Net2Brain: toolbox for comparing artificial vision models with brain responses

• Net2Brain is a toolbox to compare artificial vision models with human brain responses (Frontiers, 2025).¹
• The open tooling supports computational neuroscience, neuroinformatics, and vision–brain alignment for model benchmarking and brain-inspired design.^{1 1}

Weekly enrichment (2026-07-20)

• Net2Brain was published in Frontiers in Neuroinformatics (2025; DOI 10.3389/fninf.2025.1515873) as an open-source, Python-based toolbox providing an end-to-end pipeline for incorporating deep neural networks (DNNs) into neuroscience research.^{2 3}
• The toolbox provides access to over 600 pre-trained DNNs spanning multiple modalities (vision, language, audio, and multimodal data), organized through a structured taxonomy to help researchers select appropriate models.^{2 4}
• The pipeline is organized into six components: a taxonomy module, a dataset selection module, a feature extraction module, an RDM (representational dissimilarity matrix) creation module, an evaluation module, and a plotting module.^{2 3}
• It supplies a streamlined API for downloading and handling popular neuroscience datasets, including the Natural Scenes Dataset (NSD) and the THINGS dataset, and supports fMRI, EEG, and MEG brain recordings.^{2 3}
• Evaluation methods include standard and weighted representational similarity analysis (RSA), linear encoding, variance partitioning, searchlight analysis, centered kernel alignment (CKA), and stacked encoding; the default RSA uses Pearson correlation distance.^{2 4}
• Feature extraction is handled by the FeatureExtractor class, which accepts pre-trained architectures, random-weight models, and user-supplied custom models across image, video, and text inputs, with optional sparse random projection (SRP) dimensionality reduction (default 50 components).^{4 5}
• Net2Brain is a collaboration between the CVAI group and Radoslaw Cichy’s lab, and builds on the goals of the Algonauts project (Cichy, Roig, Andonian et al., 2019).^{4 5}
• The toolbox offers both a graphical user interface (GUI) and a command-line interface (CLI) suited to server environments such as Google Colab, and produces publication-ready bar and line plots with noise ceilings and statistical significance.^{2 5}

Footnotes

1. https://news.google.com/rss/articles/CBMimgFBVV95cUxOWG91b2FWLTRwbnZBMHNxbTZmUVZTX1dnN1pWaWoxcDVsMnFjal9SX3dzYjRZMVZvRDRTdWs0cnNXbTQ4VVNaQm5lSnhzX3ZPYUlWYnRFU2o3VWpsWjBjN1Y4Z1JjTmxFUW1lNGg1X3hWb1dmWkphU09NbXdSUS1JaThzbjNVT3RmbXhYMnI0UENMWS13OWVVeE53?oc=5 ↩ ↩² ↩³

2. https://www.frontiersin.org/journals/neuroinformatics/articles/10.3389/fninf.2025.1515873/full ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶

3. https://pmc.ncbi.nlm.nih.gov/articles/PMC12089098/ ↩ ↩² ↩³

4. https://github.com/cvai-roig-lab/Net2Brain ↩ ↩² ↩³ ↩⁴

5. https://arxiv.org/html/2208.09677v4 ↩ ↩² ↩³