fkong7 / heartffdnet Goto Github PK

As I've just spent some time doing this, I thought i'd add some instructions to help others.

A tf gpu ready implementation of nearest neighbours needs to be compiled before loss.py will work out of the box. I've found this can be painful as it is dependent on cuda / tensorflow versions, however these were my steps to create the tf_nndistance_so.so file.

environment: Tensorflow 1.14, Cuda 10.0

1.Download all the tf_nndistance* files from this repo and store all files in same dir
2. cd into your dir and execute sh tf_nndistance_compile.sh
3. If you're lucky this will work as is, however its likely you will have to edit tf_nndistance_compile.sh. I had to edit the usr/local/cuda dir to ensure that it was pointing at the correct cuda dir. There was also an error telling me that the ltensorflow_framework could not be found. For this error I first found the dir containing my tensorflow libraries using python -c 'import tensorflow as tf; print(tf.sysconfig.get_lib())', then cd into that dir and check that I have a libtensorflow_framework.so file. I didnt, but I did have a libtensorflow_framework.so.1 file, so I created a symbolic link as suggested here with the command ln -s libtensorflow_framework.so.1 libtensorflow_framework.so. After this I was able to compile with no problems.
4. You can test using the nn_distance.py.
5. Finally change the dir link in loss.py to your tf_nndistance_so.so file.

When creating a prediction using the MMWHS dataset, can you tell me where the dat_of_template_with_no_veins.dat file comes from? Assume it would be created from template_no_views_normalized.vtp using make_control_grid.py, however i'm getting vtkXMLPolyDataReader errors when executing the following command:

python template/make_control_grid.py --template examples/mmwhs_test/template_with_no_veins_normalized.vtp --template_im examples/place_holder_image.nii.gz --num_pts 6

Any ideas where I'm going wrong?

P.s the weblink to the MMWHS does not work. I think this is the updated one: https://zmiclab.github.io/zxh/0/mmwhs/

full error:
` ERROR: In ../Common/ExecutionModel/vtkExecutive.cxx, line 784
vtkCompositeDataPipeline (0x555a68df0770): Algorithm vtkXMLPolyDataReader(0x555a68ef3250) returned failure for request: vtkInformation (0x555a68f86840)
Debug: Off
Modified Time: 131
Reference Count: 1
Registered Events: (none)
Request: REQUEST_INFORMATION
FORWARD_DIRECTION: 0
ALGORITHM_AFTER_FORWARD: 1

Traceback (most recent call last):
File "template/make_control_grid.py", line 853, in
template = process_template(args.template, args.target_node_num, template_im_fn=args.template_im, ref_template_fn=args.ref_template)
File "template/make_control_grid.py", line 251, in process_template
template = transform_polydata(template, img_center2-img_center, transform, SIZE)
File "template/make_control_grid.py", line 47, in transform_polydata
coords = map_polydata_coords(poly, displacement, transform, size)
File "template/make_control_grid.py", line 39, in map_polydata_coords
coords = vtk_to_numpy(poly.GetPoints().GetData())
AttributeError: 'NoneType' object has no attribute 'GetData' `

Hi, when i'm running your predict.py out the box, i have an error where the dataloader returns 0 testing volumes and
0 mask volumes. When I dug a little deeper I found that the dataloader was trying to retrieve *.nii.gz, *.vti or *.nii files from inside examples/ct_test directory. However, in the google drive link you provided there are not any of those files types in the ct_test directory.

Is the program supposed to look for files in this directory? Perhaps you can tell me what you think is going wrong

P.s great work on the paper!

Hello,

I would like to use your pre-trained model, but unfortunately I bumped into errors because of the Tensorflow version. So one of your requirements was to use Tensorflow 1.14, which seems to be no longer an option in environments like Google Colab (where I currently work).
Have you updated your code to be compatible with Tensorflow 2?
Or do you have any suggestions to make it compatible with newer Tensorflow versions?

Cheers,
Nika