figures4papers README

The project's README from github.com/ChenLiu-1996/figures4papers. As one markdown file: README.md.

Figures for Papers


LinkedIn Twitter Follow Google Scholar

I am Chen Liu (刘晨), a Computer Science PhD Candidate at Yale University.

This is a centralized repository of my own Python scripts for high-quality figures.

These figures are published at top venues, including Nature Machine Intelligence, ICML, NeurIPS, ECCV, etc. 🎉

Please feel free to cite any of these papers if you find them relevant or helpful. 🎓
在符合学术规范的前提下，欢迎大家狠狠引用。 🎓


Bar plots for quantitative comparison

Bar plots for composition breakdown

3D spheres

Radar plots                   Line plots

Radar comparison Post-training comparison

Concept plots

Trend plots

Miscellaneous: figures not made end-to-end in Python

These figures were made partially in Python. I included them to acknowledge the time and efforts I spent on them.






LLM skill integration

I want to show appreciation to my friend Shan Chen who suggested doing this.

The scientific figure making skill lives in scientific-figure-making/. Demo figures live in assets/. Project-specific scripts and outputs live in figure_*/.

Skill folder hierarchy

scientific-figure-making/
├── SKILL.md                              # Quick reference: metadata, when to use, patterns, links
└── references/
    ├── api.md                            # API/conventions to implement (palette, helpers, export)
    ├── common-patterns.md                # Reusable figure patterns
    ├── demos.md                          # Real-world figure_* projects (with URLs)
    ├── design-theory.md                  # Style rationale and design principles
    └── tutorials.md                      # Step-by-step guides

Using this skill in an AI coding agent

No installation (path-based)

You can use this skill without installing anything: open this repo in your AI coding agent (e.g. Cursor, Claude Code, etc.) and reference the skill by path in your prompts. The agent reads scientific-figure-making/SKILL.md and the references/ files from the repo—no symlinks or plugins required.

Simple AI workflow

  1. Open this repository in your AI coding agent (e.g. Cursor).
  2. Ask the AI to create or update a plotting script in your target folder (for example figure_PROJECT_NAME/).
  3. In your prompt, explicitly ask it to follow scientific-figure-making/SKILL.md and scientific-figure-making/references/design-theory.md.
  4. Run the generated script and check the exported figure.

Prompt template (copy/paste)

Create a publication-quality figure script at <target_path>.
Use the Scientific Figure Making skill conventions from:
- scientific-figure-making/SKILL.md
- scientific-figure-making/references/design-theory.md
- scientific-figure-making/references/api.md (palette, helpers, export)

Implement or adapt the patterns (apply_publication_style, make_* helpers, finalize_figure). See figure_* folders for reference scripts.
Input data: <describe your data or paste arrays>.
Output files: <name>.png and <name>.pdf.
Keep the style consistent with this repository.
Install as a skill (symlink)

From the repository root, run:

Agent Commands
Cursor mkdir -p ~/.cursor/skills then ln -s "$(pwd)/scientific-figure-making" ~/.cursor/skills/scientific-figure-making
Claude Code mkdir -p ~/.claude/skills then ln -s "$(pwd)/scientific-figure-making" ~/.claude/skills/scientific-figure-making
Codex mkdir -p ~/.codex/skills then ln -s "$(pwd)/scientific-figure-making" ~/.codex/skills/scientific-figure-making

Restart the agent (or refresh its skill list) after linking. You can then invoke or cite the skill by name in addition to using path-based references when the repo is open.

ImmunoStruct (Nature Machine Intelligence 2026)

nature PDF Huggingface Huggingface GitHub Stars

@article{givechian2026immunostruct,
  title={ImmunoStruct enables multimodal deep learning for immunogenicity prediction},
  author={Givechian, Kevin Bijan and Rocha, Jo{\~a}o Felipe and Liu, Chen and Yang, Edward and Tyagi, Sidharth and Greene, Kerrie and Ying, Rex and Caron, Etienne and Iwasaki, Akiko and Krishnaswamy, Smita},
  journal={Nature Machine Intelligence},
  volume={8},
  pages={70--83},
  year={2026},
  publisher={Nature Publishing Group UK London}
}
LM-Dispersion (ICML 2026)

OpenReview ICML 2026 Project Page arXiv PDF GitHub Stars

@inproceedings{liu2026dispersion,
  title={Dispersion loss counteracts embedding condensation and improves generalization in small language models},
  author={Liu, Chen and Sun, Xingzhi and Xiao, Xi and Van Tassel, Alexandre and Xu, Ke and Reimann, Kristof and Liao, Danqi and Gerstein, Mark and Wang, Tianyang and Wang, Xiao and Krishnaswamy, Smita},
  booktitle={International Conference on Machine Learning},
  year={2026},
  organization={PMLR}
}
VIGIL (ECCV 2026)

OpenReview Project Page arXiv PDF

@inproceedings{xiao2026vigil,
  title={Staying VIGILant: Mitigating Visual Laziness via Counterfactual Visual Alignment in MLLMs},
  author={Xiao, Xi and Liu, Chen and Liao, Chih-Ting and Zhang, Yunbei and Lan, Qizhen and Wei, Yuxiang and Zhao, Lin and Wang, Janet and Gu, Jianyang and Ye, Muchao and Wang, Tianyang and Xu, Hao},
  booktitle={European Conference on Computer Vision},
  year={2026},
  organization={Springer}
}
RNAGenScape

arXiv PDF

@article{liao2025rnagenscape,
  title={RNAGenScape: Property-Guided, Optimized Generation of mRNA Sequences with Manifold Langevin Dynamics},
  author={Liao, Danqi and Liu, Chen and Sun, Xingzhi and Tang, Di{\'e} and Wang, Haochen and Youlten, Scott and Gopinath, Srikar Krishna and Lee, Haejeong and Strayer, Ethan C and Giraldez, Antonio J and Krishnaswamy, Smita},
  journal={arXiv preprint arXiv:2510.24736},
  year={2025}
}
Brainteaser (NeurIPS 2025)

OpenReview NeurIPS 2025 HuggingFace Dataset arXiv PDF GitHub Stars

@inproceedings{han2025creativity,
  title={Creativity or brute force? using brainteasers as a window into the problem-solving abilities of large language models},
  author={Han, Sophia and Dai, Howard and Xia, Stephen and Zhang, Grant and Liu, Chen and Chen, Lichang and Nguyen, Hoang H and Mei, Hongyuan and Mao, Jiayuan and McCoy, R Thomas},
  journal={Advances in Neural Information Processing Systems},
  volume={38},
  pages={146950--147004},
  year={2025}
}

figures4papers (ChenLiu-1996/figures4papers)

Searching Database…

8.7K ·GitHub
Figures4papers
What is this?A personal collection of Python scripts (plus an AI-agent skill) for making the kind of high-quality figures the author used in papers published at venues like Nature Machine Intelligence, ICML, NeurIPS, and ECCV.
Is this a Python library I pip install, or something else?Not a pip-installable library — it is a repo of standalone example scripts plus a 'scientific-figure-making' skill folder that AI coding agents (Cursor, Claude Code, Codex) read directly to write new plotting code for you.
How do I actually use the figure-making skill?Simplest way is no-install: open the repo in your AI coding agent and tell it to follow scientific-figure-making/SKILL.md and references/design-theory.md while writing a plotting script, then run the script yourself. You can also symlink the skill folder into the agent's skills directory for permanent install.
Which coding agents does the skill support?Cursor, Claude Code, and Codex are explicitly documented, each with its own symlink command (e.g. ln -s into ~/.claude/skills/scientific-figure-making for Claude Code); any agent that can read files in the open repo can use the path-based mode too.
Can I just run these scripts on my own data out of the box?Not really — most scripts are project-specific examples. The intended flow is to have your AI agent write a new script for your own target folder following the skill's guidance, then you run it and check the exported figure yourself.
What kinds of figures can I make with this skill?It covers bar plots for quantitative comparison and composition breakdown, 3D sphere and radar plots, line plots, concept plots, and trend plots — the same figure types used in the author's own ICML, NeurIPS, and ECCV papers.
Why did the author build this as an AI-agent skill instead of just a script library?Chen Liu added the skill format on a friend's suggestion so coding agents like Cursor and Claude Code can read SKILL.md and generate new publication-ready figures on demand, instead of users having to copy and adapt raw scripts by hand.
Which papers have used figures from this workflow?Work including ImmunoStruct (Nature Machine Intelligence 2026), LM-Dispersion (ICML 2026), VIGIL (ECCV 2026), and RNAGenScape Brainteaser (NeurIPS 2025) are listed as related papers that used this figure-making approach.
What are similar projects?pwc, codex-claude-academic-skills, nature-skills, skills, Research-Paper-Writing-Skills, Auto-claude-code-research-in-sleep, Auto-Empirical-Research-Skills, claude-scholar
Which topics is it in?cvpr, matplotlib, academic, data-visualization, academic-research, skill, cursor, skills, agent-skills, codex, research, claude-code
Creator
ChenLiu-1996
Size
44.6 MB
Stars
8.7K
Updated
CommitOct 12
Age
1y
Commits · 90 Days