def loader(img_path, mask_path): image, gt = utils.load_vol_brats(img_path, slicen=nslice) return image, gt
def loader(img_path, mask_path):
image, gt = utils.load_vol_brats(img_path, slicen=nslice)
return image[:, :, seq_map[seq]][:, :, None], gt
infoclasses['ET'] = (3,)
infoclasses['CT'] = (1,3)
for layer_name in layer_names:
try:
dissector = spatial.Dissector(model=model,
layer_name = layer_name, seq='all')
except:
continue
threshold_maps = dissector.get_threshold_maps(dataset_path = data_root_path,
save_path = 'results_ET/Dissection/simnet/threshold_maps/',
percentile = 85,
loader=dataloader())
image, gt = utils.load_vol_brats('../sample_vol/brats/Brats18_CBICA_AOP_1', slicen=105)
maks_path = '../sample_vol/brats/Brats18_CBICA_AOP_1/mask.nii.gz'
ROI = sitk.GetArrayFromImage(sitk.ReadImage(maks_path))[105, :, :]
dissector.apply_threshold(image, threshold_maps,
nfeatures=25,
save_path='results_ET/Dissection/simnet/feature_maps/',
ROI = ROI)
dissector.quantify_gt_features(image, gt,
threshold_maps,
nclasses=infoclasses,
nfeatures=None,
save_path='results_ET/Dissection/simnet/csv/',
save_fmaps=False,
ROI = ROI)
custom_objects={
'gen_dice_loss': gen_dice_loss,
'dice_whole_metric': dice_whole_metric,
'dice_core_metric': dice_core_metric,
'dice_en_metric': dice_en_metric
})
model.load_weights(
'/home/pi/Projects/BioExp/trained_models/densedrop/model_lrsch.hdf5',
by_name=True)
if __name__ == '__main__':
list_ = []
for volume in [32, 20, 24, 53, 12, 14]:
for slice_ in range(20, 140, 5):
test_image, gt = load_vol_brats(test_path[volume], slice_, pad=0)
D = uncertainty(test_image)
# for aleatoric
mean, var = D.aleatoric(model, iterations=50)
# for epistemic
mean, var = D.epistemic(model, iterations=50)
# for combined
mean, var = D.combined(model, iterations=50)
print(np.mean(var, axis=(0, 1, 2)))
list_.append(np.mean(var, axis=(0, 1, 2)))
custom_objects={
'gen_dice_loss': gen_dice_loss,
'dice_whole_metric': dice_whole_metric,
'dice_core_metric': dice_core_metric,
'dice_en_metric': dice_en_metric
})
weights_path = '/media/balaji/CamelyonProject/parth/checkpoints/double_headed_autoencoder/autoencoder_double_headed_no_skip_weights.hdf5'
model.save_weights(weights_path)
for layer_name in layers_to_consider:
# for i, file in enumerate(glob(data_root_path +'*')[5:5 +num_images]):
model.load_weights(weights_path)
image, gt = utils.load_vol_brats(
'/media/balaji/CamelyonProject/parth/brats_2018/val/Brats18_2013_3_1',
slicen=78)
A = ablation.Ablate(model,
weights_path,
metric,
layer_name,
image[None, ...],
gt,
classes=infoclasses,
image_name=str(i))
path = 'results/'
os.makedirs(path, exist_ok=True)
df1 = A.ablate_filters(save_path='./results/', step=2)
# if i == 0:
data_root_path = '../sample_vol/'
model_path = '../trained_models/U_resnet/ResUnet.h5'
weights_path = '../trained_models/U_resnet/ResUnet.40_0.559.hdf5'
model = load_model(model_path, custom_objects={'gen_dice_loss':gen_dice_loss,
'dice_whole_metric':dice_whole_metric,
'dice_core_metric':dice_core_metric,
'dice_en_metric':dice_en_metric})
model.load_weights(weights_path)
infoclasses = {}
for i in range(1): infoclasses['class_'+str(i)] = (i,)
infoclasses['whole'] = (1,2,3)
data_root_path = '../sample_vol/'
layer_name = 'conv2d_3'
test_image, gt = utils.load_vol_brats('../sample_vol/Brats18_CBICA_ARZ_1', slicen=78)
A = spatial.Ablation(model = moedl,
weights_pth = weights_path,
metric = dice_label_coef,
layer_name = layer_name,
test_image = test_image,
gt = gt,
classes = infoclasses,
nclasses = 4)
df = A.ablate_filter(step = 1)
'gen_dice_loss': gen_dice_loss,
'dice_whole_metric': dice_whole_metric,
'dice_core_metric': dice_core_metric,
'dice_en_metric': dice_en_metric
})
model_no_drop.load_weights(
'/home/parth/Interpretable_ML/saved_models/densenet/densenet.55_0.522.hdf5',
by_name=True)
if __name__ == '__main__':
list_ = []
for volume in [10]:
for slice_ in range(78, 79):
test_image, gt = load_vol_brats(glob(test_path)[volume],
slice_,
pad=8)
D = uncertainty(test_image)
# for aleatoric
mean, var = D.aleatoric(model_no_drop, iterations=50)
D.save(mean, var, gt)
# for epistemic
mean, var = D.epistemic(model, iterations=50)
D.save(mean, var, gt)
# # for combined
for i in range(4): infoclasses['class_'+str(i)] = (i,)
infoclasses['whole'] = (1,2,3,)
infoclasses['ET'] = (3,)
infoclasses['CT'] = (1,3,)
num_images = 5
model = load_model(model_path, compile=False, custom_objects={'gen_dice_loss':gen_dice_loss,
'dice_whole_metric':dice_whole_metric,
'dice_core_metric':dice_core_metric,
'dice_en_metric':dice_en_metric})
for layer_name in layers_to_consider:
for i, file in tqdm(enumerate(glob(data_root_path +'*')[5 : 5 + num_images])):
model.load_weights(weights_path)
image, gt = utils.load_vol_brats(file, slicen=78)
image = image[None, ...]
A = ablation.Ablate(model, weights_path, dice_label_coef, layer_name, image, gt, classes = infoclasses, image_name=str(i))
path = 'results/'
os.makedirs(path, exist_ok=True)
df1 = A.ablate_filters(save_path='./results/', step=4)
if i == 0: df = df1
else: df.iloc[:,1:] += df1.iloc[:,1:]
df.iloc[:,1:] = df.iloc[:,1:]/(1. * num_images)
print (df)
