Skip to content
Snippets Groups Projects
Commit 9aaf495b authored by mmmmimic's avatar mmmmimic
Browse files

init

parent 697777bf
No related branches found
No related tags found
No related merge requests found
LICENSE 0 → 100644
MIT License
Copyright (c) 2019 Máté Kisantal
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.
# backboned-unet
U-Nets for image segmentation with pre-trained backbones in PyTorch.
## Why another U-Net implementation?
I was looking for an U-Net PyTorch implementation that can use pre-trained
torchvision models as backbones in the encoder path. There is a great
[repo](https://github.com/qubvel/segmentation_models)
for this in Keras, but I didn't find a good PyTorch implementation that works
with multiple torchvision models. So I decided to create one.
### WIP
## Getting started
So far VGG, ResNet and DenseNet backbones have been implemented.
To make it easy for you to get started with GitLab, here's a list of recommended next steps.
### Setup
Already a pro? Just edit this README.md and make it your own. Want to make it easy? [Use the template at the bottom](#editing-this-readme)!
Installing package:
## Add your files
git clone https://github.com/mkisantal/backboned-unet.git
cd backboned-unet
pip install .
- [ ] [Create](https://docs.gitlab.com/ee/user/project/repository/web_editor.html#create-a-file) or [upload](https://docs.gitlab.com/ee/user/project/repository/web_editor.html#upload-a-file) files
- [ ] [Add files using the command line](https://docs.gitlab.com/ee/gitlab-basics/add-file.html#add-a-file-using-the-command-line) or push an existing Git repository with the following command:
### Simple usage example
The U-net model can be imported just like any other torchvision model. The user can specify
a backbone architecture, choose upsampling operation (transposed convolution or bilinear upsampling
followed by convolution), specify the number of filters in the different decoder stages, etc.
```
cd existing_repo
git remote add origin https://lab.compute.dtu.dk/manli/backboned-unet.git
git branch -M main
git push -uf origin main
```
from backboned_unet import Unet
net = Unet(backbone_name='densenet121', classes=21)
## Integrate with your tools
The module loads the backbone torchvision model, and builds a decoder on top of it using specified
internal features of the backbone.
- [ ] [Set up project integrations](https://lab.compute.dtu.dk/manli/backboned-unet/-/settings/integrations)
## Collaborate with your team
### Results on Pascal VOC
- [ ] [Invite team members and collaborators](https://docs.gitlab.com/ee/user/project/members/)
- [ ] [Create a new merge request](https://docs.gitlab.com/ee/user/project/merge_requests/creating_merge_requests.html)
- [ ] [Automatically close issues from merge requests](https://docs.gitlab.com/ee/user/project/issues/managing_issues.html#closing-issues-automatically)
- [ ] [Enable merge request approvals](https://docs.gitlab.com/ee/user/project/merge_requests/approvals/)
- [ ] [Automatically merge when pipeline succeeds](https://docs.gitlab.com/ee/user/project/merge_requests/merge_when_pipeline_succeeds.html)
The figure below illustrates the training results for U-Nets with different ImageNet pretrained backbones.
## Test and Deploy
![pascal training](images/model_ious.png?raw=true "Pascal VOC training results")
Use the built-in continuous integration in GitLab.
Setup:
- [ ] [Get started with GitLab CI/CD](https://docs.gitlab.com/ee/ci/quick_start/index.html)
- [ ] [Analyze your code for known vulnerabilities with Static Application Security Testing(SAST)](https://docs.gitlab.com/ee/user/application_security/sast/)
- [ ] [Deploy to Kubernetes, Amazon EC2, or Amazon ECS using Auto Deploy](https://docs.gitlab.com/ee/topics/autodevops/requirements.html)
- [ ] [Use pull-based deployments for improved Kubernetes management](https://docs.gitlab.com/ee/user/clusters/agent/)
- [ ] [Set up protected environments](https://docs.gitlab.com/ee/ci/environments/protected_environments.html)
network = Unet(backbone_name=model_name, classes=21, decoder_filters=(512, 256, 128, 64, 32)
criterion = torch.nn.CrossEntropyLoss(ignore_index=255)
optimizer = torch.optim.Adam([{'params': network.get_pretrained_parameters(), 'lr':1e-5},
{'params': network.get_random_initialized_parameters()}], lr=1e-4)
***
Images were resized to 224x224. Batch size 32. No dataset augmentation. IoU is calculated with the background ignored.
# Editing this README
Example segmentation results:
When you're ready to make this README your own, just edit this file and use the handy template below (or feel free to structure it however you want - this is just a starting point!). Thank you to [makeareadme.com](https://www.makeareadme.com/) for this template.
## Suggestions for a good README
Every project is different, so consider which of these sections apply to yours. The sections used in the template are suggestions for most open source projects. Also keep in mind that while a README can be too long and detailed, too long is better than too short. If you think your README is too long, consider utilizing another form of documentation rather than cutting out information.
## Name
Choose a self-explaining name for your project.
## Description
Let people know what your project can do specifically. Provide context and add a link to any reference visitors might be unfamiliar with. A list of Features or a Background subsection can also be added here. If there are alternatives to your project, this is a good place to list differentiating factors.
## Badges
On some READMEs, you may see small images that convey metadata, such as whether or not all the tests are passing for the project. You can use Shields to add some to your README. Many services also have instructions for adding a badge.
## Visuals
Depending on what you are making, it can be a good idea to include screenshots or even a video (you'll frequently see GIFs rather than actual videos). Tools like ttygif can help, but check out Asciinema for a more sophisticated method.
## Installation
Within a particular ecosystem, there may be a common way of installing things, such as using Yarn, NuGet, or Homebrew. However, consider the possibility that whoever is reading your README is a novice and would like more guidance. Listing specific steps helps remove ambiguity and gets people to using your project as quickly as possible. If it only runs in a specific context like a particular programming language version or operating system or has dependencies that have to be installed manually, also add a Requirements subsection.
## Usage
Use examples liberally, and show the expected output if you can. It's helpful to have inline the smallest example of usage that you can demonstrate, while providing links to more sophisticated examples if they are too long to reasonably include in the README.
## Support
Tell people where they can go to for help. It can be any combination of an issue tracker, a chat room, an email address, etc.
## Roadmap
If you have ideas for releases in the future, it is a good idea to list them in the README.
## Contributing
State if you are open to contributions and what your requirements are for accepting them.
For people who want to make changes to your project, it's helpful to have some documentation on how to get started. Perhaps there is a script that they should run or some environment variables that they need to set. Make these steps explicit. These instructions could also be useful to your future self.
You can also document commands to lint the code or run tests. These steps help to ensure high code quality and reduce the likelihood that the changes inadvertently break something. Having instructions for running tests is especially helpful if it requires external setup, such as starting a Selenium server for testing in a browser.
## Authors and acknowledgment
Show your appreciation to those who have contributed to the project.
## License
For open source projects, say how it is licensed.
## Project status
If you have run out of energy or time for your project, put a note at the top of the README saying that development has slowed down or stopped completely. Someone may choose to fork your project or volunteer to step in as a maintainer or owner, allowing your project to keep going. You can also make an explicit request for maintainers.
![pascal results](images/examples.png?raw=true "Example segmentations from Pascal")
backboned-unet @ 8e60ce62
Subproject commit 8e60ce62481c56eacf4a1441c376bb580da41bc5
from .unet import Unet
from .utils import iou, soft_iou, dice_score, DiceLoss
\ No newline at end of file
import torch
import torch.nn as nn
from torchvision import models, datasets, transforms
from torch.nn import functional as F
def get_backbone(name, pretrained=True):
""" Loading backbone, defining names for skip-connections and encoder output. """
# TODO: More backbones
# loading backbone model
if name == 'resnet18':
backbone = models.resnet18(pretrained=pretrained)
elif name == 'resnet34':
backbone = models.resnet34(pretrained=pretrained)
elif name == 'resnet50':
backbone = models.resnet50(pretrained=pretrained)
elif name == 'resnet101':
backbone = models.resnet101(pretrained=pretrained)
elif name == 'resnet152':
backbone = models.resnet152(pretrained=pretrained)
elif name == 'vgg16':
backbone = models.vgg16_bn(pretrained=pretrained).features
elif name == 'vgg19':
backbone = models.vgg19_bn(pretrained=pretrained).features
# elif name == 'inception_v3':
# backbone = models.inception_v3(pretrained=pretrained, aux_logits=False)
elif name == 'densenet121':
backbone = models.densenet121(pretrained=True).features
elif name == 'densenet161':
backbone = models.densenet161(pretrained=True).features
elif name == 'densenet169':
backbone = models.densenet169(pretrained=True).features
elif name == 'densenet201':
backbone = models.densenet201(pretrained=True).features
elif name == 'unet_encoder':
from unet_backbone import UnetEncoder
backbone = UnetEncoder(3)
else:
raise NotImplemented('{} backbone model is not implemented so far.'.format(name))
# specifying skip feature and output names
if name.startswith('resnet'):
feature_names = [None, 'relu', 'layer1', 'layer2', 'layer3']
backbone_output = 'layer4'
elif name == 'vgg16':
# TODO: consider using a 'bridge' for VGG models, there is just a MaxPool between last skip and backbone output
feature_names = ['5', '12', '22', '32', '42']
backbone_output = '43'
elif name == 'vgg19':
feature_names = ['5', '12', '25', '38', '51']
backbone_output = '52'
# elif name == 'inception_v3':
# feature_names = [None, 'Mixed_5d', 'Mixed_6e']
# backbone_output = 'Mixed_7c'
elif name.startswith('densenet'):
feature_names = [None, 'relu0', 'denseblock1', 'denseblock2', 'denseblock3']
backbone_output = 'denseblock4'
elif name == 'unet_encoder':
feature_names = ['module1', 'module2', 'module3', 'module4']
backbone_output = 'module5'
else:
raise NotImplemented('{} backbone model is not implemented so far.'.format(name))
return backbone, feature_names, backbone_output
class UpsampleBlock(nn.Module):
# TODO: separate parametric and non-parametric classes?
# TODO: skip connection concatenated OR added
def __init__(self, ch_in, ch_out=None, skip_in=0, use_bn=True, parametric=False):
super(UpsampleBlock, self).__init__()
self.parametric = parametric
ch_out = ch_in/2 if ch_out is None else ch_out
# first convolution: either transposed conv, or conv following the skip connection
if parametric:
# versions: kernel=4 padding=1, kernel=2 padding=0
self.up = nn.ConvTranspose2d(in_channels=ch_in, out_channels=ch_out, kernel_size=(4, 4),
stride=2, padding=1, output_padding=0, bias=(not use_bn))
self.bn1 = nn.BatchNorm2d(ch_out) if use_bn else None
else:
self.up = None
ch_in = ch_in + skip_in
self.conv1 = nn.Conv2d(in_channels=ch_in, out_channels=ch_out, kernel_size=(3, 3),
stride=1, padding=1, bias=(not use_bn))
self.bn1 = nn.BatchNorm2d(ch_out) if use_bn else None
self.relu = nn.ReLU(inplace=True)
# second convolution
conv2_in = ch_out if not parametric else ch_out + skip_in
self.conv2 = nn.Conv2d(in_channels=conv2_in, out_channels=ch_out, kernel_size=(3, 3),
stride=1, padding=1, bias=(not use_bn))
self.bn2 = nn.BatchNorm2d(ch_out) if use_bn else None
def forward(self, x, skip_connection=None):
x = self.up(x) if self.parametric else F.interpolate(x, size=None, scale_factor=2, mode='bilinear',
align_corners=None)
if self.parametric:
x = self.bn1(x) if self.bn1 is not None else x
x = self.relu(x)
if skip_connection is not None:
x = torch.cat([x, skip_connection], dim=1)
if not self.parametric:
x = self.conv1(x)
x = self.bn1(x) if self.bn1 is not None else x
x = self.relu(x)
x = self.conv2(x)
x = self.bn2(x) if self.bn2 is not None else x
x = self.relu(x)
return x
class Unet(nn.Module):
""" U-Net (https://arxiv.org/pdf/1505.04597.pdf) implementation with pre-trained torchvision backbones."""
def __init__(self,
backbone_name='resnet50',
pretrained=True,
encoder_freeze=False,
classes=21,
decoder_filters=(256, 128, 64, 32, 16),
parametric_upsampling=True,
shortcut_features='default',
decoder_use_batchnorm=True):
super(Unet, self).__init__()
self.backbone_name = backbone_name
self.backbone, self.shortcut_features, self.bb_out_name = get_backbone(backbone_name, pretrained=pretrained)
shortcut_chs, bb_out_chs = self.infer_skip_channels()
if shortcut_features != 'default':
self.shortcut_features = shortcut_features
# build decoder part
self.upsample_blocks = nn.ModuleList()
decoder_filters = decoder_filters[:len(self.shortcut_features)] # avoiding having more blocks than skip connections
decoder_filters_in = [bb_out_chs] + list(decoder_filters[:-1])
num_blocks = len(self.shortcut_features)
for i, [filters_in, filters_out] in enumerate(zip(decoder_filters_in, decoder_filters)):
print('upsample_blocks[{}] in: {} out: {}'.format(i, filters_in, filters_out))
self.upsample_blocks.append(UpsampleBlock(filters_in, filters_out,
skip_in=shortcut_chs[num_blocks-i-1],
parametric=parametric_upsampling,
use_bn=decoder_use_batchnorm))
self.final_conv = nn.Conv2d(decoder_filters[-1], classes, kernel_size=(1, 1))
if encoder_freeze:
self.freeze_encoder()
self.replaced_conv1 = False # for accommodating inputs with different number of channels later
def freeze_encoder(self):
""" Freezing encoder parameters, the newly initialized decoder parameters are remaining trainable. """
for param in self.backbone.parameters():
param.requires_grad = False
def forward(self, *input):
""" Forward propagation in U-Net. """
x, features = self.forward_backbone(*input)
for skip_name, upsample_block in zip(self.shortcut_features[::-1], self.upsample_blocks):
skip_features = features[skip_name]
x = upsample_block(x, skip_features)
x = self.final_conv(x)
return x
def forward_backbone(self, x):
""" Forward propagation in backbone encoder network. """
features = {None: None} if None in self.shortcut_features else dict()
for name, child in self.backbone.named_children():
x = child(x)
if name in self.shortcut_features:
features[name] = x
if name == self.bb_out_name:
break
return x, features
def infer_skip_channels(self):
""" Getting the number of channels at skip connections and at the output of the encoder. """
x = torch.zeros(1, 3, 224, 224)
has_fullres_features = self.backbone_name.startswith('vgg') or self.backbone_name == 'unet_encoder'
channels = [] if has_fullres_features else [0] # only VGG has features at full resolution
# forward run in backbone to count channels (dirty solution but works for *any* Module)
for name, child in self.backbone.named_children():
x = child(x)
if name in self.shortcut_features:
channels.append(x.shape[1])
if name == self.bb_out_name:
out_channels = x.shape[1]
break
return channels, out_channels
def get_pretrained_parameters(self):
for name, param in self.backbone.named_parameters():
if not (self.replaced_conv1 and name == 'conv1.weight'):
yield param
def get_random_initialized_parameters(self):
pretrained_param_names = set()
for name, param in self.backbone.named_parameters():
if not (self.replaced_conv1 and name == 'conv1.weight'):
pretrained_param_names.add('backbone.{}'.format(name))
for name, param in self.named_parameters():
if name not in pretrained_param_names:
yield param
if __name__ == "__main__":
# simple test run
net = Unet(backbone_name='resnet18')
criterion = nn.MSELoss()
optimizer = torch.optim.Adam(net.parameters())
print('Network initialized. Running a test batch.')
for _ in range(1):
with torch.set_grad_enabled(True):
batch = torch.empty(1, 3, 224, 224).normal_()
targets = torch.empty(1, 21, 224, 224).normal_()
out = net(batch)
loss = criterion(out, targets)
loss.backward()
optimizer.step()
print(out.shape)
print('fasza.')
from torch import nn
class UnetDownModule(nn.Module):
""" U-Net downsampling block. """
def __init__(self, in_channels, out_channels, downsample=True):
super(UnetDownModule, self).__init__()
# layers: optional downsampling, 2 x (conv + bn + relu)
self.maxpool = nn.MaxPool2d((2,2)) if downsample else None
self.conv1 = nn.Conv2d(in_channels=in_channels, out_channels=out_channels,
kernel_size=3, padding=1)
self.bn1 = nn.BatchNorm2d(out_channels)
self.conv2 = nn.Conv2d(in_channels=out_channels, out_channels=out_channels,
kernel_size=3, padding=1)
self.bn2 = nn.BatchNorm2d(out_channels)
self.relu = nn.ReLU(inplace=True)
def forward(self, x):
if self.maxpool is not None:
x = self.maxpool(x)
x = self.conv1(x)
x = self.bn1(x)
x = self.relu(x)
x = self.conv2(x)
x = self.bn2(x)
x = self.relu(x)
return x
class UnetEncoder(nn.Module):
""" U-Net encoder. https://arxiv.org/pdf/1505.04597.pdf """
def __init__(self, num_channels):
super(UnetEncoder, self,).__init__()
self.module1 = UnetDownModule(num_channels, 64, downsample=False)
self.module2 = UnetDownModule(64, 128)
self.module3 = UnetDownModule(128, 256)
self.module4 = UnetDownModule(256, 512)
self.module5 = UnetDownModule(512, 1024)
def forward(self, x):
x = self.module1(x)
x = self.module2(x)
x = self.module3(x)
x = self.module4(x)
x = self.module5(x)
return x
import torch
def iou(predictions, labels, threshold=None, average=True, device=torch.device("cpu"), classes=21,
ignore_index=255, ignore_background=True):
""" Calculating Intersection over Union score for semantic segmentation. """
gt = labels.long().unsqueeze(1).to(device)
# getting mask for valid pixels, then converting "void class" to background
valid = gt != ignore_index
gt[gt == ignore_index] = 0
# converting to onehot image whith class channels
onehot_gt_tensor = torch.LongTensor(gt.shape[0], classes, gt.shape[-2], gt.shape[-1]).zero_().to(device)
onehot_gt_tensor.scatter_(1, gt, 1) # write ones along "channel" dimension
classes_in_image = onehot_gt_tensor.sum([2, 3]) > 0
# check if it's only background
if ignore_background:
only_bg = (classes_in_image[:, 1:].sum(dim=1) == 0).sum() > 0
if only_bg:
raise ValueError('Image only contains background. Since background is set to ' +
'ignored, IoU is invalid.')
if threshold is None:
# taking the argmax along channels
pred = torch.argmax(predictions, dim=1).unsqueeze(1)
pred_tensor = torch.LongTensor(pred.shape[0], classes, pred.shape[-2], pred.shape[-1]).zero_().to(device)
pred_tensor.scatter_(1, pred, 1)
else:
# counting predictions above threshold
pred_tensor = (predictions > threshold).long()
onehot_gt_tensor *= valid.long()
pred_tensor *= valid.long().to(device)
intersection = (pred_tensor & onehot_gt_tensor).sum([2, 3]).float()
union = (pred_tensor | onehot_gt_tensor).sum([2, 3]).float()
iou = intersection / (union + 1e-12)
start_id = 1 if ignore_background else 0
if average:
average_iou = iou[:, start_id:].sum(dim=1) /\
(classes_in_image[:, start_id:].sum(dim=1)).float() # discard background IoU
iou = average_iou
return iou.cpu().numpy()
def soft_iou(pred_tensor, labels, average=True, device=torch.device("cpu"), classes=21,
ignore_index=255, ignore_background=True):
""" Soft IoU score for semantic segmentation, based on 10.1109/ICCV.2017.372 """
gt = labels.long().unsqueeze(1).to(device)
# getting mask for valid pixels, then converting "void class" to background
valid = gt != ignore_index
gt[gt == ignore_index] = 0
valid = valid.float().to(device)
# converting to onehot image with class channels
onehot_gt_tensor = torch.LongTensor(gt.shape[0], classes, gt.shape[-2], gt.shape[-1]).zero_().to(device)
onehot_gt_tensor.scatter_(1, gt, 1) # write ones along "channel" dimension
classes_in_image = onehot_gt_tensor.sum([2, 3]) > 0
onehot_gt_tensor = onehot_gt_tensor.float().to(device)
# check if it's only background
if ignore_background:
only_bg = (classes_in_image[:, 1:].sum(dim=1) == 0).sum() > 0
if only_bg:
raise ValueError('Image only contains background. Since background is set to ' +
'ignored, IoU is invalid.')
onehot_gt_tensor *= valid
pred_tensor *= valid
# intersection = torch.max(torch.tensor(0).float(), torch.min(pred_tensor, onehot_gt_tensor).sum(dim=[2, 3]))
# union = torch.min(torch.tensor(1).float(), torch.max(pred_tensor, onehot_gt_tensor).sum(dim=[2, 3]))
intersection = (pred_tensor * onehot_gt_tensor).sum(dim=[2, 3])
union = (pred_tensor + onehot_gt_tensor).sum(dim=[2, 3]) - intersection
iou = intersection / (union + 1e-12)
start_id = 1 if ignore_background else 0
if average:
average_iou = iou[:, start_id:].sum(dim=1) /\
(classes_in_image[:, start_id:].sum(dim=1)).float() # discard background IoU
iou = average_iou
return iou.cpu().numpy()
def dice_score(input, target, classes, ignore_index=-100):
""" Functional dice score calculation. """
target = target.long().unsqueeze(1)
# getting mask for valid pixels, then converting "void class" to background
valid = target != ignore_index
target[target == ignore_index] = 0
valid = valid.float()
# converting to onehot image with class channels
onehot_target = torch.LongTensor(target.shape[0], classes, target.shape[-2], target.shape[-1]).zero_().cuda()
onehot_target.scatter_(1, target, 1) # write ones along "channel" dimension
# classes_in_image = onehot_gt_tensor.sum([2, 3]) > 0
onehot_target = onehot_target.float()
# keeping the valid pixels only
onehot_target = onehot_target * valid
input = input * valid
dice = 2 * (input * onehot_target).sum([2, 3]) / ((input**2).sum([2, 3]) + (onehot_target**2).sum([2, 3]))
return dice.mean(dim=1)
class DiceLoss(torch.nn.Module):
""" Dice score implemented as a nn.Module. """
def __init__(self, classes, loss_mode='negative_log', ignore_index=255, activation=None):
super(DiceLoss, self).__init__()
self.classes = classes
self.ignore_index = ignore_index
self.loss_mode = loss_mode
self.activation = activation
def forward(self, input, target):
if self.activation is not None:
input = self.activation(input)
score = dice_score(input, target, self.classes, self.ignore_index)
if self.loss_mode == 'negative_log':
eps = 1e-12
return (-(score+eps).log()).mean()
elif self.loss_mode == 'one_minus':
return (1 - score).mean()
else:
raise ValueError('Loss mode unknown. Please use \'negative_log\' or \'one_minus\'!')
if __name__ == "__main__":
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
predictions = torch.softmax(torch.empty(7, 21, 224, 224).normal_(), dim=1)
conv = torch.nn.Conv2d(21, 21, 1, 1, 0)
labels = (torch.empty(7, 224, 224).normal_(10, 10) % 21)
criterion = DiceLoss(21, activation=torch.nn.Softmax(dim=1))
loss = criterion(conv(predictions), labels)
loss.backward()
print(loss)
print('done.')
images/examples.png

1.38 MiB

images/model_ious.png

69.3 KiB

setup.py 0 → 100644
from setuptools import setup
setup(name='backboned_unet',
version='0.0.1',
description='U-Net built with TorchVision backbones.',
url='https://github.com/mkisantal/backboned-unet',
keywords='machine deep learning neural networks pytorch torchvision segmentation unet',
author='mate Kisantal',
author_email='kisantal.mate@gmail.com',
license='MIT',
packages=['backboned_unet'],
install_requires=[
'torch',
'torchvision'
],
zip_safe=False)
0% Loading or .
You are about to add 0 people to the discussion. Proceed with caution.
Please register or to comment