Question

使用niftynet deepmedic分割图像尺寸和标签大小是相同的，但是空间窗口大小不是一样吧？
我尝试对其进行评估，并说"operands could not be broadcast together with shapes (101,101,77,1) (101,101,77,1,1)"，但是输出图像大小为（75,75,51,1,2），标签大小为（75,75,51 ）。我应该贴标签（75,75,51,1）然后撤离吗？

例如，使用niftynet deepmedic进行细分，训练10000次，原始adc /标签图像为（256,256,64），配置文件：

[ADC_IMAGE]
path_to_search = /data_PROCESS_3D_75_75_51/1ADC/
filename_contains = .nii.gz,.nii
spatial_window_size = (75, 75, 51)
interp_order = 1
[label]
path_to_search = /data_PROCESS_3D_75_75_51/3LABEL/
filename_contains =  .nii.gz,.nii
spatial_window_size = (27, 27, 3)

[NETWORK]
activation_function = prelu 
name = deepmedic 
batch_size = 2
volume_padding_size = 13
keep_prob=1     
queue_length = 50  
decay = 0 
reg_type = L2 
normalisation = False 
whitening = True
normalise_foreground_only=False
[TRAINING] 
sample_per_volume = 16 
OPTIMISER = adam
lr = 0.0005
loss_type = CrossEntropy
starting_iter = 0
save_every_n = 200 
max_iter = 100000
tensorboard_every_n = 100
max_checkpoints = 100
validation_every_n =100
validation_max_iter =25
exclude_fraction_for_validation = 0.2

[INFERENCE] 
border = (24,24,24)
save_seg_dir = /data_PROCESS_3D_75_75_51/2deepmedic_zmap/INFERENCE
inference_iter = 98200 
dataset_to_infer = validation
[EVALUATION]
save_csv_dir = /data_PROCESS_3D_75_75_51/2deepmedic_zmap/EVALUATION
evaluations = dice
evaluation_units = foreground

[SEGMENTATION]

image = ADC_IMAGE
label = label    
output_prob = True
num_classes = 2
label_normalisation = False

使用niftynet deepmedic，无法评估

0 个答案: