Find your next open-source project

Search by name, purpose, or keyword — or paste a GitHub repository URL

qwdwqfwq/topconf-paper-figure-gallery

Learning Resources

A searchable gallery of Figure 1/teaser designs from six top ML conferences, with optional AI-assisted figure drafting.

1,887
93
JavaScript
NOASSERTION
2 months old

What it is

Top-Conf Paper Figure Gallery collects Figure 1 / teaser images from ICLR, ICML, NeurIPS, CVPR, ACL and AAAI papers published between 2023 and 2026. Its purpose is layout inspiration: before drawing a paper's overview figure, researchers can see how accepted papers structure panels, arrows, comparisons and color schemes. The gallery currently lists 3,452 figures, tagged by visual mode (conceptual, framework, pipeline, architecture, taxonomy, teaser) and searchable by venue, year, acceptance level, title, author and keyword. Oral, Spotlight and Best/Outstanding entries carry tier badges for ICLR, ICML and NeurIPS.

Each card links back to the original paper. The site is static: metadata lives in assets/figures.js, so it can be opened offline after cloning.

Quick start

For browsing, use the online gallery. To run locally:

git clone https://github.com/qwdwqfwq/topconf-paper-figure-gallery.git
cd topconf-paper-figure-gallery
python -m http.server 8000
# open http://localhost:8000

Alternatively, open index.html directly. No backend or build step is needed because the data is inlined.

To try the optional FigureForge drafting tool, open the Forge page, paste a title and abstract or upload a PDF, choose a visual mode, select reference figures from the retrieved results, and then enter a model API key to generate an SVG or bitmap draft. The key is kept in the browser. Forge relies on an external vision/language model, so output quality depends on the key and model you provide; the gallery itself does not generate images without one.

Data pipeline

The repository also ships the scripts used to build the collection. With Python 3.10+ and pip install -r requirements.txt, the pipeline downloads PDFs, crops Figure 1, scores candidates, deduplicates with dHash, and assembles figures.json. Candidate-pool construction, extraction and assembly commands for the main venues are documented in METHODOLOGY.md. Manual review sheets are generated at 40 figures per page, and the README states each image passed human inspection after rule-based rejection of default matplotlib plots, plain tables and screenshots.

Using constraints

  • Image copyright stays with the original authors and publishers. The repository is an attributed index for non-commercial education and research; the stated policy is removal within 72 hours after a verified request, with details in IMAGES_POLICY.md. AAAI material is described as fair-use academic reference.
  • Code is licensed MIT; images are under a separate policy, so redistributing figures requires checking that policy rather than assuming the code license covers them.
  • FigureForge needs a third-party model API key and local browser storage; it is a beta, and the README frames results as first drafts to refine in Figma, draw.io, Illustrator or PowerPoint.
  • The README contains sponsor blocks for commercial LLM API platforms with referral links and sign-up credits. These are advertising placements inside the project, not part of the gallery or the figure-generation code.

HotRepo's take

The one recorded snapshot is 2026-11-18 on the JavaScript daily list, showing 1,882 stars and 84 new stars for that day, with a best position of 9 on that list. That is a single day's figure, so it supports saying the project had a strong day on the JavaScript chart, not a trend. The repository pairs a large, manually reviewed image index with a working browser-side retrieval and drafting page, and several community pull requests added accessibility, sandboxing and data-validation checks. Readers who want pure layout references can use the gallery without any account; the AI drafting step and the sponsor promotions are separate from that core browsing experience.

Written by AI from the project's README, updated . The project's own docs are the source of truth.

You might also like

  1. ywEdAi/teachable-figuresTurns a lesson claim into an interactive explainer figure; an agent skill for teachers and instructional designers.JavaScript200