import keras import json import numpy as np import pandas as pd # import nibabel as nib import matplotlib.pyplot as plt from tensorflow.keras import backend as K # import util # from public_tests import * # from test_utils import * import nibabel as nib import tensorflow as tf tf.compat.v1.logging.set_verbosity(tf.compat.v1.logging.ERROR) import src.utils.xraypp as xraypp import src.utils.plotter as pltter import src.utils.batchgenerator as dg import src.utils.models as dl_models
0.1. Initialize the xray preprocessor object
HOME_DIR_2D = "./data/nih/images-small/"
DATA_DIR_2D = HOME_DIR_2D
train_df = pd.read_csv("./data/nih/train-small.csv")
valid_df = pd.read_csv("./data/nih/valid-small.csv")
test_df = pd.read_csv("./data/nih/test.csv")
labels = ['Cardiomegaly',
'Emphysema',
'Effusion',
'Hernia',
'Infiltration',
'Mass',
'Nodule',
'Atelectasis',
'Pneumothorax',
'Pleural_Thickening',
'Pneumonia',
'Fibrosis',
'Edema',
'Consolidation']
xray_pp_obj = xraypp.XrayPP(
train_df=train_df,
valid_df=valid_df,
test_df=test_df,
img_dir=DATA_DIR_2D,
target_w=320,
target_h=320,
labels=labels,
)
06/04//2024 01:21:1712389902 PM - INFO - XrayPP: Initializing the Medical Image Preprocessor Class
06/04//2024 01:21:1712389902 PM - INFO - ComputeStats: Initializing the Compute Stats Class
06/04//2024 01:21:1712389902 PM - INFO - XrayPP: Initializing the X-ray image preprocessing class
06/04//2024 01:21:1712389902 PM - INFO - XrayPP: Initializing the X-ray image preprocessing class
06/04//2024 01:21:1712389902 PM - INFO - XrayPP: Image Directory: data/nih/images-small
06/04//2024 01:21:1712389902 PM - INFO - XrayPP: Image Directory: data/nih/images-small
dg_xray = xray_pp_obj.get_generator(batch_size=32, num_channels=2)
06/04//2024 01:21:1712389903 PM - INFO - XrayPP: Initializing the Data Generator For Training, Testing and Validation generators
06/04//2024 01:21:1712389903 PM - INFO - XrayPP: Initializing the Data Generator For Training, Testing and Validation generators
Found 1000 validated image filenames.
Found 1000 validated image filenames.
Found 200 validated image filenames.
Found 420 validated image filenames.
xray_pp_obj.update_kwargs()
06/04//2024 01:21:1712389904 PM - INFO - XrayPP: Updating the kwargs for 2D model
06/04//2024 01:21:1712389904 PM - INFO - XrayPP: Updating the kwargs for 2D model
{'data_frames': {'train_df': Image Atelectasis Cardiomegaly Consolidation Edema \
0 00008270_015.png 0 0 0 0
1 00029855_001.png 1 0 0 0
2 00001297_000.png 0 0 0 0
3 00012359_002.png 0 0 0 0
4 00017951_001.png 0 0 0 0
.. ... ... ... ... ...
995 00015869_010.png 0 0 0 0
996 00020113_005.png 0 0 0 0
997 00019939_000.png 0 0 0 0
998 00030496_000.png 0 0 0 0
999 00012363_002.png 0 0 0 0
Effusion Emphysema Fibrosis Hernia Infiltration Mass Nodule \
0 0 0 0 0 0 0 0
1 1 0 0 0 1 0 0
2 0 0 0 0 0 0 0
3 0 0 0 0 0 0 0
4 0 0 0 0 1 0 0
.. ... ... ... ... ... ... ...
995 0 0 0 0 0 0 0
996 1 0 0 0 0 0 0
997 0 0 0 0 0 0 0
998 0 0 0 0 0 0 0
999 0 0 0 0 0 0 0
PatientId Pleural_Thickening Pneumonia Pneumothorax
0 8270 0 0 0
1 29855 0 0 0
2 1297 1 0 0
3 12359 0 0 0
4 17951 0 0 0
.. ... ... ... ...
995 15869 0 0 0
996 20113 0 0 0
997 19939 0 0 0
998 30496 0 0 0
999 12363 0 0 0
[1000 rows x 16 columns],
'valid_df': Image Atelectasis Cardiomegaly Consolidation Edema \
0 00008270_015.png 0 0 0 0
1 00029855_001.png 1 0 0 0
2 00001297_000.png 0 0 0 0
3 00012359_002.png 0 0 0 0
4 00017951_001.png 0 0 0 0
.. ... ... ... ... ...
195 00025695_017.png 0 0 0 0
196 00029454_000.png 0 0 0 0
197 00020213_004.png 0 0 0 0
198 00003272_000.png 0 0 0 0
199 00001108_004.png 1 0 0 0
Effusion Emphysema Fibrosis Hernia Infiltration Mass Nodule \
0 0 0 0 0 0 0 0
1 1 0 0 0 1 0 0
2 0 0 0 0 0 0 0
3 0 0 0 0 0 0 0
4 0 0 0 0 1 0 0
.. ... ... ... ... ... ... ...
195 1 0 0 0 0 0 0
196 0 0 0 0 0 0 0
197 0 0 0 0 1 0 0
198 0 0 0 1 0 0 0
199 0 0 0 0 0 0 0
PatientId Pleural_Thickening Pneumonia Pneumothorax
0 8270 0 0 0
1 29855 0 0 0
2 1297 1 0 0
3 12359 0 0 0
4 17951 0 0 0
.. ... ... ... ...
195 25695 0 0 1
196 29454 0 0 0
197 20213 0 0 0
198 3272 0 0 0
199 1108 0 1 0
[200 rows x 16 columns],
'test_df': Image PatientId Cardiomegaly Emphysema Effusion Hernia \
0 00021463_005.png 21463 0 0 1 1
1 00019733_002.png 19733 0 0 0 1
2 00021463_001.png 21463 0 0 0 1
3 00017136_020.png 17136 0 0 0 1
4 00027477_000.png 27477 0 0 0 1
.. ... ... ... ... ... ...
415 00010497_000.png 10497 0 0 0 0
416 00017262_001.png 17262 0 0 0 0
417 00021653_000.png 21653 0 0 0 0
418 00017243_001.png 17243 0 0 0 0
419 00016142_013.png 16142 0 0 0 0
Infiltration Mass Nodule Atelectasis Pneumothorax \
0 1 0 0 1 0
1 0 0 0 1 0
2 0 0 0 0 0
3 0 0 0 0 0
4 1 0 0 0 0
.. ... ... ... ... ...
415 0 1 0 0 0
416 0 1 0 0 0
417 0 1 0 0 0
418 0 1 0 0 1
419 1 1 0 1 0
Pleural_Thickening Pneumonia Fibrosis Edema Consolidation
0 0 0 0 1 0
1 0 0 0 0 0
2 0 0 0 0 0
3 0 0 0 0 0
4 0 0 0 0 0
.. ... ... ... ... ...
415 0 0 0 0 0
416 0 0 0 0 0
417 1 0 0 0 0
418 1 0 0 0 0
419 1 0 0 0 0
[420 rows x 16 columns]},
'batch_size': 32,
'dim': (320, 320, 16),
'num_channels': 2,
'num_classes': 14,
'shuffle': False,
'x_col': 'Image',
'y_cols': ['Cardiomegaly',
'Emphysema',
'Effusion',
'Hernia',
'Infiltration',
'Mass',
'Nodule',
'Atelectasis',
'Pneumothorax',
'Pleural_Thickening',
'Pneumonia',
'Fibrosis',
'Edema',
'Consolidation'],
'train_generator': None,
'test_generator': None,
'valid_generator': None,
'labels': ['Cardiomegaly',
'Emphysema',
'Effusion',
'Hernia',
'Infiltration',
'Mass',
'Nodule',
'Atelectasis',
'Pneumothorax',
'Pleural_Thickening',
'Pneumonia',
'Fibrosis',
'Edema',
'Consolidation'],
'pos_weights': None,
'neg_weights': None,
'orig_x': 240,
'orig_y': 240,
'orig_z': 155,
'output_x': 160,
'output_y': 160,
'output_z': 16,
'max_tries': 1000,
'background_threshold': 0.95,
'model_2d_or_3d': '2D'}
xray_pp_obj.kwargs
{'data_frames': {'train_df': Image Atelectasis Cardiomegaly Consolidation Edema \
0 00008270_015.png 0 0 0 0
1 00029855_001.png 1 0 0 0
2 00001297_000.png 0 0 0 0
3 00012359_002.png 0 0 0 0
4 00017951_001.png 0 0 0 0
.. ... ... ... ... ...
995 00015869_010.png 0 0 0 0
996 00020113_005.png 0 0 0 0
997 00019939_000.png 0 0 0 0
998 00030496_000.png 0 0 0 0
999 00012363_002.png 0 0 0 0
Effusion Emphysema Fibrosis Hernia Infiltration Mass Nodule \
0 0 0 0 0 0 0 0
1 1 0 0 0 1 0 0
2 0 0 0 0 0 0 0
3 0 0 0 0 0 0 0
4 0 0 0 0 1 0 0
.. ... ... ... ... ... ... ...
995 0 0 0 0 0 0 0
996 1 0 0 0 0 0 0
997 0 0 0 0 0 0 0
998 0 0 0 0 0 0 0
999 0 0 0 0 0 0 0
PatientId Pleural_Thickening Pneumonia Pneumothorax
0 8270 0 0 0
1 29855 0 0 0
2 1297 1 0 0
3 12359 0 0 0
4 17951 0 0 0
.. ... ... ... ...
995 15869 0 0 0
996 20113 0 0 0
997 19939 0 0 0
998 30496 0 0 0
999 12363 0 0 0
[1000 rows x 16 columns],
'valid_df': Image Atelectasis Cardiomegaly Consolidation Edema \
0 00008270_015.png 0 0 0 0
1 00029855_001.png 1 0 0 0
2 00001297_000.png 0 0 0 0
3 00012359_002.png 0 0 0 0
4 00017951_001.png 0 0 0 0
.. ... ... ... ... ...
195 00025695_017.png 0 0 0 0
196 00029454_000.png 0 0 0 0
197 00020213_004.png 0 0 0 0
198 00003272_000.png 0 0 0 0
199 00001108_004.png 1 0 0 0
Effusion Emphysema Fibrosis Hernia Infiltration Mass Nodule \
0 0 0 0 0 0 0 0
1 1 0 0 0 1 0 0
2 0 0 0 0 0 0 0
3 0 0 0 0 0 0 0
4 0 0 0 0 1 0 0
.. ... ... ... ... ... ... ...
195 1 0 0 0 0 0 0
196 0 0 0 0 0 0 0
197 0 0 0 0 1 0 0
198 0 0 0 1 0 0 0
199 0 0 0 0 0 0 0
PatientId Pleural_Thickening Pneumonia Pneumothorax
0 8270 0 0 0
1 29855 0 0 0
2 1297 1 0 0
3 12359 0 0 0
4 17951 0 0 0
.. ... ... ... ...
195 25695 0 0 1
196 29454 0 0 0
197 20213 0 0 0
198 3272 0 0 0
199 1108 0 1 0
[200 rows x 16 columns],
'test_df': Image PatientId Cardiomegaly Emphysema Effusion Hernia \
0 00021463_005.png 21463 0 0 1 1
1 00019733_002.png 19733 0 0 0 1
2 00021463_001.png 21463 0 0 0 1
3 00017136_020.png 17136 0 0 0 1
4 00027477_000.png 27477 0 0 0 1
.. ... ... ... ... ... ...
415 00010497_000.png 10497 0 0 0 0
416 00017262_001.png 17262 0 0 0 0
417 00021653_000.png 21653 0 0 0 0
418 00017243_001.png 17243 0 0 0 0
419 00016142_013.png 16142 0 0 0 0
Infiltration Mass Nodule Atelectasis Pneumothorax \
0 1 0 0 1 0
1 0 0 0 1 0
2 0 0 0 0 0
3 0 0 0 0 0
4 1 0 0 0 0
.. ... ... ... ... ...
415 0 1 0 0 0
416 0 1 0 0 0
417 0 1 0 0 0
418 0 1 0 0 1
419 1 1 0 1 0
Pleural_Thickening Pneumonia Fibrosis Edema Consolidation
0 0 0 0 1 0
1 0 0 0 0 0
2 0 0 0 0 0
3 0 0 0 0 0
4 0 0 0 0 0
.. ... ... ... ... ...
415 0 0 0 0 0
416 0 0 0 0 0
417 1 0 0 0 0
418 1 0 0 0 0
419 1 0 0 0 0
[420 rows x 16 columns]},
'batch_size': 32,
'dim': (320, 320, 16),
'num_channels': 2,
'num_classes': 14,
'shuffle': False,
'x_col': 'Image',
'y_cols': ['Cardiomegaly',
'Emphysema',
'Effusion',
'Hernia',
'Infiltration',
'Mass',
'Nodule',
'Atelectasis',
'Pneumothorax',
'Pleural_Thickening',
'Pneumonia',
'Fibrosis',
'Edema',
'Consolidation'],
'train_generator': None,
'test_generator': None,
'valid_generator': None,
'labels': ['Cardiomegaly',
'Emphysema',
'Effusion',
'Hernia',
'Infiltration',
'Mass',
'Nodule',
'Atelectasis',
'Pneumothorax',
'Pleural_Thickening',
'Pneumonia',
'Fibrosis',
'Edema',
'Consolidation'],
'pos_weights': None,
'neg_weights': None,
'orig_x': 240,
'orig_y': 240,
'orig_z': 155,
'output_x': 160,
'output_y': 160,
'output_z': 16,
'max_tries': 1000,
'background_threshold': 0.95,
'model_2d_or_3d': '2D'}
train_generator = dg_xray.get_train_generator() valid_generator, test_generator = dg_xray.get_test_and_valid_generator()
Found 1000 validated image filenames.
Found 1000 validated image filenames.
Found 200 validated image filenames.
Found 420 validated image filenames.
x, y = train_generator.__getitem__(0) plt.imshow(x[0]);
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
1.0. Applying DenseNet Model using models and compute stats modules to imbalanced Chest X-Ray Datasets
import numpy as np import pandas as pd import seaborn as sns import matplotlib.pyplot as plt from keras.preprocessing.image import ImageDataGenerator from keras.applications.densenet import DenseNet121 from keras.layers import Dense, GlobalAveragePooling2D from keras.models import Model from keras import backend as K from keras.models import load_model # import util # from public_tests import * # from test_utils import * import tensorflow as tf tf.compat.v1.logging.set_verbosity(tf.compat.v1.logging.ERROR)
1.1. Get contributions and frequencies of positive, negative classes using training data
model_obj = dl_models.DLModels(model_2d_dir='./models/nih/densenet.hdf5', model_3d_dir='./models/pretrained_model.h5', pre_trained_wts='./models/nih/pretrained_model.h5') labels = xray_pp_obj.labels freq_pos, freq_neg = model_obj.cs.compute_class_freqs(train_generator.labels) pos_neg_dict = model_obj.cs.calc_pos_neg_weights(freq_pos=freq_pos, freq_neg=freq_neg)
06/04//2024 01:21:1712389910 PM - INFO - DLModels: Initializing the Deep Learning Model Class
06/04//2024 01:21:1712389910 PM - INFO - ComputeStats: Initializing the Compute Stats Class
06/04//2024 01:21:1712389910 PM - INFO - ComputeStats: Initializing the Compute Stats Class
# Create DataFrame for positive values
pos_data = pd.DataFrame({"Class": labels, "Label": "Positive", "Value": freq_pos})
# Create DataFrame for negative values
neg_data = pd.DataFrame([{"Class": labels[l], "Label": "Negative", "Value": v} for l, v in enumerate(freq_neg)])
# Concatenate positive and negative data
data = pd.concat([pos_data, neg_data], ignore_index=True)
# Plot data
plt.xticks(rotation=90)
f = sns.barplot(x="Class", y="Value", hue="Label", data=data)
f.set_title('Class distribution Before Balancing')
Text(0.5, 1.0, 'Class distribution Before Balancing')
# Create DataFrame for positive contributions
pos_data = pd.DataFrame({"Class": labels, "Label": "Positive", "Value": pos_neg_dict['pos_contribution'] })
# Create DataFrame for negative contributions
neg_data = pd.DataFrame([{"Class": labels[l], "Label": "Negative", "Value": v} for l, v in enumerate(pos_neg_dict['neg_contribution'])])
# Concatenate positive and negative contributions
data = pd.concat([pos_data, neg_data], ignore_index=True)
# Plot data
plt.xticks(rotation=90)
g = sns.barplot(x="Class", y="Value", hue="Label" ,data=data);
g.set_title('Class distribution After Balancing')
Text(0.5, 1.0, 'Class distribution After Balancing')
1.2. Get DenseNet Model Summary
kwargs = {
'test_generator': test_generator,
'labels': labels,
'pos_weights': pos_neg_dict['pos_weights'],
'neg_weights': pos_neg_dict['neg_weights'],
}
model_2d = model_obj.weighted_dense_net_model_2d(**kwargs)
model_2d.summary()
Model: "model"
__________________________________________________________________________________________________
Layer (type) Output Shape Param # Connected to
==================================================================================================
input_1 (InputLayer) [(None, None, None, 0 []
3)]
zero_padding2d (ZeroPadding2D) (None, None, None, 0 ['input_1[0][0]']
3)
conv1/conv (Conv2D) (None, None, None, 9408 ['zero_padding2d[0][0]']
64)
conv1/bn (BatchNormalization) (None, None, None, 256 ['conv1/conv[0][0]']
64)
conv1/relu (Activation) (None, None, None, 0 ['conv1/bn[0][0]']
64)
zero_padding2d_1 (ZeroPadding2 (None, None, None, 0 ['conv1/relu[0][0]']
D) 64)
pool1 (MaxPooling2D) (None, None, None, 0 ['zero_padding2d_1[0][0]']
64)
conv2_block1_0_bn (BatchNormal (None, None, None, 256 ['pool1[0][0]']
ization) 64)
conv2_block1_0_relu (Activatio (None, None, None, 0 ['conv2_block1_0_bn[0][0]']
n) 64)
conv2_block1_1_conv (Conv2D) (None, None, None, 8192 ['conv2_block1_0_relu[0][0]']
128)
conv2_block1_1_bn (BatchNormal (None, None, None, 512 ['conv2_block1_1_conv[0][0]']
ization) 128)
conv2_block1_1_relu (Activatio (None, None, None, 0 ['conv2_block1_1_bn[0][0]']
n) 128)
conv2_block1_2_conv (Conv2D) (None, None, None, 36864 ['conv2_block1_1_relu[0][0]']
32)
conv2_block1_concat (Concatena (None, None, None, 0 ['pool1[0][0]',
te) 96) 'conv2_block1_2_conv[0][0]']
conv2_block2_0_bn (BatchNormal (None, None, None, 384 ['conv2_block1_concat[0][0]']
ization) 96)
conv2_block2_0_relu (Activatio (None, None, None, 0 ['conv2_block2_0_bn[0][0]']
n) 96)
conv2_block2_1_conv (Conv2D) (None, None, None, 12288 ['conv2_block2_0_relu[0][0]']
128)
conv2_block2_1_bn (BatchNormal (None, None, None, 512 ['conv2_block2_1_conv[0][0]']
ization) 128)
conv2_block2_1_relu (Activatio (None, None, None, 0 ['conv2_block2_1_bn[0][0]']
n) 128)
conv2_block2_2_conv (Conv2D) (None, None, None, 36864 ['conv2_block2_1_relu[0][0]']
32)
conv2_block2_concat (Concatena (None, None, None, 0 ['conv2_block1_concat[0][0]',
te) 128) 'conv2_block2_2_conv[0][0]']
conv2_block3_0_bn (BatchNormal (None, None, None, 512 ['conv2_block2_concat[0][0]']
ization) 128)
conv2_block3_0_relu (Activatio (None, None, None, 0 ['conv2_block3_0_bn[0][0]']
n) 128)
conv2_block3_1_conv (Conv2D) (None, None, None, 16384 ['conv2_block3_0_relu[0][0]']
128)
conv2_block3_1_bn (BatchNormal (None, None, None, 512 ['conv2_block3_1_conv[0][0]']
ization) 128)
conv2_block3_1_relu (Activatio (None, None, None, 0 ['conv2_block3_1_bn[0][0]']
n) 128)
conv2_block3_2_conv (Conv2D) (None, None, None, 36864 ['conv2_block3_1_relu[0][0]']
32)
conv2_block3_concat (Concatena (None, None, None, 0 ['conv2_block2_concat[0][0]',
te) 160) 'conv2_block3_2_conv[0][0]']
conv2_block4_0_bn (BatchNormal (None, None, None, 640 ['conv2_block3_concat[0][0]']
ization) 160)
conv2_block4_0_relu (Activatio (None, None, None, 0 ['conv2_block4_0_bn[0][0]']
n) 160)
conv2_block4_1_conv (Conv2D) (None, None, None, 20480 ['conv2_block4_0_relu[0][0]']
128)
conv2_block4_1_bn (BatchNormal (None, None, None, 512 ['conv2_block4_1_conv[0][0]']
ization) 128)
conv2_block4_1_relu (Activatio (None, None, None, 0 ['conv2_block4_1_bn[0][0]']
n) 128)
conv2_block4_2_conv (Conv2D) (None, None, None, 36864 ['conv2_block4_1_relu[0][0]']
32)
conv2_block4_concat (Concatena (None, None, None, 0 ['conv2_block3_concat[0][0]',
te) 192) 'conv2_block4_2_conv[0][0]']
conv2_block5_0_bn (BatchNormal (None, None, None, 768 ['conv2_block4_concat[0][0]']
ization) 192)
conv2_block5_0_relu (Activatio (None, None, None, 0 ['conv2_block5_0_bn[0][0]']
n) 192)
conv2_block5_1_conv (Conv2D) (None, None, None, 24576 ['conv2_block5_0_relu[0][0]']
128)
conv2_block5_1_bn (BatchNormal (None, None, None, 512 ['conv2_block5_1_conv[0][0]']
ization) 128)
conv2_block5_1_relu (Activatio (None, None, None, 0 ['conv2_block5_1_bn[0][0]']
n) 128)
conv2_block5_2_conv (Conv2D) (None, None, None, 36864 ['conv2_block5_1_relu[0][0]']
32)
conv2_block5_concat (Concatena (None, None, None, 0 ['conv2_block4_concat[0][0]',
te) 224) 'conv2_block5_2_conv[0][0]']
conv2_block6_0_bn (BatchNormal (None, None, None, 896 ['conv2_block5_concat[0][0]']
ization) 224)
conv2_block6_0_relu (Activatio (None, None, None, 0 ['conv2_block6_0_bn[0][0]']
n) 224)
conv2_block6_1_conv (Conv2D) (None, None, None, 28672 ['conv2_block6_0_relu[0][0]']
128)
conv2_block6_1_bn (BatchNormal (None, None, None, 512 ['conv2_block6_1_conv[0][0]']
ization) 128)
conv2_block6_1_relu (Activatio (None, None, None, 0 ['conv2_block6_1_bn[0][0]']
n) 128)
conv2_block6_2_conv (Conv2D) (None, None, None, 36864 ['conv2_block6_1_relu[0][0]']
32)
conv2_block6_concat (Concatena (None, None, None, 0 ['conv2_block5_concat[0][0]',
te) 256) 'conv2_block6_2_conv[0][0]']
pool2_bn (BatchNormalization) (None, None, None, 1024 ['conv2_block6_concat[0][0]']
256)
pool2_relu (Activation) (None, None, None, 0 ['pool2_bn[0][0]']
256)
pool2_conv (Conv2D) (None, None, None, 32768 ['pool2_relu[0][0]']
128)
pool2_pool (AveragePooling2D) (None, None, None, 0 ['pool2_conv[0][0]']
128)
conv3_block1_0_bn (BatchNormal (None, None, None, 512 ['pool2_pool[0][0]']
ization) 128)
conv3_block1_0_relu (Activatio (None, None, None, 0 ['conv3_block1_0_bn[0][0]']
n) 128)
conv3_block1_1_conv (Conv2D) (None, None, None, 16384 ['conv3_block1_0_relu[0][0]']
128)
conv3_block1_1_bn (BatchNormal (None, None, None, 512 ['conv3_block1_1_conv[0][0]']
ization) 128)
conv3_block1_1_relu (Activatio (None, None, None, 0 ['conv3_block1_1_bn[0][0]']
n) 128)
conv3_block1_2_conv (Conv2D) (None, None, None, 36864 ['conv3_block1_1_relu[0][0]']
32)
conv3_block1_concat (Concatena (None, None, None, 0 ['pool2_pool[0][0]',
te) 160) 'conv3_block1_2_conv[0][0]']
conv3_block2_0_bn (BatchNormal (None, None, None, 640 ['conv3_block1_concat[0][0]']
ization) 160)
conv3_block2_0_relu (Activatio (None, None, None, 0 ['conv3_block2_0_bn[0][0]']
n) 160)
conv3_block2_1_conv (Conv2D) (None, None, None, 20480 ['conv3_block2_0_relu[0][0]']
128)
conv3_block2_1_bn (BatchNormal (None, None, None, 512 ['conv3_block2_1_conv[0][0]']
ization) 128)
conv3_block2_1_relu (Activatio (None, None, None, 0 ['conv3_block2_1_bn[0][0]']
n) 128)
conv3_block2_2_conv (Conv2D) (None, None, None, 36864 ['conv3_block2_1_relu[0][0]']
32)
conv3_block2_concat (Concatena (None, None, None, 0 ['conv3_block1_concat[0][0]',
te) 192) 'conv3_block2_2_conv[0][0]']
conv3_block3_0_bn (BatchNormal (None, None, None, 768 ['conv3_block2_concat[0][0]']
ization) 192)
conv3_block3_0_relu (Activatio (None, None, None, 0 ['conv3_block3_0_bn[0][0]']
n) 192)
conv3_block3_1_conv (Conv2D) (None, None, None, 24576 ['conv3_block3_0_relu[0][0]']
128)
conv3_block3_1_bn (BatchNormal (None, None, None, 512 ['conv3_block3_1_conv[0][0]']
ization) 128)
conv3_block3_1_relu (Activatio (None, None, None, 0 ['conv3_block3_1_bn[0][0]']
n) 128)
conv3_block3_2_conv (Conv2D) (None, None, None, 36864 ['conv3_block3_1_relu[0][0]']
32)
conv3_block3_concat (Concatena (None, None, None, 0 ['conv3_block2_concat[0][0]',
te) 224) 'conv3_block3_2_conv[0][0]']
conv3_block4_0_bn (BatchNormal (None, None, None, 896 ['conv3_block3_concat[0][0]']
ization) 224)
conv3_block4_0_relu (Activatio (None, None, None, 0 ['conv3_block4_0_bn[0][0]']
n) 224)
conv3_block4_1_conv (Conv2D) (None, None, None, 28672 ['conv3_block4_0_relu[0][0]']
128)
conv3_block4_1_bn (BatchNormal (None, None, None, 512 ['conv3_block4_1_conv[0][0]']
ization) 128)
conv3_block4_1_relu (Activatio (None, None, None, 0 ['conv3_block4_1_bn[0][0]']
n) 128)
conv3_block4_2_conv (Conv2D) (None, None, None, 36864 ['conv3_block4_1_relu[0][0]']
32)
conv3_block4_concat (Concatena (None, None, None, 0 ['conv3_block3_concat[0][0]',
te) 256) 'conv3_block4_2_conv[0][0]']
conv3_block5_0_bn (BatchNormal (None, None, None, 1024 ['conv3_block4_concat[0][0]']
ization) 256)
conv3_block5_0_relu (Activatio (None, None, None, 0 ['conv3_block5_0_bn[0][0]']
n) 256)
conv3_block5_1_conv (Conv2D) (None, None, None, 32768 ['conv3_block5_0_relu[0][0]']
128)
conv3_block5_1_bn (BatchNormal (None, None, None, 512 ['conv3_block5_1_conv[0][0]']
ization) 128)
conv3_block5_1_relu (Activatio (None, None, None, 0 ['conv3_block5_1_bn[0][0]']
n) 128)
conv3_block5_2_conv (Conv2D) (None, None, None, 36864 ['conv3_block5_1_relu[0][0]']
32)
conv3_block5_concat (Concatena (None, None, None, 0 ['conv3_block4_concat[0][0]',
te) 288) 'conv3_block5_2_conv[0][0]']
conv3_block6_0_bn (BatchNormal (None, None, None, 1152 ['conv3_block5_concat[0][0]']
ization) 288)
conv3_block6_0_relu (Activatio (None, None, None, 0 ['conv3_block6_0_bn[0][0]']
n) 288)
conv3_block6_1_conv (Conv2D) (None, None, None, 36864 ['conv3_block6_0_relu[0][0]']
128)
conv3_block6_1_bn (BatchNormal (None, None, None, 512 ['conv3_block6_1_conv[0][0]']
ization) 128)
conv3_block6_1_relu (Activatio (None, None, None, 0 ['conv3_block6_1_bn[0][0]']
n) 128)
conv3_block6_2_conv (Conv2D) (None, None, None, 36864 ['conv3_block6_1_relu[0][0]']
32)
conv3_block6_concat (Concatena (None, None, None, 0 ['conv3_block5_concat[0][0]',
te) 320) 'conv3_block6_2_conv[0][0]']
conv3_block7_0_bn (BatchNormal (None, None, None, 1280 ['conv3_block6_concat[0][0]']
ization) 320)
conv3_block7_0_relu (Activatio (None, None, None, 0 ['conv3_block7_0_bn[0][0]']
n) 320)
conv3_block7_1_conv (Conv2D) (None, None, None, 40960 ['conv3_block7_0_relu[0][0]']
128)
conv3_block7_1_bn (BatchNormal (None, None, None, 512 ['conv3_block7_1_conv[0][0]']
ization) 128)
conv3_block7_1_relu (Activatio (None, None, None, 0 ['conv3_block7_1_bn[0][0]']
n) 128)
conv3_block7_2_conv (Conv2D) (None, None, None, 36864 ['conv3_block7_1_relu[0][0]']
32)
conv3_block7_concat (Concatena (None, None, None, 0 ['conv3_block6_concat[0][0]',
te) 352) 'conv3_block7_2_conv[0][0]']
conv3_block8_0_bn (BatchNormal (None, None, None, 1408 ['conv3_block7_concat[0][0]']
ization) 352)
conv3_block8_0_relu (Activatio (None, None, None, 0 ['conv3_block8_0_bn[0][0]']
n) 352)
conv3_block8_1_conv (Conv2D) (None, None, None, 45056 ['conv3_block8_0_relu[0][0]']
128)
conv3_block8_1_bn (BatchNormal (None, None, None, 512 ['conv3_block8_1_conv[0][0]']
ization) 128)
conv3_block8_1_relu (Activatio (None, None, None, 0 ['conv3_block8_1_bn[0][0]']
n) 128)
conv3_block8_2_conv (Conv2D) (None, None, None, 36864 ['conv3_block8_1_relu[0][0]']
32)
conv3_block8_concat (Concatena (None, None, None, 0 ['conv3_block7_concat[0][0]',
te) 384) 'conv3_block8_2_conv[0][0]']
conv3_block9_0_bn (BatchNormal (None, None, None, 1536 ['conv3_block8_concat[0][0]']
ization) 384)
conv3_block9_0_relu (Activatio (None, None, None, 0 ['conv3_block9_0_bn[0][0]']
n) 384)
conv3_block9_1_conv (Conv2D) (None, None, None, 49152 ['conv3_block9_0_relu[0][0]']
128)
conv3_block9_1_bn (BatchNormal (None, None, None, 512 ['conv3_block9_1_conv[0][0]']
ization) 128)
conv3_block9_1_relu (Activatio (None, None, None, 0 ['conv3_block9_1_bn[0][0]']
n) 128)
conv3_block9_2_conv (Conv2D) (None, None, None, 36864 ['conv3_block9_1_relu[0][0]']
32)
conv3_block9_concat (Concatena (None, None, None, 0 ['conv3_block8_concat[0][0]',
te) 416) 'conv3_block9_2_conv[0][0]']
conv3_block10_0_bn (BatchNorma (None, None, None, 1664 ['conv3_block9_concat[0][0]']
lization) 416)
conv3_block10_0_relu (Activati (None, None, None, 0 ['conv3_block10_0_bn[0][0]']
on) 416)
conv3_block10_1_conv (Conv2D) (None, None, None, 53248 ['conv3_block10_0_relu[0][0]']
128)
conv3_block10_1_bn (BatchNorma (None, None, None, 512 ['conv3_block10_1_conv[0][0]']
lization) 128)
conv3_block10_1_relu (Activati (None, None, None, 0 ['conv3_block10_1_bn[0][0]']
on) 128)
conv3_block10_2_conv (Conv2D) (None, None, None, 36864 ['conv3_block10_1_relu[0][0]']
32)
conv3_block10_concat (Concaten (None, None, None, 0 ['conv3_block9_concat[0][0]',
ate) 448) 'conv3_block10_2_conv[0][0]']
conv3_block11_0_bn (BatchNorma (None, None, None, 1792 ['conv3_block10_concat[0][0]']
lization) 448)
conv3_block11_0_relu (Activati (None, None, None, 0 ['conv3_block11_0_bn[0][0]']
on) 448)
conv3_block11_1_conv (Conv2D) (None, None, None, 57344 ['conv3_block11_0_relu[0][0]']
128)
conv3_block11_1_bn (BatchNorma (None, None, None, 512 ['conv3_block11_1_conv[0][0]']
lization) 128)
conv3_block11_1_relu (Activati (None, None, None, 0 ['conv3_block11_1_bn[0][0]']
on) 128)
conv3_block11_2_conv (Conv2D) (None, None, None, 36864 ['conv3_block11_1_relu[0][0]']
32)
conv3_block11_concat (Concaten (None, None, None, 0 ['conv3_block10_concat[0][0]',
ate) 480) 'conv3_block11_2_conv[0][0]']
conv3_block12_0_bn (BatchNorma (None, None, None, 1920 ['conv3_block11_concat[0][0]']
lization) 480)
conv3_block12_0_relu (Activati (None, None, None, 0 ['conv3_block12_0_bn[0][0]']
on) 480)
conv3_block12_1_conv (Conv2D) (None, None, None, 61440 ['conv3_block12_0_relu[0][0]']
128)
conv3_block12_1_bn (BatchNorma (None, None, None, 512 ['conv3_block12_1_conv[0][0]']
lization) 128)
conv3_block12_1_relu (Activati (None, None, None, 0 ['conv3_block12_1_bn[0][0]']
on) 128)
conv3_block12_2_conv (Conv2D) (None, None, None, 36864 ['conv3_block12_1_relu[0][0]']
32)
conv3_block12_concat (Concaten (None, None, None, 0 ['conv3_block11_concat[0][0]',
ate) 512) 'conv3_block12_2_conv[0][0]']
pool3_bn (BatchNormalization) (None, None, None, 2048 ['conv3_block12_concat[0][0]']
512)
pool3_relu (Activation) (None, None, None, 0 ['pool3_bn[0][0]']
512)
pool3_conv (Conv2D) (None, None, None, 131072 ['pool3_relu[0][0]']
256)
pool3_pool (AveragePooling2D) (None, None, None, 0 ['pool3_conv[0][0]']
256)
conv4_block1_0_bn (BatchNormal (None, None, None, 1024 ['pool3_pool[0][0]']
ization) 256)
conv4_block1_0_relu (Activatio (None, None, None, 0 ['conv4_block1_0_bn[0][0]']
n) 256)
conv4_block1_1_conv (Conv2D) (None, None, None, 32768 ['conv4_block1_0_relu[0][0]']
128)
conv4_block1_1_bn (BatchNormal (None, None, None, 512 ['conv4_block1_1_conv[0][0]']
ization) 128)
conv4_block1_1_relu (Activatio (None, None, None, 0 ['conv4_block1_1_bn[0][0]']
n) 128)
conv4_block1_2_conv (Conv2D) (None, None, None, 36864 ['conv4_block1_1_relu[0][0]']
32)
conv4_block1_concat (Concatena (None, None, None, 0 ['pool3_pool[0][0]',
te) 288) 'conv4_block1_2_conv[0][0]']
conv4_block2_0_bn (BatchNormal (None, None, None, 1152 ['conv4_block1_concat[0][0]']
ization) 288)
conv4_block2_0_relu (Activatio (None, None, None, 0 ['conv4_block2_0_bn[0][0]']
n) 288)
conv4_block2_1_conv (Conv2D) (None, None, None, 36864 ['conv4_block2_0_relu[0][0]']
128)
conv4_block2_1_bn (BatchNormal (None, None, None, 512 ['conv4_block2_1_conv[0][0]']
ization) 128)
conv4_block2_1_relu (Activatio (None, None, None, 0 ['conv4_block2_1_bn[0][0]']
n) 128)
conv4_block2_2_conv (Conv2D) (None, None, None, 36864 ['conv4_block2_1_relu[0][0]']
32)
conv4_block2_concat (Concatena (None, None, None, 0 ['conv4_block1_concat[0][0]',
te) 320) 'conv4_block2_2_conv[0][0]']
conv4_block3_0_bn (BatchNormal (None, None, None, 1280 ['conv4_block2_concat[0][0]']
ization) 320)
conv4_block3_0_relu (Activatio (None, None, None, 0 ['conv4_block3_0_bn[0][0]']
n) 320)
conv4_block3_1_conv (Conv2D) (None, None, None, 40960 ['conv4_block3_0_relu[0][0]']
128)
conv4_block3_1_bn (BatchNormal (None, None, None, 512 ['conv4_block3_1_conv[0][0]']
ization) 128)
conv4_block3_1_relu (Activatio (None, None, None, 0 ['conv4_block3_1_bn[0][0]']
n) 128)
conv4_block3_2_conv (Conv2D) (None, None, None, 36864 ['conv4_block3_1_relu[0][0]']
32)
conv4_block3_concat (Concatena (None, None, None, 0 ['conv4_block2_concat[0][0]',
te) 352) 'conv4_block3_2_conv[0][0]']
conv4_block4_0_bn (BatchNormal (None, None, None, 1408 ['conv4_block3_concat[0][0]']
ization) 352)
conv4_block4_0_relu (Activatio (None, None, None, 0 ['conv4_block4_0_bn[0][0]']
n) 352)
conv4_block4_1_conv (Conv2D) (None, None, None, 45056 ['conv4_block4_0_relu[0][0]']
128)
conv4_block4_1_bn (BatchNormal (None, None, None, 512 ['conv4_block4_1_conv[0][0]']
ization) 128)
conv4_block4_1_relu (Activatio (None, None, None, 0 ['conv4_block4_1_bn[0][0]']
n) 128)
conv4_block4_2_conv (Conv2D) (None, None, None, 36864 ['conv4_block4_1_relu[0][0]']
32)
conv4_block4_concat (Concatena (None, None, None, 0 ['conv4_block3_concat[0][0]',
te) 384) 'conv4_block4_2_conv[0][0]']
conv4_block5_0_bn (BatchNormal (None, None, None, 1536 ['conv4_block4_concat[0][0]']
ization) 384)
conv4_block5_0_relu (Activatio (None, None, None, 0 ['conv4_block5_0_bn[0][0]']
n) 384)
conv4_block5_1_conv (Conv2D) (None, None, None, 49152 ['conv4_block5_0_relu[0][0]']
128)
conv4_block5_1_bn (BatchNormal (None, None, None, 512 ['conv4_block5_1_conv[0][0]']
ization) 128)
conv4_block5_1_relu (Activatio (None, None, None, 0 ['conv4_block5_1_bn[0][0]']
n) 128)
conv4_block5_2_conv (Conv2D) (None, None, None, 36864 ['conv4_block5_1_relu[0][0]']
32)
conv4_block5_concat (Concatena (None, None, None, 0 ['conv4_block4_concat[0][0]',
te) 416) 'conv4_block5_2_conv[0][0]']
conv4_block6_0_bn (BatchNormal (None, None, None, 1664 ['conv4_block5_concat[0][0]']
ization) 416)
conv4_block6_0_relu (Activatio (None, None, None, 0 ['conv4_block6_0_bn[0][0]']
n) 416)
conv4_block6_1_conv (Conv2D) (None, None, None, 53248 ['conv4_block6_0_relu[0][0]']
128)
conv4_block6_1_bn (BatchNormal (None, None, None, 512 ['conv4_block6_1_conv[0][0]']
ization) 128)
conv4_block6_1_relu (Activatio (None, None, None, 0 ['conv4_block6_1_bn[0][0]']
n) 128)
conv4_block6_2_conv (Conv2D) (None, None, None, 36864 ['conv4_block6_1_relu[0][0]']
32)
conv4_block6_concat (Concatena (None, None, None, 0 ['conv4_block5_concat[0][0]',
te) 448) 'conv4_block6_2_conv[0][0]']
conv4_block7_0_bn (BatchNormal (None, None, None, 1792 ['conv4_block6_concat[0][0]']
ization) 448)
conv4_block7_0_relu (Activatio (None, None, None, 0 ['conv4_block7_0_bn[0][0]']
n) 448)
conv4_block7_1_conv (Conv2D) (None, None, None, 57344 ['conv4_block7_0_relu[0][0]']
128)
conv4_block7_1_bn (BatchNormal (None, None, None, 512 ['conv4_block7_1_conv[0][0]']
ization) 128)
conv4_block7_1_relu (Activatio (None, None, None, 0 ['conv4_block7_1_bn[0][0]']
n) 128)
conv4_block7_2_conv (Conv2D) (None, None, None, 36864 ['conv4_block7_1_relu[0][0]']
32)
conv4_block7_concat (Concatena (None, None, None, 0 ['conv4_block6_concat[0][0]',
te) 480) 'conv4_block7_2_conv[0][0]']
conv4_block8_0_bn (BatchNormal (None, None, None, 1920 ['conv4_block7_concat[0][0]']
ization) 480)
conv4_block8_0_relu (Activatio (None, None, None, 0 ['conv4_block8_0_bn[0][0]']
n) 480)
conv4_block8_1_conv (Conv2D) (None, None, None, 61440 ['conv4_block8_0_relu[0][0]']
128)
conv4_block8_1_bn (BatchNormal (None, None, None, 512 ['conv4_block8_1_conv[0][0]']
ization) 128)
conv4_block8_1_relu (Activatio (None, None, None, 0 ['conv4_block8_1_bn[0][0]']
n) 128)
conv4_block8_2_conv (Conv2D) (None, None, None, 36864 ['conv4_block8_1_relu[0][0]']
32)
conv4_block8_concat (Concatena (None, None, None, 0 ['conv4_block7_concat[0][0]',
te) 512) 'conv4_block8_2_conv[0][0]']
conv4_block9_0_bn (BatchNormal (None, None, None, 2048 ['conv4_block8_concat[0][0]']
ization) 512)
conv4_block9_0_relu (Activatio (None, None, None, 0 ['conv4_block9_0_bn[0][0]']
n) 512)
conv4_block9_1_conv (Conv2D) (None, None, None, 65536 ['conv4_block9_0_relu[0][0]']
128)
conv4_block9_1_bn (BatchNormal (None, None, None, 512 ['conv4_block9_1_conv[0][0]']
ization) 128)
conv4_block9_1_relu (Activatio (None, None, None, 0 ['conv4_block9_1_bn[0][0]']
n) 128)
conv4_block9_2_conv (Conv2D) (None, None, None, 36864 ['conv4_block9_1_relu[0][0]']
32)
conv4_block9_concat (Concatena (None, None, None, 0 ['conv4_block8_concat[0][0]',
te) 544) 'conv4_block9_2_conv[0][0]']
conv4_block10_0_bn (BatchNorma (None, None, None, 2176 ['conv4_block9_concat[0][0]']
lization) 544)
conv4_block10_0_relu (Activati (None, None, None, 0 ['conv4_block10_0_bn[0][0]']
on) 544)
conv4_block10_1_conv (Conv2D) (None, None, None, 69632 ['conv4_block10_0_relu[0][0]']
128)
conv4_block10_1_bn (BatchNorma (None, None, None, 512 ['conv4_block10_1_conv[0][0]']
lization) 128)
conv4_block10_1_relu (Activati (None, None, None, 0 ['conv4_block10_1_bn[0][0]']
on) 128)
conv4_block10_2_conv (Conv2D) (None, None, None, 36864 ['conv4_block10_1_relu[0][0]']
32)
conv4_block10_concat (Concaten (None, None, None, 0 ['conv4_block9_concat[0][0]',
ate) 576) 'conv4_block10_2_conv[0][0]']
conv4_block11_0_bn (BatchNorma (None, None, None, 2304 ['conv4_block10_concat[0][0]']
lization) 576)
conv4_block11_0_relu (Activati (None, None, None, 0 ['conv4_block11_0_bn[0][0]']
on) 576)
conv4_block11_1_conv (Conv2D) (None, None, None, 73728 ['conv4_block11_0_relu[0][0]']
128)
conv4_block11_1_bn (BatchNorma (None, None, None, 512 ['conv4_block11_1_conv[0][0]']
lization) 128)
conv4_block11_1_relu (Activati (None, None, None, 0 ['conv4_block11_1_bn[0][0]']
on) 128)
conv4_block11_2_conv (Conv2D) (None, None, None, 36864 ['conv4_block11_1_relu[0][0]']
32)
conv4_block11_concat (Concaten (None, None, None, 0 ['conv4_block10_concat[0][0]',
ate) 608) 'conv4_block11_2_conv[0][0]']
conv4_block12_0_bn (BatchNorma (None, None, None, 2432 ['conv4_block11_concat[0][0]']
lization) 608)
conv4_block12_0_relu (Activati (None, None, None, 0 ['conv4_block12_0_bn[0][0]']
on) 608)
conv4_block12_1_conv (Conv2D) (None, None, None, 77824 ['conv4_block12_0_relu[0][0]']
128)
conv4_block12_1_bn (BatchNorma (None, None, None, 512 ['conv4_block12_1_conv[0][0]']
lization) 128)
conv4_block12_1_relu (Activati (None, None, None, 0 ['conv4_block12_1_bn[0][0]']
on) 128)
conv4_block12_2_conv (Conv2D) (None, None, None, 36864 ['conv4_block12_1_relu[0][0]']
32)
conv4_block12_concat (Concaten (None, None, None, 0 ['conv4_block11_concat[0][0]',
ate) 640) 'conv4_block12_2_conv[0][0]']
conv4_block13_0_bn (BatchNorma (None, None, None, 2560 ['conv4_block12_concat[0][0]']
lization) 640)
conv4_block13_0_relu (Activati (None, None, None, 0 ['conv4_block13_0_bn[0][0]']
on) 640)
conv4_block13_1_conv (Conv2D) (None, None, None, 81920 ['conv4_block13_0_relu[0][0]']
128)
conv4_block13_1_bn (BatchNorma (None, None, None, 512 ['conv4_block13_1_conv[0][0]']
lization) 128)
conv4_block13_1_relu (Activati (None, None, None, 0 ['conv4_block13_1_bn[0][0]']
on) 128)
conv4_block13_2_conv (Conv2D) (None, None, None, 36864 ['conv4_block13_1_relu[0][0]']
32)
conv4_block13_concat (Concaten (None, None, None, 0 ['conv4_block12_concat[0][0]',
ate) 672) 'conv4_block13_2_conv[0][0]']
conv4_block14_0_bn (BatchNorma (None, None, None, 2688 ['conv4_block13_concat[0][0]']
lization) 672)
conv4_block14_0_relu (Activati (None, None, None, 0 ['conv4_block14_0_bn[0][0]']
on) 672)
conv4_block14_1_conv (Conv2D) (None, None, None, 86016 ['conv4_block14_0_relu[0][0]']
128)
conv4_block14_1_bn (BatchNorma (None, None, None, 512 ['conv4_block14_1_conv[0][0]']
lization) 128)
conv4_block14_1_relu (Activati (None, None, None, 0 ['conv4_block14_1_bn[0][0]']
on) 128)
conv4_block14_2_conv (Conv2D) (None, None, None, 36864 ['conv4_block14_1_relu[0][0]']
32)
conv4_block14_concat (Concaten (None, None, None, 0 ['conv4_block13_concat[0][0]',
ate) 704) 'conv4_block14_2_conv[0][0]']
conv4_block15_0_bn (BatchNorma (None, None, None, 2816 ['conv4_block14_concat[0][0]']
lization) 704)
conv4_block15_0_relu (Activati (None, None, None, 0 ['conv4_block15_0_bn[0][0]']
on) 704)
conv4_block15_1_conv (Conv2D) (None, None, None, 90112 ['conv4_block15_0_relu[0][0]']
128)
conv4_block15_1_bn (BatchNorma (None, None, None, 512 ['conv4_block15_1_conv[0][0]']
lization) 128)
conv4_block15_1_relu (Activati (None, None, None, 0 ['conv4_block15_1_bn[0][0]']
on) 128)
conv4_block15_2_conv (Conv2D) (None, None, None, 36864 ['conv4_block15_1_relu[0][0]']
32)
conv4_block15_concat (Concaten (None, None, None, 0 ['conv4_block14_concat[0][0]',
ate) 736) 'conv4_block15_2_conv[0][0]']
conv4_block16_0_bn (BatchNorma (None, None, None, 2944 ['conv4_block15_concat[0][0]']
lization) 736)
conv4_block16_0_relu (Activati (None, None, None, 0 ['conv4_block16_0_bn[0][0]']
on) 736)
conv4_block16_1_conv (Conv2D) (None, None, None, 94208 ['conv4_block16_0_relu[0][0]']
128)
conv4_block16_1_bn (BatchNorma (None, None, None, 512 ['conv4_block16_1_conv[0][0]']
lization) 128)
conv4_block16_1_relu (Activati (None, None, None, 0 ['conv4_block16_1_bn[0][0]']
on) 128)
conv4_block16_2_conv (Conv2D) (None, None, None, 36864 ['conv4_block16_1_relu[0][0]']
32)
conv4_block16_concat (Concaten (None, None, None, 0 ['conv4_block15_concat[0][0]',
ate) 768) 'conv4_block16_2_conv[0][0]']
conv4_block17_0_bn (BatchNorma (None, None, None, 3072 ['conv4_block16_concat[0][0]']
lization) 768)
conv4_block17_0_relu (Activati (None, None, None, 0 ['conv4_block17_0_bn[0][0]']
on) 768)
conv4_block17_1_conv (Conv2D) (None, None, None, 98304 ['conv4_block17_0_relu[0][0]']
128)
conv4_block17_1_bn (BatchNorma (None, None, None, 512 ['conv4_block17_1_conv[0][0]']
lization) 128)
conv4_block17_1_relu (Activati (None, None, None, 0 ['conv4_block17_1_bn[0][0]']
on) 128)
conv4_block17_2_conv (Conv2D) (None, None, None, 36864 ['conv4_block17_1_relu[0][0]']
32)
conv4_block17_concat (Concaten (None, None, None, 0 ['conv4_block16_concat[0][0]',
ate) 800) 'conv4_block17_2_conv[0][0]']
conv4_block18_0_bn (BatchNorma (None, None, None, 3200 ['conv4_block17_concat[0][0]']
lization) 800)
conv4_block18_0_relu (Activati (None, None, None, 0 ['conv4_block18_0_bn[0][0]']
on) 800)
conv4_block18_1_conv (Conv2D) (None, None, None, 102400 ['conv4_block18_0_relu[0][0]']
128)
conv4_block18_1_bn (BatchNorma (None, None, None, 512 ['conv4_block18_1_conv[0][0]']
lization) 128)
conv4_block18_1_relu (Activati (None, None, None, 0 ['conv4_block18_1_bn[0][0]']
on) 128)
conv4_block18_2_conv (Conv2D) (None, None, None, 36864 ['conv4_block18_1_relu[0][0]']
32)
conv4_block18_concat (Concaten (None, None, None, 0 ['conv4_block17_concat[0][0]',
ate) 832) 'conv4_block18_2_conv[0][0]']
conv4_block19_0_bn (BatchNorma (None, None, None, 3328 ['conv4_block18_concat[0][0]']
lization) 832)
conv4_block19_0_relu (Activati (None, None, None, 0 ['conv4_block19_0_bn[0][0]']
on) 832)
conv4_block19_1_conv (Conv2D) (None, None, None, 106496 ['conv4_block19_0_relu[0][0]']
128)
conv4_block19_1_bn (BatchNorma (None, None, None, 512 ['conv4_block19_1_conv[0][0]']
lization) 128)
conv4_block19_1_relu (Activati (None, None, None, 0 ['conv4_block19_1_bn[0][0]']
on) 128)
conv4_block19_2_conv (Conv2D) (None, None, None, 36864 ['conv4_block19_1_relu[0][0]']
32)
conv4_block19_concat (Concaten (None, None, None, 0 ['conv4_block18_concat[0][0]',
ate) 864) 'conv4_block19_2_conv[0][0]']
conv4_block20_0_bn (BatchNorma (None, None, None, 3456 ['conv4_block19_concat[0][0]']
lization) 864)
conv4_block20_0_relu (Activati (None, None, None, 0 ['conv4_block20_0_bn[0][0]']
on) 864)
conv4_block20_1_conv (Conv2D) (None, None, None, 110592 ['conv4_block20_0_relu[0][0]']
128)
conv4_block20_1_bn (BatchNorma (None, None, None, 512 ['conv4_block20_1_conv[0][0]']
lization) 128)
conv4_block20_1_relu (Activati (None, None, None, 0 ['conv4_block20_1_bn[0][0]']
on) 128)
conv4_block20_2_conv (Conv2D) (None, None, None, 36864 ['conv4_block20_1_relu[0][0]']
32)
conv4_block20_concat (Concaten (None, None, None, 0 ['conv4_block19_concat[0][0]',
ate) 896) 'conv4_block20_2_conv[0][0]']
conv4_block21_0_bn (BatchNorma (None, None, None, 3584 ['conv4_block20_concat[0][0]']
lization) 896)
conv4_block21_0_relu (Activati (None, None, None, 0 ['conv4_block21_0_bn[0][0]']
on) 896)
conv4_block21_1_conv (Conv2D) (None, None, None, 114688 ['conv4_block21_0_relu[0][0]']
128)
conv4_block21_1_bn (BatchNorma (None, None, None, 512 ['conv4_block21_1_conv[0][0]']
lization) 128)
conv4_block21_1_relu (Activati (None, None, None, 0 ['conv4_block21_1_bn[0][0]']
on) 128)
conv4_block21_2_conv (Conv2D) (None, None, None, 36864 ['conv4_block21_1_relu[0][0]']
32)
conv4_block21_concat (Concaten (None, None, None, 0 ['conv4_block20_concat[0][0]',
ate) 928) 'conv4_block21_2_conv[0][0]']
conv4_block22_0_bn (BatchNorma (None, None, None, 3712 ['conv4_block21_concat[0][0]']
lization) 928)
conv4_block22_0_relu (Activati (None, None, None, 0 ['conv4_block22_0_bn[0][0]']
on) 928)
conv4_block22_1_conv (Conv2D) (None, None, None, 118784 ['conv4_block22_0_relu[0][0]']
128)
conv4_block22_1_bn (BatchNorma (None, None, None, 512 ['conv4_block22_1_conv[0][0]']
lization) 128)
conv4_block22_1_relu (Activati (None, None, None, 0 ['conv4_block22_1_bn[0][0]']
on) 128)
conv4_block22_2_conv (Conv2D) (None, None, None, 36864 ['conv4_block22_1_relu[0][0]']
32)
conv4_block22_concat (Concaten (None, None, None, 0 ['conv4_block21_concat[0][0]',
ate) 960) 'conv4_block22_2_conv[0][0]']
conv4_block23_0_bn (BatchNorma (None, None, None, 3840 ['conv4_block22_concat[0][0]']
lization) 960)
conv4_block23_0_relu (Activati (None, None, None, 0 ['conv4_block23_0_bn[0][0]']
on) 960)
conv4_block23_1_conv (Conv2D) (None, None, None, 122880 ['conv4_block23_0_relu[0][0]']
128)
conv4_block23_1_bn (BatchNorma (None, None, None, 512 ['conv4_block23_1_conv[0][0]']
lization) 128)
conv4_block23_1_relu (Activati (None, None, None, 0 ['conv4_block23_1_bn[0][0]']
on) 128)
conv4_block23_2_conv (Conv2D) (None, None, None, 36864 ['conv4_block23_1_relu[0][0]']
32)
conv4_block23_concat (Concaten (None, None, None, 0 ['conv4_block22_concat[0][0]',
ate) 992) 'conv4_block23_2_conv[0][0]']
conv4_block24_0_bn (BatchNorma (None, None, None, 3968 ['conv4_block23_concat[0][0]']
lization) 992)
conv4_block24_0_relu (Activati (None, None, None, 0 ['conv4_block24_0_bn[0][0]']
on) 992)
conv4_block24_1_conv (Conv2D) (None, None, None, 126976 ['conv4_block24_0_relu[0][0]']
128)
conv4_block24_1_bn (BatchNorma (None, None, None, 512 ['conv4_block24_1_conv[0][0]']
lization) 128)
conv4_block24_1_relu (Activati (None, None, None, 0 ['conv4_block24_1_bn[0][0]']
on) 128)
conv4_block24_2_conv (Conv2D) (None, None, None, 36864 ['conv4_block24_1_relu[0][0]']
32)
conv4_block24_concat (Concaten (None, None, None, 0 ['conv4_block23_concat[0][0]',
ate) 1024) 'conv4_block24_2_conv[0][0]']
pool4_bn (BatchNormalization) (None, None, None, 4096 ['conv4_block24_concat[0][0]']
1024)
pool4_relu (Activation) (None, None, None, 0 ['pool4_bn[0][0]']
1024)
pool4_conv (Conv2D) (None, None, None, 524288 ['pool4_relu[0][0]']
512)
pool4_pool (AveragePooling2D) (None, None, None, 0 ['pool4_conv[0][0]']
512)
conv5_block1_0_bn (BatchNormal (None, None, None, 2048 ['pool4_pool[0][0]']
ization) 512)
conv5_block1_0_relu (Activatio (None, None, None, 0 ['conv5_block1_0_bn[0][0]']
n) 512)
conv5_block1_1_conv (Conv2D) (None, None, None, 65536 ['conv5_block1_0_relu[0][0]']
128)
conv5_block1_1_bn (BatchNormal (None, None, None, 512 ['conv5_block1_1_conv[0][0]']
ization) 128)
conv5_block1_1_relu (Activatio (None, None, None, 0 ['conv5_block1_1_bn[0][0]']
n) 128)
conv5_block1_2_conv (Conv2D) (None, None, None, 36864 ['conv5_block1_1_relu[0][0]']
32)
conv5_block1_concat (Concatena (None, None, None, 0 ['pool4_pool[0][0]',
te) 544) 'conv5_block1_2_conv[0][0]']
conv5_block2_0_bn (BatchNormal (None, None, None, 2176 ['conv5_block1_concat[0][0]']
ization) 544)
conv5_block2_0_relu (Activatio (None, None, None, 0 ['conv5_block2_0_bn[0][0]']
n) 544)
conv5_block2_1_conv (Conv2D) (None, None, None, 69632 ['conv5_block2_0_relu[0][0]']
128)
conv5_block2_1_bn (BatchNormal (None, None, None, 512 ['conv5_block2_1_conv[0][0]']
ization) 128)
conv5_block2_1_relu (Activatio (None, None, None, 0 ['conv5_block2_1_bn[0][0]']
n) 128)
conv5_block2_2_conv (Conv2D) (None, None, None, 36864 ['conv5_block2_1_relu[0][0]']
32)
conv5_block2_concat (Concatena (None, None, None, 0 ['conv5_block1_concat[0][0]',
te) 576) 'conv5_block2_2_conv[0][0]']
conv5_block3_0_bn (BatchNormal (None, None, None, 2304 ['conv5_block2_concat[0][0]']
ization) 576)
conv5_block3_0_relu (Activatio (None, None, None, 0 ['conv5_block3_0_bn[0][0]']
n) 576)
conv5_block3_1_conv (Conv2D) (None, None, None, 73728 ['conv5_block3_0_relu[0][0]']
128)
conv5_block3_1_bn (BatchNormal (None, None, None, 512 ['conv5_block3_1_conv[0][0]']
ization) 128)
conv5_block3_1_relu (Activatio (None, None, None, 0 ['conv5_block3_1_bn[0][0]']
n) 128)
conv5_block3_2_conv (Conv2D) (None, None, None, 36864 ['conv5_block3_1_relu[0][0]']
32)
conv5_block3_concat (Concatena (None, None, None, 0 ['conv5_block2_concat[0][0]',
te) 608) 'conv5_block3_2_conv[0][0]']
conv5_block4_0_bn (BatchNormal (None, None, None, 2432 ['conv5_block3_concat[0][0]']
ization) 608)
conv5_block4_0_relu (Activatio (None, None, None, 0 ['conv5_block4_0_bn[0][0]']
n) 608)
conv5_block4_1_conv (Conv2D) (None, None, None, 77824 ['conv5_block4_0_relu[0][0]']
128)
conv5_block4_1_bn (BatchNormal (None, None, None, 512 ['conv5_block4_1_conv[0][0]']
ization) 128)
conv5_block4_1_relu (Activatio (None, None, None, 0 ['conv5_block4_1_bn[0][0]']
n) 128)
conv5_block4_2_conv (Conv2D) (None, None, None, 36864 ['conv5_block4_1_relu[0][0]']
32)
conv5_block4_concat (Concatena (None, None, None, 0 ['conv5_block3_concat[0][0]',
te) 640) 'conv5_block4_2_conv[0][0]']
conv5_block5_0_bn (BatchNormal (None, None, None, 2560 ['conv5_block4_concat[0][0]']
ization) 640)
conv5_block5_0_relu (Activatio (None, None, None, 0 ['conv5_block5_0_bn[0][0]']
n) 640)
conv5_block5_1_conv (Conv2D) (None, None, None, 81920 ['conv5_block5_0_relu[0][0]']
128)
conv5_block5_1_bn (BatchNormal (None, None, None, 512 ['conv5_block5_1_conv[0][0]']
ization) 128)
conv5_block5_1_relu (Activatio (None, None, None, 0 ['conv5_block5_1_bn[0][0]']
n) 128)
conv5_block5_2_conv (Conv2D) (None, None, None, 36864 ['conv5_block5_1_relu[0][0]']
32)
conv5_block5_concat (Concatena (None, None, None, 0 ['conv5_block4_concat[0][0]',
te) 672) 'conv5_block5_2_conv[0][0]']
conv5_block6_0_bn (BatchNormal (None, None, None, 2688 ['conv5_block5_concat[0][0]']
ization) 672)
conv5_block6_0_relu (Activatio (None, None, None, 0 ['conv5_block6_0_bn[0][0]']
n) 672)
conv5_block6_1_conv (Conv2D) (None, None, None, 86016 ['conv5_block6_0_relu[0][0]']
128)
conv5_block6_1_bn (BatchNormal (None, None, None, 512 ['conv5_block6_1_conv[0][0]']
ization) 128)
conv5_block6_1_relu (Activatio (None, None, None, 0 ['conv5_block6_1_bn[0][0]']
n) 128)
conv5_block6_2_conv (Conv2D) (None, None, None, 36864 ['conv5_block6_1_relu[0][0]']
32)
conv5_block6_concat (Concatena (None, None, None, 0 ['conv5_block5_concat[0][0]',
te) 704) 'conv5_block6_2_conv[0][0]']
conv5_block7_0_bn (BatchNormal (None, None, None, 2816 ['conv5_block6_concat[0][0]']
ization) 704)
conv5_block7_0_relu (Activatio (None, None, None, 0 ['conv5_block7_0_bn[0][0]']
n) 704)
conv5_block7_1_conv (Conv2D) (None, None, None, 90112 ['conv5_block7_0_relu[0][0]']
128)
conv5_block7_1_bn (BatchNormal (None, None, None, 512 ['conv5_block7_1_conv[0][0]']
ization) 128)
conv5_block7_1_relu (Activatio (None, None, None, 0 ['conv5_block7_1_bn[0][0]']
n) 128)
conv5_block7_2_conv (Conv2D) (None, None, None, 36864 ['conv5_block7_1_relu[0][0]']
32)
conv5_block7_concat (Concatena (None, None, None, 0 ['conv5_block6_concat[0][0]',
te) 736) 'conv5_block7_2_conv[0][0]']
conv5_block8_0_bn (BatchNormal (None, None, None, 2944 ['conv5_block7_concat[0][0]']
ization) 736)
conv5_block8_0_relu (Activatio (None, None, None, 0 ['conv5_block8_0_bn[0][0]']
n) 736)
conv5_block8_1_conv (Conv2D) (None, None, None, 94208 ['conv5_block8_0_relu[0][0]']
128)
conv5_block8_1_bn (BatchNormal (None, None, None, 512 ['conv5_block8_1_conv[0][0]']
ization) 128)
conv5_block8_1_relu (Activatio (None, None, None, 0 ['conv5_block8_1_bn[0][0]']
n) 128)
conv5_block8_2_conv (Conv2D) (None, None, None, 36864 ['conv5_block8_1_relu[0][0]']
32)
conv5_block8_concat (Concatena (None, None, None, 0 ['conv5_block7_concat[0][0]',
te) 768) 'conv5_block8_2_conv[0][0]']
conv5_block9_0_bn (BatchNormal (None, None, None, 3072 ['conv5_block8_concat[0][0]']
ization) 768)
conv5_block9_0_relu (Activatio (None, None, None, 0 ['conv5_block9_0_bn[0][0]']
n) 768)
conv5_block9_1_conv (Conv2D) (None, None, None, 98304 ['conv5_block9_0_relu[0][0]']
128)
conv5_block9_1_bn (BatchNormal (None, None, None, 512 ['conv5_block9_1_conv[0][0]']
ization) 128)
conv5_block9_1_relu (Activatio (None, None, None, 0 ['conv5_block9_1_bn[0][0]']
n) 128)
conv5_block9_2_conv (Conv2D) (None, None, None, 36864 ['conv5_block9_1_relu[0][0]']
32)
conv5_block9_concat (Concatena (None, None, None, 0 ['conv5_block8_concat[0][0]',
te) 800) 'conv5_block9_2_conv[0][0]']
conv5_block10_0_bn (BatchNorma (None, None, None, 3200 ['conv5_block9_concat[0][0]']
lization) 800)
conv5_block10_0_relu (Activati (None, None, None, 0 ['conv5_block10_0_bn[0][0]']
on) 800)
conv5_block10_1_conv (Conv2D) (None, None, None, 102400 ['conv5_block10_0_relu[0][0]']
128)
conv5_block10_1_bn (BatchNorma (None, None, None, 512 ['conv5_block10_1_conv[0][0]']
lization) 128)
conv5_block10_1_relu (Activati (None, None, None, 0 ['conv5_block10_1_bn[0][0]']
on) 128)
conv5_block10_2_conv (Conv2D) (None, None, None, 36864 ['conv5_block10_1_relu[0][0]']
32)
conv5_block10_concat (Concaten (None, None, None, 0 ['conv5_block9_concat[0][0]',
ate) 832) 'conv5_block10_2_conv[0][0]']
conv5_block11_0_bn (BatchNorma (None, None, None, 3328 ['conv5_block10_concat[0][0]']
lization) 832)
conv5_block11_0_relu (Activati (None, None, None, 0 ['conv5_block11_0_bn[0][0]']
on) 832)
conv5_block11_1_conv (Conv2D) (None, None, None, 106496 ['conv5_block11_0_relu[0][0]']
128)
conv5_block11_1_bn (BatchNorma (None, None, None, 512 ['conv5_block11_1_conv[0][0]']
lization) 128)
conv5_block11_1_relu (Activati (None, None, None, 0 ['conv5_block11_1_bn[0][0]']
on) 128)
conv5_block11_2_conv (Conv2D) (None, None, None, 36864 ['conv5_block11_1_relu[0][0]']
32)
conv5_block11_concat (Concaten (None, None, None, 0 ['conv5_block10_concat[0][0]',
ate) 864) 'conv5_block11_2_conv[0][0]']
conv5_block12_0_bn (BatchNorma (None, None, None, 3456 ['conv5_block11_concat[0][0]']
lization) 864)
conv5_block12_0_relu (Activati (None, None, None, 0 ['conv5_block12_0_bn[0][0]']
on) 864)
conv5_block12_1_conv (Conv2D) (None, None, None, 110592 ['conv5_block12_0_relu[0][0]']
128)
conv5_block12_1_bn (BatchNorma (None, None, None, 512 ['conv5_block12_1_conv[0][0]']
lization) 128)
conv5_block12_1_relu (Activati (None, None, None, 0 ['conv5_block12_1_bn[0][0]']
on) 128)
conv5_block12_2_conv (Conv2D) (None, None, None, 36864 ['conv5_block12_1_relu[0][0]']
32)
conv5_block12_concat (Concaten (None, None, None, 0 ['conv5_block11_concat[0][0]',
ate) 896) 'conv5_block12_2_conv[0][0]']
conv5_block13_0_bn (BatchNorma (None, None, None, 3584 ['conv5_block12_concat[0][0]']
lization) 896)
conv5_block13_0_relu (Activati (None, None, None, 0 ['conv5_block13_0_bn[0][0]']
on) 896)
conv5_block13_1_conv (Conv2D) (None, None, None, 114688 ['conv5_block13_0_relu[0][0]']
128)
conv5_block13_1_bn (BatchNorma (None, None, None, 512 ['conv5_block13_1_conv[0][0]']
lization) 128)
conv5_block13_1_relu (Activati (None, None, None, 0 ['conv5_block13_1_bn[0][0]']
on) 128)
conv5_block13_2_conv (Conv2D) (None, None, None, 36864 ['conv5_block13_1_relu[0][0]']
32)
conv5_block13_concat (Concaten (None, None, None, 0 ['conv5_block12_concat[0][0]',
ate) 928) 'conv5_block13_2_conv[0][0]']
conv5_block14_0_bn (BatchNorma (None, None, None, 3712 ['conv5_block13_concat[0][0]']
lization) 928)
conv5_block14_0_relu (Activati (None, None, None, 0 ['conv5_block14_0_bn[0][0]']
on) 928)
conv5_block14_1_conv (Conv2D) (None, None, None, 118784 ['conv5_block14_0_relu[0][0]']
128)
conv5_block14_1_bn (BatchNorma (None, None, None, 512 ['conv5_block14_1_conv[0][0]']
lization) 128)
conv5_block14_1_relu (Activati (None, None, None, 0 ['conv5_block14_1_bn[0][0]']
on) 128)
conv5_block14_2_conv (Conv2D) (None, None, None, 36864 ['conv5_block14_1_relu[0][0]']
32)
conv5_block14_concat (Concaten (None, None, None, 0 ['conv5_block13_concat[0][0]',
ate) 960) 'conv5_block14_2_conv[0][0]']
conv5_block15_0_bn (BatchNorma (None, None, None, 3840 ['conv5_block14_concat[0][0]']
lization) 960)
conv5_block15_0_relu (Activati (None, None, None, 0 ['conv5_block15_0_bn[0][0]']
on) 960)
conv5_block15_1_conv (Conv2D) (None, None, None, 122880 ['conv5_block15_0_relu[0][0]']
128)
conv5_block15_1_bn (BatchNorma (None, None, None, 512 ['conv5_block15_1_conv[0][0]']
lization) 128)
conv5_block15_1_relu (Activati (None, None, None, 0 ['conv5_block15_1_bn[0][0]']
on) 128)
conv5_block15_2_conv (Conv2D) (None, None, None, 36864 ['conv5_block15_1_relu[0][0]']
32)
conv5_block15_concat (Concaten (None, None, None, 0 ['conv5_block14_concat[0][0]',
ate) 992) 'conv5_block15_2_conv[0][0]']
conv5_block16_0_bn (BatchNorma (None, None, None, 3968 ['conv5_block15_concat[0][0]']
lization) 992)
conv5_block16_0_relu (Activati (None, None, None, 0 ['conv5_block16_0_bn[0][0]']
on) 992)
conv5_block16_1_conv (Conv2D) (None, None, None, 126976 ['conv5_block16_0_relu[0][0]']
128)
conv5_block16_1_bn (BatchNorma (None, None, None, 512 ['conv5_block16_1_conv[0][0]']
lization) 128)
conv5_block16_1_relu (Activati (None, None, None, 0 ['conv5_block16_1_bn[0][0]']
on) 128)
conv5_block16_2_conv (Conv2D) (None, None, None, 36864 ['conv5_block16_1_relu[0][0]']
32)
conv5_block16_concat (Concaten (None, None, None, 0 ['conv5_block15_concat[0][0]',
ate) 1024) 'conv5_block16_2_conv[0][0]']
bn (BatchNormalization) (None, None, None, 4096 ['conv5_block16_concat[0][0]']
1024)
relu (Activation) (None, None, None, 0 ['bn[0][0]']
1024)
global_average_pooling2d (Glob (None, 1024) 0 ['relu[0][0]']
alAveragePooling2D)
dense (Dense) (None, 14) 14350 ['global_average_pooling2d[0][0]'
]
==================================================================================================
Total params: 7,051,854
Trainable params: 6,968,206
Non-trainable params: 83,648
__________________________________________________________________________________________________
1.3. Predict vals using DenseNet Model
model_obj.model_2d_dir
'./models/nih/densenet.hdf5'
predicted_vals = model_obj.predict_2d(**kwargs)
Received kwargs: {'test_generator': <keras.preprocessing.image.DataFrameIterator object at 0x15ee0a390>, 'labels': ['Cardiomegaly', 'Emphysema', 'Effusion', 'Hernia', 'Infiltration', 'Mass', 'Nodule', 'Atelectasis', 'Pneumothorax', 'Pleural_Thickening', 'Pneumonia', 'Fibrosis', 'Edema', 'Consolidation'], 'pos_weights': array([0.98 , 0.987, 0.872, 0.998, 0.825, 0.955, 0.946, 0.894, 0.962,
0.979, 0.99 , 0.986, 0.984, 0.967]), 'neg_weights': array([0.02 , 0.013, 0.128, 0.002, 0.175, 0.045, 0.054, 0.106, 0.038,
0.021, 0.01 , 0.014, 0.016, 0.033])}
/Users/abhilashdhal/AIMedicine/src/utils/models.py:233: UserWarning: `Model.predict_generator` is deprecated and will be removed in a future version. Please use `Model.predict`, which supports generators.
predicted_vals = model.predict_generator(test_generator, steps=len(test_generator))
2024-04-06 13:21:59.504202: I tensorflow/core/common_runtime/executor.cc:1197] [/device:CPU:0] (DEBUG INFO) Executor start aborting (this does not indicate an error and you can ignore this message): INVALID_ARGUMENT: You must feed a value for placeholder tensor 'Placeholder/_0' with dtype int32
[[{{node Placeholder/_0}}]]
2024-04-06 13:21:59.533483: W tensorflow/tsl/platform/profile_utils/cpu_utils.cc:128] Failed to get CPU frequency: 0 Hz
2. Evaluation and Interpretation
2.1. Model Prediction Performance
import random random.seed(a=None, version=2)
auc_rocs = model_obj.cs.get_roc_curve(labels, predicted_vals, test_generator)
2.2. Model Interpretation
One of the challenges of using deep learning in medicine is that the complex architecture used for neural networks makes them much harder to interpret compared to traditional machine learning models (e.g. linear models). One of the most common approaches aimed at increasing the interpretability of models for computer vision tasks is to use Class Activation Maps (CAM).
Class activation maps are useful for understanding where the model is "looking" when classifying an image.
In this section we will use a GradCAM's technique to produce a heatmap highlighting the important regions in the image for predicting the pathological condition. This is done by extracting the gradients of each predicted class, flowing into our model's final convolutional layer.
Look at the compute_gradcam which has been provided for you in to see how this is done with the Keras framework.
It is worth mentioning that GradCAM does not provide a full explanation of the reasoning for each classification probability.
However, it is still a useful tool for "debugging" our model and augmenting our prediction so that an expert could validate that a prediction is indeed due to the model focusing on the right regions of the image.
# only show the labels with top 4 AUC labels_to_show = np.take(labels, np.argsort(auc_rocs)[::-1])[:4] model_obj.cs.compute_gradcam(model_2d, img='00008270_015.png', image_dir=DATA_DIR_2D, df=train_df, labels=labels, selected_labels=labels_to_show, W = 320, H=320)
1/1 [==============================] - 1s 766ms/step
Loading original image
Generating gradcam for class Cardiomegaly
Generating gradcam for class Mass
Generating gradcam for class Pneumothorax
Generating gradcam for class Edema
img_ex = '00011355_002.png' model_obj.cs.compute_gradcam(model_2d, img=img_ex, image_dir=DATA_DIR_2D, df=train_df, labels=labels, selected_labels=labels_to_show, W = 320, H=320)
1/1 [==============================] - 0s 102ms/step
Loading original image
Generating gradcam for class Cardiomegaly
Generating gradcam for class Mass
Generating gradcam for class Pneumothorax
Generating gradcam for class Edema
