Metadata-Version: 2.1
Name: bism
Version: 0.0.5
Summary: Biomedical Image Segmentation Models (BISM)
Home-page: https://github.com/buswinka/bism
Author: Chris Buswinka
Author-email: buswinka@g.harvard.edu
Classifier: Programming Language :: Python :: 3
Requires-Python: >=3.10
Description-Content-Type: text/markdown
License-File: LICENSE
Requires-Dist: torch (>=2.0.0)
Requires-Dist: torchvision (>=0.13.0)
Requires-Dist: bism
Requires-Dist: waterz (>=0.9.4)
Requires-Dist: fastremap
Requires-Dist: zarr
Requires-Dist: tqdm
Requires-Dist: scikit-image
Requires-Dist: yacs
Requires-Dist: tensorboard

# bism - Biomedical Image Segmentation Models

This is a collection of generic PyTorch model constructors usefull for biomedical segmentation tasks. 
Something akin to the `timm` package for 2D image tasks. 
When at all possible, each model will offer a 2D or 3D implementation, however we will not provide pre-trained model files. 

Current Models
---------------

| Model          | 2D  | 3D  | Scriptable |
|----------------|-----|-----|------------|
| UNet           | ✓   | ✓   | ✓          |
| UNeXT          | ✓   | ✓   | ✓          |
| Recurrent UNet | ✓   | ✓   | ✓          |
| Residual UNet  |     |     |            |
| Unet++         | ✓   | ✓   | ✓          |
| CPnet          | ✓   | ✓   | ✓          |


Current Generic Blocks
----------------------

| BLOCK NAME           | 2D   | 3D  |
|----------------------|------|-----|
| UNeXT Block          | ✓    | ✓   |
| ConcatConv           | ✓    | ✓   |
| Recurrent UNet BLock | ✓    | ✓   |
| Residual UNet BLock  | ✓    | ✓   |
| DropPath             | ✓    | ✓   |
| LayerNorm            | ✓    | ✓   |
| UpSample             | ✓    | ✓   |
| ViT Block            |      |     |

Segmentation Implementation
----------------------

| APPROACH            | 2D   | 3D  |
|---------------------|------|-----|
| Cellpose            |     |   |
| Affinities          |     | ✓   |
| Local Shape Desc.   |     | ✓   |

