"Real-time Spectral Cube Colouring and Processing with Graphics Shaders", presented at Astroinformatics 2017, Cape Town, South Africa.

With the vast adoption of the Graphics Processing Unit (GPU) for general purpose computing, one should not forget about the practicality of the GPU for fast scienti c visualisation. As astronomers have increasing access to three-dimensional (3D) data from instruments and facilities like integral eld units and radio interferometers, visualisation techniques such as volume rendering offer means to quickly explore spectral cubes as a whole. Using the open source software shwirl, we demonstrate how transfer functions and graphics shaders can be exploited to provide new astronomy- speci c explorative colouring methods. In particular, we present transfer functions speci cally designed to produce intuitive and informative 3D visualisations of spectral cube data — moving beyond classical techniques imported from medical imaging. We compare their utility to classic colour mapping. Additionally, we present how common computation like ltering, smoothing, and line ratio algorithms can be integrated as part of the graphics pipeline for rapid visual feedback. We discuss how this can be achieved by utilising the parallelism of modern GPUs along with a shading language, letting astronomers apply these new techniques at interactive frame rates.

Transcript

1. Real-time spectral cube colouring and processing with graphics shaders Dany Vohl | Astroinformatics 2017 | @danyvohl In collaboration with Christopher J. Fluke, Amr H. Hassan, and David G. Barnes (Monash University)
2. APERTIF Cube Dimensions 
 2048 x 2048 (spatial) x 16384 (spectral)
 0.25 TB Large-Scale Spectral-Cube Surveys ASKAP Cube Dimensions 
 6144 x 6144 x 16384
 ~1 TB • large # of spectral-cubes, each with large # of sources • e.g. Apertif: 20,000 spectral-cubes, each containing ~100 sources
 (Verheijen et al. 2009, Punzo et al. 2015) Introduction Petascale Astronomy Era
3. Moment maps 0th moment overall gas distribution 1st moment gas velocity field Spectral cubes in 2D Classic Visualisation Data : Katharina Lutz NGC 3261
4. Moment maps 0th moment overall gas distribution 1st moment gas velocity field Spectral cubes in 2D Classic Visualisation Data : Katharina Lutz NGC 3261 Dwarf galaxy!
5. 3D visualisation of spectral cube Hassan et al., MNRAS, (2012) Perkins et al., NA, (2013) Big Data Framework Taylor, A&C, (2015) Kent, PASP, (2013) Ferrand et al., 
 ArXiV, (2016) Explorative visualisation Punzo et al., A&C, (2015, 2016) Beaumont et al., 
 ASCL, (2014) Vohl et al.,
 PeerJ CS, (2016) encube Signal processing & 
 visual analytics
6. 3D visualisation of spectral cube THINGS and IMAGE HD 
 datasets rendered 
 with encube @ CAVE2, 
 Monash University. Vohl et al.,
 PeerJ CS, (2016) Large-scale visual analytics framework
7. 3D visualisation of spectral cube Compute algorithms commonly pre-computed New colouring techniques to enhance comprehension What can shaders do for us? Graphics Shaders background: techspot.com
8. 3D visualisation of spectral cube The Graphics Pipeline Vohl et al., MNRAS, (2017) background: techspot.com
9. Output pixel Image plane Sampled voxels Spectral cube Observer Ray Vohl et al., MNRAS, (2017) Transfer function Maximum Intensity Projection / Radiative Transfer MIP 0 1. Find maximum voxel 2. Set colour and intensity with Maximum
10. ra dec vel ra dec ra vel vel dec Vohl et al., MNRAS, (2017) 0th moment-inspired Transfer function Maximum Intensity Projection
11. Vohl et al., MNRAS, (2017) First Moment-inspired transfer function Advanced Colouring Techniques MIP 1 1a. Find maximum voxel 1b. Note voxel coordinate 2a. Set colour with Z coordinate 2b. Set transparency with Maximum Output pixel Image plane Sampled voxels Spectral cube Observer Ray
12. dec ra vel (km/s) 680 230 Vohl et al., MNRAS, (2017) 1st moment-inspired Advanced Colouring Techniques ra dec vel
13. Vohl et al., MNRAS, (2017) RGB colouring Informs about all three dimensions ra dec vel front ra dec vel back ra dec vel More exotic colouring Advanced Colouring Techniques
14. Common processing in shader Filtering / Smoothing Visualisation / GPGPU
15. 3D visualisation of spectral cube Vohl et al., MNRAS, (2017) Timing Smoothing Kernels Smoothing kernel size Smaller Larger Locally On the cloud
16. Vohl et al., MNRAS, (2017) Computing emission line ratio SAMI (4) Hα [NII] [NII]/Hα (1) [NII]/Hα (2) SAMI (1) GAMA-511867 taken from the SAMI Survey Proof of concept Visualisation / GPGPU
17. Shaders showed efficient for custom visualisations Allow real-time processing • Interactive • Opens new ways to explore our data In some cases, hybrid CPU/GPU methods? Lessons learned
18. Custom transfer functions and shaders can play an important role in the development of future visualisation and analysis astronomical software. Can be transferred to range of fields 
 (source finding, simulation, …) Final thoughts
19. Experiment sandbox Python / GLSL github.com/macrocosme/shwirl Final thoughts ascl:1704.003