A starfield renderer for the Gaia DR2 catalog.
git clone --recurse-submodules git@github.com:lancelet/star-stream.git
cd star-stream
cabal new-build
ghcid --command='cabal new-repl'The image below shows an equirectangular, galactic-coordinate projection of all 1.69 billon sources in the Gaia DR2 catalog. This "integrated flux" image was created by reading each of the Gaia source catalog files (compressed CSVs) and adding the contribution of flux from each source to each pixel it influences. The pixel influences are calculated by multiplying the measured flux in each of the three Gaia channels by a convolution filter (Mitchell-Netravali) centred on each pixel.
The amount of data to be processed (~588 GB compressed) is too big to fit in memory on the machines running this job (typically a MacBook Pro or an iMac), so it is streamed in smaller batches, currently corresponding to the original source CSV files. The project name (star-stream) comes from this necessity to stream the data and accumulate the image.
The output from the flux integration process is a floating-point HDR image, which is tweaked (manually, for now) with a non-linear exposure in an image editor.
This is only an initial exploration. Improvements planned for the future include:
- Add a galactic coordinate Mollweide projection
- Adjust filtering ellipse according to the local anisotropy of the projection
- Bucket the source observations into an acceleration structure for faster rendering of subsets
- Render temporal offsets for sources with proper motion
- Correlation of sources with a catalog that contains popular names of stars
- Etc.
This project uses the Gaia star catalog as a data source. Gaia is a space observatory launched by the European Space Agency (ESA) in 2013.
The second data release from Gaia, called GDR2, was made public on 25 April 2018. It consists of GZipped CSV files, stored on an ESA server, comprising a total of ~588 GB of compressed data.
The gaia-fetch package contains a utility to download these files. Please note that the download may be a long process. (It took 3 weeks in Sydney, Australia!)