inlyze/deeplearning-segan-segmentation

This contains an implementation of the SeGAN model for semantic segmentation introduced in https://arxiv.org/pdf/1703.10239.pdf

DeepLearning SeGAN Segmentation

This contains an implementation of the SeGAN model for semantic segmentation introduced in https://arxiv.org/pdf/1706.01805.pdf

The model serves for semantic segmentation of image data and the authors have demonstrated its utility on cranial MRT images.

A summary of the model architecture from the paper is shown below

Dependencies

• Python 3.6
• Numpy
• Keras 2.0
• Tensorflow >= 1.x
• TQDM (optional)

This work was inspired by Xue et al. as well as the excellent "Deep Learning for coders" tought by Jeremy Howard and Rachel Thomas in their MOOC

deeplearning-segan-segmentation's Issues

Some questions about gt and cropinp

You have offered a whole implementation of SEGAN, and they seems very interesting, but without the processed data, I couldn't run the whole program.
So I wanna ask about some questions.
My input are (none, 572, 572, 3), and the label is (none, 358, 358, n_labels=8), as the origin Unet, what should the gt be? Do I need change the input into gray(channel=1), then multiply the input and label to get images masked by ground-truth(I think it's the gt)? And should I do the same thing with the output of the netS(the input has 3 channel, but the output has 8, I can't simply use the multiply layer)?
Btw, I found the old version without the crop operation.. It's easier to understand, but maybe it has a lower grades?

How to use these class for segmentation

Hello sir,
Can you please share the usage of given class for segmentation

model Compile error

in train
self.rl.append(self.model.layers[-1].train_on_batch(X, y)) # Train critic

File "envs\tensorflow\lib\site-packages\keras\engine\training.py", line 1209, in train_on_batch
class_weight=class_weight)

File "\envs\tensorflow\lib\site-packages\keras\engine\training.py", line 675, in _standardize_user_data
raise RuntimeError('You must compile a model before '

RuntimeError: You must compile a model before training/testing. Use model.compile(optimizer, loss).

No Dice loss or L1 loss included in your script

Hi,
I don't see the loss functions defined in your script as mentioned in this paper. Am I missing something,

Thank you.

Can u share a usage?

can u share a usage?

thanks

Is loss used in this repo same as in original paper?

I mean, in original paper, they write about mae: "hierarchical features extracted from image x by the critic network", and it seems that they calculate L1 as difference for all critic network layers. In your implementation, you use mae just for target.

Executing the code and trying to replicate

Hi,

I'm trying to replicate the results by executing your code. But, the result of segmentor network after 500 epochs is generated image of whole brain, rather only the tumor region of brain.

Can you share the results of the experiment?

Thanks!

Wrong link to the paper

Hi!

I think, there is a mistake in the README. You put the link to the paper SeGAN: Segmenting and Generating the Invisible, but this repo is actually the implementation of SegAN: Adversarial Network with Multi-scale L1 Loss for Medical
Image Segmentation.

How to train the SegAN?

Thank you very much for sharing SegAN. Currently, I apply the SegAN to segment the rib. If I use the Unet with the resnet as the backbone, the dice can achieve 0.85. But for SegAN, I can only obtain dice of 0.7. Do you have any suggestion?