Movatterモバイル変換


[0]ホーム

URL:


Skip to content

Navigation Menu

Sign in
Appearance settings

Search code, repositories, users, issues, pull requests...

Provide feedback

We read every piece of feedback, and take your input very seriously.

Saved searches

Use saved searches to filter your results more quickly

Sign up
Appearance settings

What Drives Compositional Generalization in Visual Generative Models?

NotificationsYou must be signed in to change notification settings

lmb-freiburg/gen-comp-gen.github.io

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

15 Commits
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Static project page for the paper
“What Drives Compositional Generalization in Visual Generative Models?” (Faridet al., 2025).
The live site is a single-page app built with plain HTML/CSS/JS and is intended for GitHub Pages hosting.

Quick Start

  • View online: enable GitHub Pages on themain branch (Settings → Pages → Sourcemain / root).
  • Preview locally: run a lightweight HTTP server from the repo root and open the reported URL in your browser.
cd gen-comp-gen.github.iopython3 -m http.server 4000# visit http://127.0.0.1:4000/

Repo Layout

  • index.html – main landing page with hero, findings, animated figures, tables, and BibTeX block.
  • figures/ – figure assets referenced by the page (PDF/PNG/SVG); see.gitignore for unused variants.
  • figs/teaser.jpg – hero image shown in the header and social previews.
  • .gitignore – ignores auxiliary figure exports that are not linked from the site.

Editing the Site

  1. Update copy or structure directly inindex.html.
  2. Place new visual assets infigures/ (orfigs/ for hero art) and reference them with relative URLs.
  3. If you add or remove assets, adjust.gitignore so only unused variants remain ignored.
  4. Commit and push; GitHub Pages will redeploy automatically after the push.

Animated figures use a tiny dependency-free script at the bottom ofindex.html. To add another carousel:

<divid="fig-anim-xyz"class="anim-fig"data-dwell="1500"><!-- repeated .anim-step elements --></div><script>mountAnimFigure('fig-anim-xyz');</script>

Troubleshooting

  • Broken images: ensure paths match the filenames (case-sensitive) and that the asset is not listed in.gitignore.
  • PDF rendering issues: export to PNG (same basename) if the browser cannot inline the PDF and update thesrc.
  • No GitHub Pages output: verify that Pages is enabled on the correct branch and that the build began in the Pages dashboard.

TODO

High priority

  • Add Orbis compositional videos with REAL, MaskGIT, and DiT (one video per novel composition) – Karim
  • Document the experimental setup underlying the probe accuracy results and clarify their interpretation on the site.
  • Restructure all sections that follow the JEPA section.
  • Add a qualitative results section sourced from Shapes3D.
  • Add a world models section with quantitative and qualitative results – Karim
  • Fix image sizes.
  • Proofread the full site copy.
  • Move shapes 3d figure to the end of the page.
  • Make the shapes 3d figure more readable, clear color legend while scrolling, what should the reader see there.
  • Put the videos as the main teaser part.
  • Remove the main teaser in the beginning and keep title only.

Less priority

  • Add CelebA compostional split images - Karim
  • Add CLEVRER videos (111 videos) - Rajat
  • Fix language system prompt images - Rajat

For questions, contact the authors atfaridk@cs.uni-freiburg.de.

About

What Drives Compositional Generalization in Visual Generative Models?

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Contributors2

  •  
  •  

Languages


[8]ページ先頭

©2009-2025 Movatter.jp