Resources

PyTorch package for reducing inter-experimental variability (batch effects) in aligned, fixed-length time-series data. Implements RAVE (unsupervised) and RAVE+ (semi-supervised) and includes a quickstart notebook on synthetic data. First release v0.1.0 (November 2026), installed from GitHub.

The research implementation used for the NeurIPS 2021 paper, with notebooks that reproduce its figures from the accompanying bipolar cell dataset. To apply RAVE or RAVE+ to new data, use Neuro-RAVE.

Open-source PyTorch package for training, evaluating and comparing deep learning models of the retina. Provides a shared core-and-readout architecture, a reproducible training pipeline, data loaders and pretrained models for public retina datasets from several species.

Loads the All-GCL dataset, reproduces the paper’s main figures, and includes a tutorial notebook for plotting response traces with their stimulus triggers.

Assigns mouse ganglion cell layer neurons to functional types (following Baden et al., 2016) from their chirp and moving-bar responses. The pretrained model downloads from Hugging Face on first use. Used for All-GCL and earlier Euler Lab studies.

A trained deep network (“digital twin”) that predicts responses of retinal ganglion cell axon terminals in the mouse superior colliculus, with analysis notebooks organized by figure. The Colab notebook runs the model in the browser without local setup.

Contrastive learning framework that builds positive pairs by averaging different subsets of repeated trials, producing 2D embeddings of multi-trial neural time series. A companion repository reproduces the paper’s figures.

Scripts and notebooks for system-identification models of retinal responses that add an efficient-coding objective based on natural scene statistics.

Analysis code for the two-photon Ca²⁺ imaging and multi-electrode array datasets, to reproduce the paper’s analyses and figure panels.

Jupyter notebooks that reproduce the analyses and figures on how photoreceptor degeneration in rd10 mice affects functional retinal ganglion cell types.

Tools for the two-photon functional recordings in the Eyewire II dataset (tutorial scripts, an interactive EM ↔ function explorer and 2P-to-EM registration), plus a separate repository that reproduces the resource paper’s figures.

Two-photon Ca²⁺ recordings from more than 80,000 neurons in the mouse ganglion cell layer, in NWB format, with standardized metadata and functional type assignments. Useful for population analyses, cell-type classification and model training.

Simulated and recorded responses of mouse retinal bipolar cells to the chirp stimulus, used to develop and evaluate RAVE and RAVE+.

Data accompanying the Eyewire II resource paper: preprocessed two-photon Ca²⁺ traces, 2P-to-EM mapping tables and reconstructed neuron skeletons from the mouse retina EM volume.

Preprocessed retina datasets from several labs in openretina’s format, plus pretrained model checkpoints. Includes the training data for the superior colliculus digital twin.

Two-photon Ca²⁺ imaging and multi-electrode array data underlying the study of nitric oxide effects on mouse retinal ganglion cells.

Functional recordings of mouse retinal ganglion cells underlying the study of how early photoreceptor degeneration in rd10 mice affects functional cell types.

Data accompanying Qiu et al. (2023) on hybrid system-identification models with an efficient-coding objective.