Model Zoo
All models in Napari-OmniEM share the same core architecture: OmniEM.
OmniEM Architecture
OmniEM is a unified framework designed to support multiple electron microscopy (EM) tasks within a consistent architectural paradigm.
Model behavior is determined by:
- Task type
- Input dimensionality configuration
Task Types
Models are categorized by tasktype:
-
Image Restoration
(e.g., super-resolution, denoising) -
Image Segmentation
Dimensionality Support
OmniEM models can operate in either 2D or 3D mode.
2D Models
img_z = 1- Accept:
- 2D input
- 3D input (automatically reshaped into slice-wise 2D inference)
- 3D reshaping can be disabled by the user if inter-slice consistency is a concern.
3D Models
img_z > 1- Accept:
- 3D input only
- Require minimum Z-dimension ≥
img_z - No 2D inference allowed
Data Type Configuration
The datatypes field defines acceptable input dimensionality:
- 2D data
- 3D data
Available Models
Image Restoration
Image Segmentation
- Mitochondria Segmentation (2D & 3D variants)
Training and Fine-tuning OmniEM
If you would like to train or fine-tune OmniEM on your own dataset, please refer to the official training repository.
👉 The training repository will be publicly released soon.
Note: Models trained externally must export compatible weight files and configuration metadata to be imported into Napari-OmniEM.