---
library_name: EMber
tags:
- cryo-em
- image-feature-extraction
- model_hub_mixin
- pytorch_model_hub_mixin
---

<h1 align="center"><em>EMber</em>: Learning the Shared Latent Structure of Cryo-EM Inverse Problems</h1>

<!-- Replace each href="#" with the author's website URL. -->
<p align="center">
  <a href="https://nithinvc.github.io/">Nithin Chalapathi</a> ·
  <a href="https://www.theosternlieb.com/">Theo Sternlieb</a> ·
  <a href="#">Ethan Han</a> ·
  <a href="https://sonyahanson.github.io/">Sonya M. Hanson</a> ·
  <a href="https://pilarcossio.org/">Pilar Cossio</a> ·
  <a href="https://a1k12.github.io/">Aditi S. Krishnapriyan</a> ·
  <a href="https://ritwikgupta.me/">Ritwik Gupta</a>
</p>

EMber is a family of pretrained vision transformers for extracting representations from single-particle cryo-EM particle images. Given a particle image, EMber produces an image-level embedding and spatial patch features for use in downstream analysis. The models were trained using self-distillation on over 50 million experimental particle images from 214 EMPIAR datasets, together with simulated projections from 268,000 molecular assemblies.

Check out our preprint [here](https://arxiv.org/abs/2610.22523) and code [here](https://github.com/nithinvc/EMber).

EMber-Large has 304.2M parameters and uses a 16x16 patch size.