shiba24 / 3d-unet Goto Github PK

comparing your model.py with tf_unet/unet.py, i find that you just changed the names of conv2d and deconv2d. And the sizes of convolution kernels are never changed. Accrording to the official apis, the filter size of conv3d must be 5-D.

LUNA16 is a better 3D dataset,the annotation xml files are pixel-wise,you can transform them into 3D npy files.

tensorflow/models.py script tries to import conv3d and deconv3d wrappers from tf_unet/layers that don't exist.

Hi - I am trying to understand the kernels used for the convolution here, its quite unclear from the paper and the source. Would greatly appreciate it if you could perhaps shed some light on this.

Thanks :)

Hi,
Curious about how you've modified either the layers from tf_unet or modified the input images.
tf.nn.conv3d (which i assume is what you substituted tf.nn.conv2d in layers.py for) only takes 5D tensors but all of the tensors I'm seeing in your code are 4D (batch, height, width, channel).

when I try to use the tensorflow version 3d-unet with model.py, there is an error occured that cannot import named "conv3d"

Just to let you know that the link to the paper is not correct.

Hi looking a script on how to feed data to this 3D Unet . Also the CT SLIVER dataset site are not allowing registration to obtain there data, can you please advise on an alternative or the exact data type

Thanks

I can't download the data of SLIVER07. This network needs to input a 3D data. What label file is needed?