def __getitem__(self, index):
'Generate one batch of data'
# Generate indexes of the batch
indexes = self.indexes[index * self.batch_size:(index + 1) *
self.batch_size]
# Loading Paths & Labels
if self.label_type == 'attr':
_paths, _labels_act, _labels_dom, _labels_val = getPaths(
self.label_type, self.split_set, self.num_class)
# Find Batch list of Loading Paths
list_paths_temp = [_paths[k] for k in indexes]
list_act_temp = [_labels_act[k] for k in indexes]
list_dom_temp = [_labels_dom[k] for k in indexes]
list_val_temp = [_labels_val[k] for k in indexes]
# Generate data
data, label = self.__data_generation_attr(list_paths_temp,
list_act_temp,
list_dom_temp,
list_val_temp)
elif self.label_type == 'class':
_paths, _labels_class = getPaths(self.label_type, self.split_set,
self.num_class)
# Find Batch list of Loading Paths
list_paths_temp = [_paths[k] for k in indexes]
list_class_temp = [_labels_class[k] for k in indexes]
# Generate data
data, label = self.__data_generation_class(list_paths_temp,
list_class_temp)
return data, label
def on_epoch_end(self):
'Updates indexes after each epoch'
if self.label_type == 'attr':
_paths, _, _, _ = getPaths(self.label_type, self.split_set,
self.num_class)
elif self.label_type == 'class':
_paths, _ = getPaths(self.label_type, self.split_set,
self.num_class)
self.indexes = np.arange(len(_paths))
if self.shuffle == True:
np.random.seed(random_seed)
np.random.shuffle(self.indexes)
def on_epoch_end(self):
'Updates indexes after each epoch'
_paths, _labels = getPaths(self.label_dir, self.split_set,
self.emo_attr)
self.indexes = np.arange(len(_paths))
if self.shuffle == True:
np.random.seed(random_seed)
np.random.shuffle(self.indexes)
def createValloader(data_dirnames):
dataloader_val = thermal_loader.ThermalTestDataLoader(*utils.getPaths(data_dirnames))
val_loader = torch.utils.data.DataLoader(dataloader_val,
batch_size=1,
shuffle=False,
num_workers=opt.num_workers,
pin_memory=True,
drop_last=False)
return val_loader
def __init__(self,
root_dir,
label_dir,
batch_size,
split_set,
emo_attr,
shuffle=True):
'Initialization'
self.root_dir = root_dir
self.label_dir = label_dir
self.batch_size = batch_size
self.split_set = split_set # 'Train' or 'Validation'
self.emo_attr = emo_attr # 'Act', 'Dom' or 'Val'
self.shuffle = shuffle
# Loading Norm-Feature Parameters
self.Feat_mean = loadmat(
'./NormTerm/feat_norm_means.mat')['normal_para']
self.Feat_std = loadmat('./NormTerm/feat_norm_stds.mat')['normal_para']
# Loading Norm-Label Parameters
if emo_attr == 'Act':
self.Label_mean = loadmat(
'./NormTerm/act_norm_means.mat')['normal_para'][0][0]
self.Label_std = loadmat(
'./NormTerm/act_norm_stds.mat')['normal_para'][0][0]
elif emo_attr == 'Dom':
self.Label_mean = loadmat(
'./NormTerm/dom_norm_means.mat')['normal_para'][0][0]
self.Label_std = loadmat(
'./NormTerm/dom_norm_stds.mat')['normal_para'][0][0]
elif emo_attr == 'Val':
self.Label_mean = loadmat(
'./NormTerm/val_norm_means.mat')['normal_para'][0][0]
self.Label_std = loadmat(
'./NormTerm/val_norm_stds.mat')['normal_para'][0][0]
# Loading Data Paths/Labels
self._paths, self._labels = getPaths(label_dir, split_set, emo_attr)
self.on_epoch_end()
from keras.utils.np_utils import to_categorical from keras.applications.resnet50 import preprocess_input from keras.models import Model import queue import threading from tqdm import tqdm from networks import get_models, adversarial, null_loss, generator, discriminator from utils import producer, getPaths, scale, mean data_train = r"C:\Users\tgill\OneDrive\Documents\GD_AI\ArtGAN\wikipaintings_full\wikipaintings_train" data_test = r"C:\Users\tgill\OneDrive\Documents\GD_AI\ArtGAN\wikipaintings_full\wikipaintings_train" train_paths, y_train, classes = getPaths(data_train) test_paths, y_test, classes = getPaths(data_test) target_size = (128, 128) ls = [-np.sum(y_train == i) for i in range(25)] arg = np.argsort(ls) classement = np.argsort(arg) nb_select = 1 select = arg[:nb_select] idx_select = np.isin(y_train, select) train_paths = train_paths[idx_select] y_train = y_train[idx_select] y_train = classement[y_train] print(train_paths.shape)
num_class = args['num_class']
except:
pass
Feat_mean_All = loadmat('./NormTerm/feat_norm_means.mat')['normal_para']
Feat_std_All = loadmat('./NormTerm/feat_norm_stds.mat')['normal_para']
Label_mean_act = loadmat('./NormTerm/act_norm_means.mat')['normal_para'][0][0]
Label_std_act = loadmat('./NormTerm/act_norm_stds.mat')['normal_para'][0][0]
Label_mean_dom = loadmat('./NormTerm/dom_norm_means.mat')['normal_para'][0][0]
Label_std_dom = loadmat('./NormTerm/dom_norm_stds.mat')['normal_para'][0][0]
Label_mean_val = loadmat('./NormTerm/val_norm_means.mat')['normal_para'][0][0]
Label_std_val = loadmat('./NormTerm/val_norm_stds.mat')['normal_para'][0][0]
# Testing Task
if label_type == 'attr':
test_file_path, test_file_tar_act, test_file_tar_dom, test_file_tar_val = getPaths(
label_type, 'Test', num_class)
elif label_type == 'class':
test_file_path, test_file_tar_class = getPaths(label_type, 'Test',
num_class)
# Testing Data & Label
Test_Data = []
Test_Label_Act = []
Test_Label_Dom = []
Test_Label_Val = []
Test_Label_Class = []
for i in range(len(test_file_path)):
data = loadmat(root_dir +
test_file_path[i].replace('.wav', '.mat'))['Audio_data']
data = (data - Feat_mean_All) / Feat_std_All # Feature Normalization
data = data.reshape(-1)
conf_segnet_model.load_state_dict(checkpoint['state_dict'])
optimizer.load_state_dict(checkpoint['optimizer'])
lr_scheduler.load_state_dict(checkpoint['lr_scheduler'])
opt.epoch = checkpoint['epoch']
best_iou = checkpoint['best_iou']
print('Create validation loader daytime')
val_loader_night = createValloader([opt.testroot_night])
print('Create validation loader nighttime')
val_loader_day = createValloader([opt.testroot_day])
print('Create validation loader both')
val_loader_combined = createValloader([opt.testroot_night, opt.testroot_day])
test_stamps = getTestStamps(*utils.getPaths([opt.testroot_night, opt.testroot_day]))
print('Create training loader')
train_loader = createDataloader(test_stamps)
# Loss plot
total_loss_avgmeter_phase1 = AverageMeter()
total_loss_avgmeter_phase2 = AverageMeter()
critic_loss_avgmeter = AverageMeter()
seg_loss_avgmeter = AverageMeter()
conf_loss_avgmeter = AverageMeter()
if opt.eval is not "":
print('Starting evaluation on: %s....' % (opt.eval))
def __len__(self):
'Denotes the number of batches per epoch'
return int(
len(getPaths(self.label_dir, self.split_set, self.emo_attr)[0]) /
self.batch_size)
def __len__(self):
'Denotes the number of batches per epoch'
return int(
len(getPaths(self.label_type, self.split_set, self.num_class)[0]) /
self.batch_size)
num_class = args['num_class']
except:
pass
# Hidden Features Paths Setting
if label_type == 'attr':
root_dir = './Fusion_Features/3-attribute'
elif label_type == 'class':
if num_class == '5-class':
root_dir = './Fusion_Features/5-class'
elif num_class == '8-class':
root_dir = './Fusion_Features/8-class'
# Loading Paths & Labels
if label_type == 'class':
paths_test, labels_class_test = getPaths(label_type, split_set='Test', num_class=num_class)
elif label_type == 'attr':
# Loading Norm-Label
Label_mean_act = loadmat('./NormTerm/act_norm_means.mat')['normal_para'][0][0]
Label_std_act = loadmat('./NormTerm/act_norm_stds.mat')['normal_para'][0][0]
Label_mean_dom = loadmat('./NormTerm/dom_norm_means.mat')['normal_para'][0][0]
Label_std_dom = loadmat('./NormTerm/dom_norm_stds.mat')['normal_para'][0][0]
Label_mean_val = loadmat('./NormTerm/val_norm_means.mat')['normal_para'][0][0]
Label_std_val = loadmat('./NormTerm/val_norm_stds.mat')['normal_para'][0][0]
paths_test, labels_act_test, labels_dom_test, labels_val_test = getPaths(label_type, split_set='Test', num_class=num_class)
# Loading Hidden Features (Testing set)
X_Test = []
Y_Test_Class = []
Y_Test_Act = []
Y_Test_Dom = []
# Loading Norm-Parameters
Feat_mean = loadmat('./NormTerm/feat_norm_means.mat')['normal_para']
Feat_std = loadmat('./NormTerm/feat_norm_stds.mat')['normal_para']
if emo_attr == 'Act':
Label_mean = loadmat('./NormTerm/act_norm_means.mat')['normal_para'][0][0]
Label_std = loadmat('./NormTerm/act_norm_stds.mat')['normal_para'][0][0]
elif emo_attr == 'Dom':
Label_mean = loadmat('./NormTerm/dom_norm_means.mat')['normal_para'][0][0]
Label_std = loadmat('./NormTerm/dom_norm_stds.mat')['normal_para'][0][0]
elif emo_attr == 'Val':
Label_mean = loadmat('./NormTerm/val_norm_means.mat')['normal_para'][0][0]
Label_std = loadmat('./NormTerm/val_norm_stds.mat')['normal_para'][0][0]
# Regression Task
test_file_path, test_file_tar = getPaths(label_dir,
split_set='Test',
emo_attr=emo_attr)
#test_file_path, test_file_tar = getPaths(label_dir, split_set='Validation', emo_attr=emo_attr)
# Setting Online Prediction Model Graph (predict sentence by sentence rather than a data batch)
time_step = 62 # same as the number of frames within a chunk (i.e., m)
feat_num = 130 # number of LLDs features
if atten_type == 'GatedVec':
# LSTM Layer
inputs = Input((time_step, feat_num))
encode = LSTM(units=feat_num,
activation='tanh',
dropout=0.5,
return_sequences=True)(inputs)
encode = LSTM(units=feat_num,
"""
import numpy as np
import os
from scipy.io import loadmat, savemat
import random
from utils import getPaths
# Ignore warnings & Fix random seed
import warnings
warnings.filterwarnings("ignore")
random.seed(999)
if __name__ == '__main__':
data_root = '/media/winston/UTD-MSP/Speech_Datasets/MSP-Face/Features/OpenSmile_func_IS13ComParE/feat_mat/'
fnames, Train_Label_act, Train_Label_dom, Train_Label_val = getPaths(
label_type='attr', split_set='Train', num_class=None)
# Output normalize parameters folder based on the training set
if not os.path.isdir('./NormTerm/'):
os.makedirs('./NormTerm/')
# Acoustic-Feature Normalization based on Training Set
Train_Data = []
for i in range(len(fnames)):
data = loadmat(data_root +
fnames[i].replace('.wav', '.mat'))['Audio_data']
data = data.reshape(-1)
Train_Data.append(data)
Train_Data = np.array(Train_Data)
# Feature Normalization Parameters
except:
pass
# Hidden Features Paths Setting
if label_type == 'attr':
root_dir = './Fusion_Features/3-attribute'
elif label_type == 'class':
if num_class == '5-class':
root_dir = './Fusion_Features/5-class'
elif num_class == '8-class':
root_dir = './Fusion_Features/8-class'
# Loading Paths & Labels
if label_type == 'class':
paths_valid, labels_class_valid = getPaths(label_type,
split_set='Validation',
num_class=num_class)
paths_train, labels_class_train = getPaths(label_type,
split_set='Train',
num_class=num_class)
elif label_type == 'attr':
# Loading Norm-Label
Label_mean_act = loadmat(
'./NormTerm/act_norm_means.mat')['normal_para'][0][0]
Label_std_act = loadmat(
'./NormTerm/act_norm_stds.mat')['normal_para'][0][0]
Label_mean_dom = loadmat(
'./NormTerm/dom_norm_means.mat')['normal_para'][0][0]
Label_std_dom = loadmat(
'./NormTerm/dom_norm_stds.mat')['normal_para'][0][0]
Label_mean_val = loadmat(