vis15

vis15: Reproducing VIS'15 Diderot Paper Figures

This page documents how to reproduce the Diderot-generated figures in the VIS 2015 paper about Diderot. In most cases, the figures can be reproduced by compiling and running the Diderot programs that appeared within the published paper, modulo typo fixes and other superficial changes. Where the figures used data that was not publicly available, this shows how to make a related figure with public or synthetic data.

The figures are listed here, and below are more details about running the scripts. Please ask questions about these instructions, and give feedback, in the diderot-language Google group.

(00-cmap) Needed for other figures: colormap generation

(01-mousehand) Figure 1: 2D image data, field, and isocontours

Figure 2: Listing of isoptcl.diderot program for uniformly sampling a 2D isocontour with a system of interacting particles.

(Figure 3: Diagram of bulk synchronous parallel (BSP) execution model)

Figure 4: Listing of volrend.diderot program for volume rendering a scalar field, which is modified (as described in Section 4.1) to produce Figures 5, 6, and 7.

(05-canny) Figure 5: Volume rendering Canny edges

(06-vrflow) Figure 6: Volume rendering flow properties

(07a-vrstress) Figure 7a: Volume rendering stress tensors

(07b-vrdti) Figure 7b: Volume rendering diffusion tensors

Figure 8: Listing of lic2d-published.diderot program for doing line integral convolution (LIC) in a vector field; see Figure 10.

Figure 9: Listing of sline.diderot program for computing streamlines in a vector field; see Figure 10.

(10-2dflow) Figure 10: LIC and streamlines in 2D flow

(Figure 11: A comparison of parallel speedup on benchmark programs)

Running these programs requires some set-up (this is checked in 0-common.sh):

• These Diderot programs depend on the vis15 branch of the compiler (called diderotc), which can be built by following these directions (search for: Building Diderot and these examples). You will need to add the vis15 diderotc to your path.
• The scripts assume that you have built Teem from its latest source. August 2017 updates (revision r6319) added the ability, used in these scripts, to specify gamma to unu quantize and overrgb with -g srgb. Following the Diderot build instructions, you should add $DDRO_ROOT/teem-util to your path so that you have the latest unu and overrgb.
• The epsdraw.c program has to be compiled (and linked with Teem) to create some of the figures. This depends on setting environment variable TEEM_INSTALL. If you've followed the Diderot build instructions, you can do this with (in bash): export TEEM_INSTALL=$DDRO_ROOT/teem-util.
• Some of the programs assume that you have epstopdf and convert in your path.
• Finally, some of the figures use data or programs from the Diderot examples themselves, so you should set environment variable DDRO_EXAMPLES to the path of your checkout of https://github.com/Diderot-Language/examples

While Diderot is used as the computational core for all the figures, various pre- and post-processing steps are also needed. Gordon Kindlmann (GLK) uses bash and unu for these steps, because GLK's day-to-day work uses bash and unu for command-line data hacking, and also because pieces of image metadata (in NRRD files) that Diderot relies on can be easily manipulated with unu. With time, we may develop comparable solutions based on python or IPython, or other GUI-based tools that exploit how Diderot programs can be compiled to libraries.

Thus, each figure is made with a fairly low-tech sequence of scripts, each starting with an integer (such as, in 05-canny, 1-dataprep.sh, 2-vr-iso.sh, 3-vr-canny.sh, and then 4-cleanup.sh). These scripts should be run in sequence (with no additional command-line arguments), though the final N-cleanup.sh script is optional. All the scripts have detailed comments to describe what is going on, as well as commentary to explain the various (generally superficial) differences between what was in the paper, and what is generated here. Some additional commentary is within the Diderot programs. The intended final output of the scripts is saved in the ref subdirectory of each per-figure directory.