Skip to content

in browser image viewer with ranking - pairwise-comparison sort through large collections of images, distill the best ones (no generalizing ML though)

License

Unknown, LGPL-2.1 licenses found

Licenses found

Unknown
COPYING.file
LGPL-2.1
COPYING.libgnurx
Notifications You must be signed in to change notification settings

martingalevoiceless/reimagined-happiness

Repository files navigation

In browser image viewer with ranking.

mac/linux setup (needs node.js installed already):

  1. cp pyramid.ini.example pyramid.ini
  2. vim pyramid.ini, edit to point to your image database and to a correct tempdir
  3. bash setup_web.sh -- or literally just cd web; npm install; node_modules/.bin/webpack
  4. bash run_pyramid.sh -- if needed, downloads fully isolated copy of pyenv, uses it to install 3.7.0, then sets up a virtualenv; then runs the server

windows:

  1. install chocolatey: http://chocolatey.org
  2. make a directory somewhere to put your code, open a terminal, cd C:/Users/username/that/directory
  3. set up:
choco install python nodejs git ffmpeg
git clone https://github.com/martingalevoiceless/reimagined-happiness.git
cd reimagined-happiness/web/
npm install
node_modules\.bin\webpack --config webpack.config.js
cd ..
python -m venv ve
ve\Scripts\activate
pip install python-magic-bin
pip install https://download.pytorch.org/whl/cu100/torch-1.0.0-cp37-cp37m-win_amd64.whl
pip install -e .
copy pyramid.ini.example pyramid.ini
  1. edit pyramid.ini:
base = C:/Users/username/path/to/your/files/
tempdir =  C:/Users/username/AppData/Local/Temp/images_app/
  1. run it: pserve pyramid.ini

usage

now open http://localhost:8844 and have fun!

make sure to not expose your site to the internet without authentication (you don't by default) if you don't want to accidentally offer your image library to the world or have randos providing ranking feedback!

views:

  • ranking compare view: http://localhost:8844/_/compare/
  • similarity compare view: http://localhost:8844/_/similarity/ - answer the question "which of these things is the least like the others?" repeatedly, eventually get a bunch of vectors representing your images. Pulls from the ranked data, rather than the full dataset, and prioritizes high-ranking items, so you'll need to rank some first to use this at the moment. currently no tools for making use of the similarity data, though I'm intending to use it to try to predict variation in ranking from session to session.
  • file browser view: http://localhost:8844/_/p/
  • history view: http://localhost:8844/_/history/<file hash> (can be opened by shift-j or shift-o in compare)

keyboard layout in main view:

layout image link to layout editor

0 top/right strongly preferred (use sparingly, normal prefer tracks how long you take to respond to estimate how easy it was to rank)
o top/right preferred
i too close to easily call
j bottom/left preferred
h bottom/left strongly preferred

p view top/right full
l show top/right with a different pairing

k incomparable: doesn't even make sense to compare these images
, goes well: they're so incomparable that viewing them together is better than either one on its own

n view bottom/left full
m show bottom/left with a different pairing

Shift+o view history for top/right
Shift+j view history for bottom/left

on windows, Super is the win key, and on mac, it's command.
Super+o lock to top/right image
Super+i unlock, go back to showing the most useful comparison
Super+j lock to the bottom/left image

[ alias for browser back
] alias for browser forward
backspace alias for browser back
shift-backspace alias for browser forward

t switch between horizontal and vertical view
d show/hide numerical info

keyboard layout in file browser view

j go to previous image
i compare this image
o go to next image

p return to last compare
n return to last compare

d show/hide numerical info

[ alias for browser back
] alias for browser forward
backspace alias for browser back
shift-backspace alias for browser forward

About

in browser image viewer with ranking - pairwise-comparison sort through large collections of images, distill the best ones (no generalizing ML though)

Resources

License

Unknown, LGPL-2.1 licenses found

Licenses found

Unknown
COPYING.file
LGPL-2.1
COPYING.libgnurx

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published