ImageGen = ImageDataGenerator(fill_mode='nearest',
horizontal_flip=True,
rescale=None,
preprocessing_function=preprocess_input,
data_format="channels_last",
validation_split=0.1
)
class_csv_path = 'groupings-csv/' + class_name + '_Imagenet.csv'
df_classes = pd.read_csv(class_csv_path, usecols=['wnid'])
classes = sorted([i for i in df_classes['wnid']])
good_train_generator, steps = create_good_generator(ImageGen,
imagenet_train,
batch_size=bs,
target_size = (img_rows, img_cols),
class_mode='sparse',
subset= 'training',
classes=classes)
good_validation_generator, steps_val = create_good_generator(ImageGen,
imagenet_train,
batch_size=bs,
target_size = (img_rows, img_cols),
class_mode='sparse',
subset= 'validation',
classes=classes)
#Build the model
def get_ATT_model(str_layer_index, class_name):
model = load_model(model_file + model_path,
custom_objects={
'SinglyConnected': SinglyConnected,
'CustomModel': KO.CustomModel
})
cross_df.loc[i] = 0
cross_df['model_name'].iloc[i] = model_path[:-3]
for class_name in ctgry_list:
if class_name != ctgry:
class_csv_path = 'groupings-csv/' + class_name + '_Imagenet.csv'
df_classes = pd.read_csv(class_csv_path, usecols=['wnid'])
classes = sorted([i for i in df_classes['wnid']])
in_context_generator, in_context_steps = create_good_generator(
ImageGen,
imagenet_test,
batch_size=bs,
target_size=(img_rows, img_cols),
class_mode='sparse',
AlextNetAug=False,
classes=classes)
ic_loss, ic_acc = model.evaluate_generator(in_context_generator,
in_context_steps,
verbose=1)
cross_df[class_name].iloc[i] = ic_acc
print(cross_df)
save_path = 'single_att_results/cross_evaluate_' + ctgry + '.csv'
cross_df.to_csv(save_path)