unknown error

[zswgzx]

I'm trying to perform inference using the pre-trained model, but have the following error.

configuration options:

modalities : ['FLAIR', 'T1']
fract_negative_positive : 2.0
test_folder : /path/to/testing
weight_paths : None
load_weights : True
t_bin : 0.5
min_error : 0.5
num_layers : None
max_epochs : 400
pretrained_model : baseline_2ch
image_tags : ['flair', 't1']
gpu_number : 2
denoise_iter : 3
roi_tags : ['lesion']
register_modalities : True
save_tmp : True
denoise : True
output_folder : /output
experiment : baseline_2ch
mod3_tags : ['None']
t1_tags : ['t1']
batch_size : 128
x_names : ['FLAIR_brain.nii.gz', 'T1_brain.nii.gz']
current_scan : scan
full_train : False
train_folder : /path/to/training
mod4_tags : ['None']
skull_stripping : True
net_verbose : 1
patience : 50
flair_tags : ['flair']
patch_size : (11, 11, 11)
balanced_training : False
l_min : 10
out_name : out_seg.nii.gz
min_th : 0.5
debug : True
randomize_train : True
train_split : 0.25
pretrained : None
fully_convolutional : False

DEBUG: 2
Using TensorFlow backend.
Traceback (most recent call last):
File "nic_infer_segmentation_batch.py", line 97, in
model = cascade_model(options)
File "/home/docker/src/libs/CNN/build_model.py", line 95, in cascade_model
model = get_network(options)
File "/home/docker/src/libs/CNN/nets.py", line 24, in get_network
net_input = Input(name='in1', shape=(channels,) + options['patch_size'])
File "/miniconda/lib/python2.7/site-packages/keras/engine/topology.py", line 1375, in Input
input_tensor=tensor)
File "/miniconda/lib/python2.7/site-packages/keras/engine/topology.py", line 1286, in init
name=self.name)
File "/miniconda/lib/python2.7/site-packages/keras/backend/tensorflow_backend.py", line 349, in placeholder
x = tf.placeholder(dtype, shape=shape, name=name)
AttributeError: 'module' object has no attribute 'placeholder'

docker@19764b608857:~/src$ ls
init.py config nets utils
app.py libs nic_infer_segmentation_batch.py
cnn_scripts.py logonic.png nic_train_network_batch.py

GPU is completely new to me, so any help would be appreciated.

[Sergi Valverde]

Hi, If you are using the docker, then you don't have to install anything apart from docker and be sure that the machine running the code has the proper nvidia-drivers and cuda. I would encourage to use this version of nicmslesions:

Then go to nicmslesions and follow the instructions. It's basically using a docker file from the Docker hub and running everything from a shell script. We have this version running in several machines without problems.

Let met know,
s

[zswgzx]

It's run on NVIDIA Corporation GK210GL [Tesla K80] (rev a1). I installed nvidia driver 384 and cuda-9-0 (https://www.tensorflow.org/install/gpu#ubuntu_1604_cuda_90_for_tensorflow_1130, checked with nvidia-smi). libcudnn7=7.0.3.11-1+cuda9.0. cuda-repo-ubuntu1604_9.0.176-1_amd64.deb.

If Docker image is NOT used, try to install the required python packages first (keras instroduced tensorflow/tensorboard2.0.0, and should be uninstalled), then install tensorflow-gpu.

Everything looks good so far. Thanks again Sergi.