def evaluate_endtoend_identification(training = False, num_folds = 10, fcn_filters = 128, augm = False, augm_type=AugmentationType.RND):
model_name = generate_model_name(representation_type=RepresentationType.EE, fcn_filters = fcn_filters, augm = augm, augm_type = augm_type)
if training == True:
df_idnet = pd.read_csv("input_csv_gait/idnet_raw.csv", header = None)
if augm == True:
df_idnet = get_augmented_dataset(df_idnet, augm_type)
df_idnet = normalize_rows( df_idnet, st.NormalizationType.ZSCORE)
train_model(df_idnet, model_name, fcn_filters = fcn_filters, representation_learning=True)
df1 = pd.read_csv("input_csv_gait/zju_session1_frames_raw.csv", header = None)
df2 = pd.read_csv("input_csv_gait/zju_session2_frames_raw.csv", header = None)
df1 = normalize_rows( df1, st.NormalizationType.ZSCORE)
df2 = normalize_rows( df2, st.NormalizationType.ZSCORE)
features1 = get_model_output_features( df1, model_name)
features2 = get_model_output_features( df2, model_name)
print("\nSession 1")
evaluate_identification_CV(features1, num_folds = num_folds)
print("\nSession 2")
evaluate_identification_CV(features2, num_folds = num_folds)
print("\nCross-Session")
def evaluate_endtoend_identification(training=False,
augm=False,
augm_type=AugmentationType.RND):
if augm == True:
if augm_type == AugmentationType.RND:
model_name = "gait_fcn_rnd.h5"
else:
model_name = "gait_fcn_cshift.h5"
else:
model_name = "gait_fcn.h5"
if training == True:
df_idnet = pd.read_csv("input_csv/idnet.csv", header=None)
if augm == True:
df_idnet = get_augmented_dataset(df_idnet, augm_type)
df_idnet = normalize_rows(df_idnet, st.NormalizationType.ZSCORE)
train_model(df_idnet, model_name)
df1 = pd.read_csv("input_csv/zju_session1_frames_raw.csv", header=None)
df2 = pd.read_csv("input_csv/zju_session2_frames_raw.csv", header=None)
df1 = normalize_rows(df1, st.NormalizationType.ZSCORE)
df2 = normalize_rows(df2, st.NormalizationType.ZSCORE)
features1 = get_model_output_features(df1, model_name)
features2 = get_model_output_features(df2, model_name)
evaluate_identification_CV(features1, num_folds=10)
evaluate_identification_CV(features2, num_folds=10)
evaluate_identification_Train_Test(features1, features2)
def evaluate_autoencoder_authentication(
training=False,
augm=False,
augm_type=AugmentationType.RND,
verbose=False,
filename='results/autoencoder_scores.csv'):
if augm == True:
if augm_type == AugmentationType.RND:
model_name = "gait_autoencoder_fcn_rnd.h5"
else:
model_name = "gait_autoencoder_fcn_cshift.h5"
else:
model_name = "gait_autoencoder_fcn.h5"
if training == True:
df_idnet = pd.read_csv("input_csv/idnet.csv", header=None)
df_idnet = normalize_rows(df_idnet, st.NormalizationType.ZSCORE)
if augm == True:
df_idnet = get_augmented_dataset(df_idnet, augm_type)
train_autoencoder(df_idnet, model_name=model_name, num_epochs=100)
encoder_name = "encoder_" + model_name
df1 = pd.read_csv("input_csv/zju_session1_frames_raw.csv", header=None)
df2 = pd.read_csv("input_csv/zju_session2_frames_raw.csv", header=None)
df1 = normalize_rows(df1, st.NormalizationType.ZSCORE)
df2 = normalize_rows(df2, st.NormalizationType.ZSCORE)
features1 = get_autoencoder_output_features(df1, encoder_name)
features2 = get_autoencoder_output_features(df2, encoder_name)
auc1, eer1 = evaluate_authentication(features1, verbose)
auc2, eer2 = evaluate_authentication(features2, verbose)
auc_cross, eer_cross = evaluate_authentication_cross_day(
features1, features2, verbose)
# dictionary of lists
dict = {
'AUC1': auc1,
'EER1': eer1,
'AUC2': auc2,
'EER2': eer2,
'AUC_cross': auc_cross,
'EER_cross': eer_cross
}
df = pd.DataFrame(dict)
df.to_csv(filename, index=False)
def evaluate_autoencoder_authentication(
training=False,
fcn_filters=128,
augm=False,
augm_type=AugmentationType.RND,
verbose=False,
filename='results/autoencoder_scores.csv'):
representation_type = RepresentationType.AE
model_name = generate_model_name(representation_type,
fcn_filters=fcn_filters,
augm=augm,
augm_type=augm_type)
if training == True:
df_idnet = pd.read_csv("input_csv_gait/idnet_raw.csv", header=None)
df_idnet = normalize_rows(df_idnet, st.NormalizationType.ZSCORE)
if augm == True:
df_idnet = get_augmented_dataset(df_idnet, augm_type)
train_autoencoder(df_idnet,
model_name=model_name,
fcn_filters=fcn_filters)
encoder_name = "encoder_" + model_name
df1 = pd.read_csv("input_csv_gait/zju_session1_frames_raw.csv",
header=None)
df2 = pd.read_csv("input_csv_gait/zju_session2_frames_raw.csv",
header=None)
df1 = normalize_rows(df1, st.NormalizationType.ZSCORE)
df2 = normalize_rows(df2, st.NormalizationType.ZSCORE)
features1 = get_autoencoder_output_features(df1, encoder_name)
features2 = get_autoencoder_output_features(df2, encoder_name)
roc_data_filename = 'results/roc_session1_' + str(
representation_type.value) + '_' + str(data_type.value) + '.csv'
auc1, eer1 = evaluate_authentication(features1,
data_type,
representation_type,
verbose=verbose,
roc_data=True,
roc_data_filename=roc_data_filename)
roc_data_filename = 'results/roc_session2_' + str(
representation_type.value) + '_' + str(data_type.value) + '.csv'
auc2, eer2 = evaluate_authentication(features1,
data_type,
representation_type,
verbose=verbose,
roc_data=True,
roc_data_filename=roc_data_filename)
roc_data_filename = 'results/roc_cross_session_' + str(
representation_type.value) + '_' + str(data_type.value) + '.csv'
auc_cross, eer_cross = evaluate_authentication_cross_day(
features1,
features2,
data_type,
representation_type,
verbose=verbose,
roc_data=True,
roc_data_filename=roc_data_filename)
scores_filename = 'results/auc_eer_users_' + str(
representation_type.value) + '_' + str(data_type.value) + '.csv'
dict = {
'AUC1': auc1,
'EER1': eer1,
'AUC2': auc2,
'EER2': eer2,
'AUC_cross': auc_cross,
'EER_cross': eer_cross
}
df = pd.DataFrame(dict)
df.to_csv(scores_filename, index=False)