shirgur/umis

Official PyTorch implementation of "Unsupervised Microvascular Image Segmentation Using an Active Contours Mimicking Neural Network"

UMIS

Official PyTorch implementation of "Unsupervised Microvascular Image Segmentation Using an Active Contours Mimicking Neural Network" (link)

Prerequisites

• Python 3.6
• Pytorch +1.4
• Numpy
• Scipy
• OpenCV
• Path
• tqdm
• h5py
• tifffile
• libtiff

Morphological Pooling Layer

In order to build the Morphological Pooling layer on your own machine, run the following line

python src/setup.py install

Train

You can now train using the Euler-Lagrange (original paper), or the PDE (level-set) loss with additional regularization for stability.

python train_unsup.py --loss <EL/LS>

Citation

@inproceedings{gur2019unsupervised,
  title={Unsupervised Microvascular Image Segmentation Using an Active Contours Mimicking Neural Network},
  author={Gur, Shir and Wolf, Lior and Golgher, Lior and Blinder, Pablo},
  booktitle={Proceedings of the IEEE International Conference on Computer Vision},
  pages={10722--10731},
  year={2019}
}

umis's Issues

CUDA_HOME environment variable is not set

Dear @shirgur ,

Thanks for sharing the great work.

When I installed the morphologicalpool by python morphologicalpool/setup.py install, following error occurred:

Traceback (most recent call last):
  File "morphologicalpool/setup.py", line 9, in <module>
    'morphpool_cuda_kernel.cu',
  File "/home/mia/anaconda3/lib/python3.7/site-packages/torch/utils/cpp_extension.py", line 479, in CUDAExtension
    library_dirs += library_paths(cuda=True)
  File "/home/mia/anaconda3/lib/python3.7/site-packages/torch/utils/cpp_extension.py", line 558, in library_paths
    if (not os.path.exists(_join_cuda_home(lib_dir)) and
  File "/home/mia/anaconda3/lib/python3.7/site-packages/torch/utils/cpp_extension.py", line 1216, in _join_cuda_home
    raise EnvironmentError('CUDA_HOME environment variable is not set. '
OSError: CUDA_HOME environment variable is not set. Please set it to your CUDA install root.

My environment is ubuntu 18.04 + cuda 10.0.
I have installed the cuda and cudnn before installing the pytorch.

I can't find any useful solutions by google.

Could you share your insight about how to solve this problem?

Looking forward to your reply.
Best regards,
Jun

setup error

hi~great work.
When I try to compile, this occurs

How can I fix it?
And also could you tell me your cuda and pytorch version?
It will be very helpful.

Cannot get segmentation results

Dear Shigur,
Thank you for your good work.

I used the dataset VesselNN and followed the parameters setting of your code.
But I cannot get the results of segmentation.

I just got the results as following.

Inference results are not saved.

Dear @shirgur

It seems the inference results are not saved into local hard drives in train_unsup.py.

UMIS/train_unsup.py

Lines 226 to 235 in 1baf4b1

	output = output / idx_sum
	output = torch.Tensor(output).unsqueeze(0).cuda()
	input_gt = torch.Tensor(input_gt).unsqueeze(0)
	# Normalize
	output = norm_range(output)
	# Plot
	input = torch.Tensor(input).unsqueeze(0).unsqueeze(0)
	summary.visualize_image_val(writer, input, output, epoch)

Best,
Jun

License?

Hi

Very interesting work. Are you officially releasing under any specific license, or is this public domain?

Thanks
Phil

High level question: why AC loss can do better than Morph-ACWE?

Dear @shirgur ,

Again, I really enjoy reading this paper.

However, I still do not figure out: why AC loss can do better than Morph-ACWE?,

both the proposed Morph Layer with AC loss and Morph-ACWE are unsupervised methods, and Morph Layer with AC loss is a reformulation of Morph-ACWE.

The plain Morph-ACWE doesn't work well for vessel segmentation.

But the Morph-ACWE's reformulation (Morph Layer with AC loss) can work so well.

Could you share your insight on the reason?

Looking forward to your reply.

Best regards,
Jun

Where is the dataset?

Dear @shirgur ,

Thanks for sharing the code.

Where can I get the following dataset?

Best,
Jun