation

https://docs.nvidia.com/clara-viz/index.html

Quick Start

Installation

This will install all Clara Viz packages use pip:

$ pip install clara-viz

Clara Viz is using namespace packages. The main functionality is implemented in the 'clara-viz-core' package, Jupyter Notebook widgets are found in the 'clara-viz-widgets' package. So for example if you just need the renderer use

$ pip install clara-viz-core

Use interactive widget in Jupyter Notebook

Install the Jupyter notebook widgets.

$ pip install clara-viz-widgets

Start Jupyter Lab, open the notebooks in the notebooks folder. Make sure Git LFS is installed when cloning the repo, else the data files are not downloaded correctly and you will see file open errors when using the example notebooks.

from clara.viz.widgets import Widget
from clara.viz.core import Renderer
import numpy as np

# load a RAW CT data file (volume is 512x256x256 voxels)
input = np.fromfile("CT.raw", dtype=np.int16)
input = input.reshape((512, 256, 256))

display(Widget(Renderer(input)))

Render CT data from Python

from PIL import Image
import clara.viz.core
import numpy as np

# load a RAW CT data file (volume is 512x256x256 voxels)
input = np.fromfile("CT.raw", dtype=np.int16)
input = input.reshape((512, 256, 256))

# create the renderer
renderer = clara.viz.core.Renderer(input)

# render to a raw numpy array
output = renderer.render_image(1024, 768, image_type=clara.viz.core.RAW_RGB_U8_DEPTH_U8)
rgb_data = np.asarray(output)[:, :, :3]

# show with PIL
image = Image.fromarray(rgb_data)
image.show()

Use within a Docker container

Clara Viz requires CUDA, use a base container from https://hub.docker.com/r/nvidia/cuda for example nvidia/cuda:11.4.2-base-ubuntu20.04. By default the CUDA container exposes the compute and utility capabilities only. Clara Viz additionally needs the graphics and video capabilites. Therefore the docker container needs to be run with the NVIDIA_DRIVER_CAPABILITIES env variable set:

$ docker run -it --rm -e NVIDIA_DRIVER_CAPABILITIES=graphics,video,compute,utility nvidia/cuda:11.4.2-base-ubuntu20.04

or add:

ENV NVIDIA_DRIVER_CAPABILITIES graphics,video,compute,utility

to your docker build file. See https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/user-guide.html#driver-capabilities for more information.

Dockerfile

After building the Dockerfile and tagging it e.g. claraviz:latest run in terminal or use run_docker.sh

docker run -it --rm \
    --gpus=all \
    -p 8888:8888 \
    --name clara-viz \
    --volume $(pwd):/home/jovyan/work \
    -e NVIDIA_DRIVER_CAPABILITIES=graphics,video,compute,utility \
    claraviz:latest

WSL (Windows Subsystem for Linux)

Currently Clara Viz won't run under WSL because OptiX is not supported in that environment.

Acknowledgments

Without awesome third-party open source software, this project wouldn't exist.

Please find LICENSE-3rdparty.md to see which third-party open source software is used in this project.

License

Apache-2.0 License (see LICENSE file).

Copyright (c) 2020-2021, NVIDIA CORPORATION.