data and plots for "Reconfiguring the Imaging Pipeline for Computer Vision"
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
data
vg-tooltip
.gitignore
LICENSE
Makefile
README.md
error.meta.js
flatten.py
index.html
package.json
ressweep.meta.js
single.meta.js
singles.html
sweep.meta.js
sweeps.html
table.py

README.md

Visualization for the Vision Mode Paper

This is a set of scripts for visualizing the data from the paper "Reconfiguring the Imaging Pipeline for Computer Vision" in ICCV 2017 by Mark Buckler, Suren Jayasuriya, and Adrian Sampson.

We use Vega-Lite to draw plots. To produce plots, first install the dependencies using Yarn. Just type:

$ yarn

Then, you can produce a lot of SVG files by typing make. Or, to get PDF versions, make sure you have librsvg's rsvg-convert and then run make pdf. You can try using -j9 or something to parallelize the plot generation.

There are also two HTML files here that let you view the visualizations interactively. For example, you can hover over bars to get exact numbers.

Hacking

The Vega-Lite source for the plots is in the files *.meta.js. These are MetaJSON files with a few different parameters that can be used to generate a bunch of similar but slightly different plots. See the Makefile for how to translate these into complete Vega-Lite JSON source programs.

Credits

If you use this data, please cite our ICCV paper. The license is MIT.