Universal scale-free representations in human visual cortex

This repository contains code required to reproduce the experimental results published in PLOS Computational Biology and as a preprint on arXiv.

System requirements

Hardware

These analyses were run on a workstation with

• Processor: 13th Gen Intel(R) Core(TM) i9-13900K
• RAM: 128 GB
• GPU: NVIDIA GeForce RTX 4090 (24 GB VRAM)
• Storage: 2 TB NVMe SSD

If you plan to reproduce these results, I'd recommend using a machine with

• >= 64 GB RAM
• a GPU with >= 8 GB VRAM
• >= 1.5 TB of free disk space

Caution

RAM usage peaks when preparing the datasets for use the first time: we concatenate 40 sessions of fMRI data (!) for each subject in the Natural Scenes Dataset, which uses almost 64 GB of memory.

Tip

Throughout the code, there are batch_size parameters that control the amount of GPU memory used. You might want to adjust this especially if you run permutation tests or bootstrap resampling when computing covariance spectra.

Software

The code has been tested on RHEL 9.3 (Linux 5.14.0-503.23.1.el9_5.x86_64). Any standard Linux distribution should work. Other operating systems are not supported.

We use Python 3.13.7 for all analyses; the other required Python dependencies are specified in uv.lock. Do not attempt to install them directly; follow the installation guide below.

Installation (~5 min)

1. Clone this repository.

    git clone https://github.com/BonnerLab/scale-free-visual-cortex.git

2. Edit .env.

   • AWS_SHARED_CREDENTIALS_FILE should be the path to an AWS credentials file that gives you access to the Natural Scenes Dataset.
   • PROJECT_HOME controls where figures and results are saved. If you leave this unset, it will default to ~/scale-free-visual-cortex.
   • The other environment variables can be left unset: simply delete those lines.

3. Set up the Python environment.

   • Install the uv package manager.
   • Run uv sync --no-sources --extra {cpu, cu126, cu128, rocm} from $PROJECT_HOME to create the environment at $PROJECT_HOME/.venv.
     • If you have no GPU, use cpu.
     • If you have an Nvidia GPU, use cu126, cu128, or cu129 depending on your driver version.
     • If you have an AMD GPU, use rocm (note that this hasn't been tested).
   • Activate the environment (e.g. source $PROJECT_HOME/.venv/bin/activate if you're using bash)

Demo (~5 min)

We provide a simple high-level overview of the analysis in demo.ipynb using a small subset of the data. After installing this package, simply run the notebook file: this will automatically download ~300 MB of data and run the within- and between-subject analyses for one pair of subjects.

Reproducing the analyses

Important

You will need access to the Natural Scenes Dataset to reproduce the analyses in the paper. Specifically, you will need to obtain an AWS credentials file and set the AWS_SHARED_CREDENTIALS_FILE environment variable (see Step 2 of Installation).

manuscript/notebooks contains Jupyter notebooks that generate the figures in the paper, PDFs of which are located in manuscript/figures:

Figure	PDF	Notebook
1	schematic.pdf	schematic.ipynb
2	general.pdf	general-region.ipynb
3	significance-test.pdf	significance-test.ipynb
4	cross-correlations.pdf	cross-similarity.ipynb
5	visual-regions.pdf	visual-regions.ipynb
6	between-region-heatmaps.pdf	visual-regions.ipynb
7	rsa.pdf	rsa.ipynb
S1	vary-n-voxels.pdf	vary-n-voxels.ipynb
S2	vary-n-stimuli.pdf	vary-n-stimuli.ipynb
S3	cross-detectability.pdf	cross-similarity.ipynb
S4	gabor-model.pdf	gabor-filter-bank.ipynb
S5	between-region-spectra.pdf	visual-regions.ipynb
S6	general-all.pdf	general-region.ipynb
S7	cross-correlations-all.pdf	cross-similarity.ipynb
S8	singular-vectors.pdf	singular-vectors.ipynb
S9	dimensions-V1.pdf and dimensions-V1-synthetic.pdf	visualize-dimensions.ipynb
S10	dimensions-general.pdf and dimensions-general-synthetic.pdf	visualize-dimensions.ipynb
S11	things.pdf	things.ipynb
S12	between-monkey.pdf	tvsd.ipynb
S13	dimensions-general-semantic.pdf	visualize-dimensions.ipynb
S14	things-dimensions.pdf	things.ipynb

Warning

Running a notebook for the first time will likely take ages since the datasets used will be downloaded, processed and cached. Subsequent runs will be much faster.

The manuscript and responses to reviewers can be found in manuscript, with compiled PDFs (.pdf) and source files in Markdown (.qmd) and LaTeX (.tex) format.

Citation

If you find this work useful in your own research, please cite our manuscript published in PLOS Computational Biology.

A BibTeX entry is provided below for your convenience.

BibTeX

@article{Gauthaman2025,
  title = {Universal scale-free representations in human visual cortex},
  volume = {21},
  ISSN = {1553-7358},
  url = {http://dx.doi.org/10.1371/journal.pcbi.1013714},
  DOI = {10.1371/journal.pcbi.1013714},
  number = {11},
  journal = {PLOS Computational Biology},
  publisher = {Public Library of Science (PLoS)},
  author = {Gauthaman,  Raj Magesh and Ménard,  Brice and Bonner,  Michael F.},
  editor = {Christian Kietzmann,  Tim},
  year = {2025},
  month = nov,
  pages = {e1013714}
}

Contributing

If you run into any trouble, notice any bugs, or have any feedback, please create an issue in this repository or contact me by email at science@raj-magesh.org! Thanks for your interest in our work :)