qwdwqfwq/topconf-paper-figure-gallery

3,516 curated Figure 1 / teasers from ICLR, ICML, NeurIPS, CVPR, ACL, AAAI (2023-2026), with tier badges and FigureForge retrieval-augmented figure drafting

What it solves

Researchers writing papers often struggle to design a compelling Figure 1 (the main teaser figure) because they lack a sense of what top-conference papers actually look like. This project collects thousands of Figure 1 images from six top AI/ML conferences (ICLR, ICML, NeurIPS, CVPR, ACL, AAAI) across 2023–2026, making them searchable and filterable as a design reference. It also provides FigureForge, a browser-based tool that drafts a Figure 1 for a user's own paper, grounded in these reference figures instead of producing generic flowcharts.

How it works

The gallery is a static website (HTML/CSS/JS) with data inlined in a JavaScript file, so it works offline with no backend. Each figure is manually reviewed after an automated pipeline: conference index → PDF download → Figure 1 crop → quality scoring with heuristic rules (removing default matplotlib charts, plain tables, GUI screenshots, etc.) → dHash de-duplication → human page-by-page check. Filters let users sort by venue, year, acceptance tier (Best/Oral/Spotlight with color badges), and visual pattern (conceptual, framework, pipeline, architecture, taxonomy, teaser). Full-text search covers titles and authors.

FigureForge runs entirely in the browser. Users describe their paper (paste abstract/PDF, or give a one-line input→method→output story), choose a visual pattern, and select reference figures. A CLIP + BM25 hybrid search ranks all 3,452 figures by relevance. In two-stage mode, a vision model first summarizes the common layout, elements, palette, and hierarchy of the selected references, then generates a draft in either SVG (editable, accurate text) or bitmap (publication-style layout). The user brings their own API key for the LLM; models from many providers (OpenAI, DeepSeek, Anthropic, etc.) are supported. The CLIP model and gallery index are bundled locally, so paper content and keys never leave the device.

Who it's for

Researchers and students in AI/ML who want to design better paper figures before submission. It is useful for both inspiration (browsing curated high-tier figures) and practical draft generation (using FigureForge to get a first-draft layout). The project is open source and welcomes community contributions to add figures, fix labels, or extend to new venues like CHI, CoRL, RSS.

Highlights

  • Curated, hand-reviewed gallery of 3,452 Figure 1/teaser images from six top conferences (2023–2026), each linked to the source paper.
  • Filter and search by venue, year, acceptance tier (Best/Oral/Spotlight badges), and visual pattern; masonry layout with lightbox.
  • FigureForge: browser-based, no-signup tool that uses CLIP+BM25 retrieval to ground a generated draft in real top-conference layouts, avoiding generic flowcharts.
  • Two output modes: SVG for editable text and bitmap for quick layout checking; supports many LLM providers through bring-your-own-key.
  • Fully static and offline-capable; complete open-source pipeline for reproducing the gallery from conference proceedings.
  • Clear image policy: educational use with attribution, and a 72-hour takedown on request.

Related

  • Project
    OpenEnvision/Awesome-Multimodal-Modeling

    Awesome‑Multimodal‑Modeling is a community‑curated “awesome‑list” that surveys multimodal AI models. It defines four clear families—Traditional, MLLMs, Unified, and Native—explains their architectural differences, and links to open‑source model collections on Hugging Face. The repo serves researchers and engineers who need a structured view of the rapidly evolving multimodal landscape.

  • Project
    zhengxuJosh/Awesome-RAG-Vision

    Awesome‑RAG‑Vision is a curated “awesome‑list” that gathers papers, surveys, tutorials and benchmarks on Retrieval‑Augmented Generation for computer‑vision tasks (image/video understanding, generation, embodied AI, document‑centric RAG, etc.). It provides organized links to recent research (2023‑2026) and resources for building multimodal RAG pipelines, but contains no code of its own.

  • Project
    xtreme1-io/xtreme1

    Xtreme1 is an open‑source, Docker‑based platform for annotating, curating, and managing multimodal training data (2‑D images, 3‑D LiDAR, sensor‑fusion). It offers UI‑driven labeling tools, ontology management, built‑in pre‑labeling models (YOLOR, OpenPCDet), quality monitoring, and a beta RLHF interface for LLMs. Installable via a single `docker compose up` command, it runs on any OS with Docker and can optionally use GPU‑accelerated model containers. Licensed under Apache‑2.0 and hosted by the LF AI & Data Foundation.

  • Project
    VigoZhao/AI-Visual-Prompt-Cookbook

    A curated library of 169 JSON‑based visual‑style prompts for text‑to‑image models. Users copy a `style.json`, fill in a few variables, paste it into a multimodal LLM (e.g., ChatGPT‑4‑Vision, Gemini‑3‑Pro Image) and get a consistent, stylized image. The repo includes previews, copy‑ready prompts, and a validator script.

  • Dispatch
    LLM Attention Visualization Tool Shows How Tokens Influence Generation

    A new browser‑based visualizer lets users hover over generated tokens to see which past tokens the model attended to, revealing how attention drives copying and synthesis in LLMs.