shreyaspadhy/unet-zoo

A collection of UNet and hybrid architectures in PyTorch for 2D and 3D Biomedical Image segmentation

unet-zoo's Issues

Question about prediction output

Hi,

Is there a way to quickly output prediction segmentation pngs?

Thank you!

How to realize BDC-LSTM in Keras?

Inference using UNET + BD-LSTM?

I used main_bdclstm.py to train the network.
To do the inference using the above network, we need to use one image or 3 images. Kindly provide inputs.

Question about BDC-LSTM input.

Hi, I read the article of BDC-LSTM, it seems the inputs of it are just the mask resluts of U-net segmentation, is it right?

Which BraTS set was used for training?

2012, 2015, 2017 or 2018?
Thanks,

kU-NET implementation

Hey,

really nice collection of UNets! I was wondering whether you put the "kU-Net" on the line already. I've seen that you wrote: "kUNet - Combining multiple UNets for increasing heirarchial preservation of information (coming soon) [reference]"

Would be nice to have a working version for rapid prototyping. I'm a bit affraid of implementing it as I'm not the expert in NNs yet.

Thank you very much for your help!
Best
beniroquai

about the database

@shreyaspadhy I am using brats 2017 database which is in .nii format. It is necessary to convert that data into .png format?

Missing 'losses' module

The module contain the lost function appears to be missing
Stacktrace after running main.py:

Traceback (most recent call last):
File "main.py", line 22, in
from losses import DICELossMultiClass
ModuleNotFoundError: No module named 'losses'

Hi, just a quick thought: difference comparision to 3d-unet

I just wonder, all these nets, like what we have in 2d tasks: yolo, faster, cascade, they could by summarized as one-stage or two-stage.
When it comes to 3d, U-net is now also available to 3d data, so except that U-net more often used to work with medical data, what else (and why) that are different? in design ideas or senses of intuition?

No module named dataParser

Seems like when I attempt to run main.py, I get a 'No module named dataParser.' Might I be doing something wrong?

A question about the DICELossMultiClass

Can you make a explanation about the implementation of DICELossMultiClass:
why use output[:, 1, :, :]?

Questions about BraTS Masks

Do you have to convert the ground truth masks to a binary segmentation map? The original BraTS masks are multi-class and it seems like the network only has two classes.

Dice Score

Just wanted to make sure I'm looking at this correctly: the Average Dice Coefficient which is printed out after you run test should be different then the DICE score which is shown on the 'dice.png'

Loading BRATS dataset and # of training epochs

I managed to get access of the BRATS dataset, but the scans are in nii format. I managed to use nibabel to load them into a 3d tensors then stacked the different channels. How did you get them in png format? And how did you convert the 3d scans to 2d slices? did you just iterate over the depth of the scan? If so, how did you deal with slices that contain only background?
Also, how many epochs did you need to get these dice scores?

how to prepare data for bdclstm.py

Hi, how to prepare my data for training using bdclstm.py?

Thank you!

3D Unet unable to locate

Could you please point out the 3D Unet code in the repository? I could only find 2D Unet implementation.

Thanks