Metadata-Version: 2.5
Name: x-unet
Version: 0.5.0
Summary: X-Unet
Project-URL: Homepage, https://github.com/lucidrains/x-unet
Project-URL: Repository, https://github.com/lucidrains/x-unet
Author-email: Phil Wang <lucidrains@gmail.com>
License: MIT License
        
        Copyright (c) 2022 Phil Wang
        
        Permission is hereby granted, free of charge, to any person obtaining a copy
        of this software and associated documentation files (the "Software"), to deal
        in the Software without restriction, including without limitation the rights
        to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
        copies of the Software, and to permit persons to whom the Software is
        furnished to do so, subject to the following conditions:
        
        The above copyright notice and this permission notice shall be included in all
        copies or substantial portions of the Software.
        
        THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
        IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
        FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
        AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
        LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
        OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
        SOFTWARE.
License-File: LICENSE
Keywords: artificial intelligence,biomedical segmentation,deep learning,medical deep learning,unets
Classifier: Development Status :: 4 - Beta
Classifier: Intended Audience :: Developers
Classifier: License :: OSI Approved :: MIT License
Classifier: Programming Language :: Python :: 3
Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
Requires-Python: >=3.9
Requires-Dist: einops>=0.8.0
Requires-Dist: torch-einops-utils>=0.1.29
Requires-Dist: torch>=2.5
Provides-Extra: test
Requires-Dist: pytest; extra == 'test'
Description-Content-Type: text/markdown

<img src="./unet.png" width="450px"></img>

## x-unet

Implementation of a U-net complete with efficient attention as well as the latest research findings

## Install

```bash
$ pip install x-unet
```

## Usage

```python
import torch
from x_unet import XUnet

unet = XUnet(
    dim = 64,
    channels = 3,
    dim_mults = (1, 2, 4, 8),
    nested_unet_depths = (7, 4, 2, 1),     # nested unet depths, from unet-squared paper
    consolidate_upsample_fmaps = True,     # whether to consolidate outputs from all upsample blocks, used in unet-squared paper
    use_xattnres = True                    # enable cross-stage attention residuals
)

img = torch.randn(1, 3, 256, 256)
out = unet(img) # (1, 3, 256, 256)
```

You can also completely replace predetermined skip connections with learned cross-stage attention residuals:

```python
unet = XUnet(
    dim = 64,
    channels = 3,
    dim_mults = (1, 2, 4, 8),
    use_xattnres = True,
    xattnres_replace_skips = True          # remove skip connections and rely purely on learned attention residuals
)
```

For 3d (video or CT / MRI scans)

```python
import torch
from x_unet import XUnet

unet = XUnet(
    dim = 64,
    frame_kernel_size = 3,                 # set this to greater than 1
    channels = 3,
    dim_mults = (1, 2, 4, 8),
    nested_unet_depths = (5, 4, 2, 1),     # nested unet depths, from unet-squared paper
    consolidate_upsample_fmaps = True,     # whether to consolidate outputs from all upsample blocks, used in unet-squared paper
    weight_standardize = True,
    use_xattnres = True
)

video = torch.randn(1, 3, 10, 128, 128)    # (batch, channels, frames, height, width)
out = unet(video) # (1, 3, 10, 128, 128)
```

Or use `XAttnRes` standalone as a lightweight learned feature router across multi-scale stages:

```python
import torch
from x_unet import XAttnRes

xattnres = XAttnRes(
    dim = 64,
    dim_ins = (32, 64, 128),
    heads = 1
)

x = torch.randn(1, 64, 32, 32)
history = [
    torch.randn(1, 32, 64, 64),
    torch.randn(1, 64, 32, 32),
    torch.randn(1, 128, 16, 16)
]

out = xattnres(x, history) # (1, 64, 32, 32)
```

## Todo

- [ ] memory efficiency for 3d - reversible blocks, checkpointing, memory efficient unet
- [ ] offer option for axial convolutions (placing frame convolutions at end of the resnet chain)

## Citations

```bibtex
@article{Ronneberger2015UNetCN,
    title   = {U-Net: Convolutional Networks for Biomedical Image Segmentation},
    author  = {Olaf Ronneberger and Philipp Fischer and Thomas Brox},
    journal = {ArXiv},
    year    = {2015},
    volume  = {abs/1505.04597}
}
```

```bibtex
@article{Qin2020U2NetGD,
    title   = {U2-Net: Going Deeper with Nested U-Structure for Salient Object Detection},
    author  = {Xuebin Qin and Zichen Vincent Zhang and Chenyang Huang and Masood Dehghan and Osmar R Zaiane and Martin J{\"a}gersand},
    journal = {ArXiv},
    year    = {2020},
    volume  = {abs/2005.09007}
}
```

```bibtex
@inproceedings{Henry2020QueryKeyNF,
    title   = {Query-Key Normalization for Transformers},
    author  = {Alex Henry and Prudhvi Raj Dachapally and Shubham Vivek Pawar and Yuxuan Chen},
    booktitle = {FINDINGS},
    year    = {2020}
}
```

```bibtex
@article{Qiao2019WeightS,
    title   = {Weight Standardization},
    author  = {Siyuan Qiao and Huiyu Wang and Chenxi Liu and Wei Shen and Alan Loddon Yuille},
    journal = {ArXiv},
    year    = {2019},
    volume  = {abs/1903.10520}
}
```

```bibtex
@article{Shleifer2021NormFormerIT,
    title   = {NormFormer: Improved Transformer Pretraining with Extra Normalization},
    author  = {Sam Shleifer and Jason Weston and Myle Ott},
    journal = {ArXiv},
    year    = {2021},
    volume  = {abs/2110.09456}
}
```

```bibtex
@article{Sunkara2022NoMS,
    title   = {No More Strided Convolutions or Pooling: A New CNN Building Block for Low-Resolution Images and Small Objects},
    author  = {Raja Sunkara and Tie Luo},
    journal = {ArXiv},
    year    = {2022},
    volume  = {abs/2208.03641}
}
```

```bibtex
@inproceedings{Woo2023ConvNeXtVC,
    title   = {ConvNeXt V2: Co-designing and Scaling ConvNets with Masked Autoencoders},
    author  = {Sanghyun Woo and Shoubhik Debnath and Ronghang Hu and Xinlei Chen and Zhuang Liu and In-So Kweon and Saining Xie},
    year    = {2023}
}
```

```bibtex
@article{Liu2026XAttnRes,
    title   = {XAttnRes: Cross-Stage Attention Residuals for Medical Image Segmentation},
    author  = {Xinyu Liu and Qing Xu and Zhen Chen},
    journal = {arXiv preprint arXiv:2604.03297},
    year    = {2026}
}
```
