def _make_data_set(root, video_paths, class_to_idx, init_shuffle,
video_index):
def _is_video_file(filename_):
return any(
filename_.endswith(extension)
for extension in VIDEO_EXTENSIONS)
if init_shuffle and not video_index:
list_shuffle(video_paths) # shuffle
videos = list()
frames_per_video = list()
frames_per_class = [0] * len(class_to_idx)
frames_counter = 0
for filename in tqdm(video_paths, ncols=80):
class_ = filename.split('/')[0]
data_path = join(root, filename)
if _is_video_file(data_path):
video_meta = ffprobe(data_path)
start_idx = frames_counter
frames = int(video_meta['video'].get('@nb_frames'))
frames_per_video.append(frames)
frames_per_class[class_to_idx[class_]] += frames
frames_counter += frames
item = ((frames_counter - 1, start_idx),
(filename, class_to_idx[class_]))
videos.append(item)
sleep(0.5) # allows for progress bar completion
return videos, frames_counter, frames_per_video, frames_per_class
def source_data_expand(data_info_file, img_dir, rand_val=False, gender=None):
"""Read sample information, get split train- and test-dataset."""
# config sample number per class
train_sample_num = 1000
val_sample_num = 1000
# read sample info
all_info = open(data_info_file).readlines()
all_info.pop(0)
all_info = [line.strip().split(',') for line in all_info]
# select samples of specific gender
if gender=='m':
all_info = [line for line in all_info if int(line[1][16])%2==1]
elif gender=='f':
all_info = [line for line in all_info if int(line[1][16])%2==0]
else:
pass
imgs = [os.path.join(img_dir, line[2]) for line in all_info]
vals = [float(line[3]) for line in all_info]
ages = []
for line in all_info:
birth_year = int(line[1][6:10])
birth_month = int(line[1][10:12])
a = 2008 - birth_year + (12-birth_month)*1.0/12
ages.append(a)
# select training samples within specific age range
timgs = [imgs[i] for i in range(len(ages)) if ages[i]>=1.5]
tvals = [vals[i] for i in range(len(ages)) if ages[i]>=1.5]
print('%s training samples collected'%(len(timgs)))
# select validation samples within specific age range
vimgs = [imgs[i] for i in range(len(ages)) if (ages[i]>=1.0) and (ages[i]<1.5)]
vvals = [vals[i] for i in range(len(ages)) if (ages[i]>=1.0) and (ages[i]<1.5)]
print('%s validation samples collected'%(len(vimgs)))
# sort the IQs, and split dataset into high and low parts
train_sorted_idx = np.argsort(tvals)
train_low_part = train_sorted_idx[0:train_sample_num]
train_high_part = train_sorted_idx[(-1*train_sample_num):]
train_idx = train_low_part.tolist() + train_high_part.tolist()
train_imgs = [timgs[i] for i in train_idx]
train_labels = [0]*train_sample_num + [1]*train_sample_num
val_sorted_idx = np.argsort(vvals)
val_low_part = val_sorted_idx[0:val_sample_num]
val_high_part = val_sorted_idx[(-1*val_sample_num):]
val_idx = val_low_part.tolist() + val_high_part.tolist()
val_imgs = [vimgs[i] for i in val_idx]
val_labels = [0]*val_sample_num + [1]*val_sample_num
if rand_val:
list_shuffle(val_labels)
print('Training samples %s'%(len(train_imgs)))
print('Validation samples %s'%(len(val_imgs)))
return train_imgs, train_labels, val_imgs, val_labels
def source_data(beh_file, dist_file):
"""Read sample information, get split train- and test-dataset."""
# config sample number per class
all_sample_num = 1500
train_sample_num = 1350
# read behavior info
beh_info = open(beh_file).readlines()
beh_info = [line.strip().split(',') for line in beh_info]
vals = [float(line[1]) for line in beh_info]
# read landmark dist info
dist = np.load(dist_file)
dist_mean = dist.mean(axis=0)
dist = dist - dist_mean
sorted_idx = np.argsort(vals)
low_part = sorted_idx[0:all_sample_num]
high_part = sorted_idx[(-1 * all_sample_num):]
low_dist = dist[low_part, :, :]
high_dist = dist[high_part, :, :]
rand_low_dist_idx = range(low_dist.shape[0])
list_shuffle(rand_low_dist_idx)
low_dist = low_dist[rand_low_dist_idx, :, :]
rand_high_dist_idx = range(high_dist.shape[0])
list_shuffle(rand_high_dist_idx)
high_dist = high_dist[rand_high_dist_idx, :, :]
train_dist = np.concatenate((low_dist[:train_sample_num, :, :],
high_dist[:train_sample_num, :, :]),
axis=0)
val_dist = np.concatenate((low_dist[train_sample_num:, :, :],
high_dist[train_sample_num:, :, :]),
axis=0)
train_labels = [0] * train_sample_num + [1] * train_sample_num
val_labels = [0]*(all_sample_num-train_sample_num) + \
[1]*(all_sample_num-train_sample_num)
print 'Training set: %s' % (train_dist.shape[0])
print 'Validation set: %s' % (val_dist.shape[0])
return train_dist, train_labels, val_dist, val_labels
def _shuffle(self):
"""
Shuffles the video list
by regenerating the sequence to sample sequentially
"""
def _is_video_file(filename_):
return any(
filename_.endswith(extension)
for extension in VIDEO_EXTENSIONS)
root = self.root
video_paths = self.video_paths
class_to_idx = self.class_to_idx
list_shuffle(video_paths) # shuffle
videos = list()
frames_per_video = list()
frames_counter = 0
for filename in tqdm(video_paths, ncols=80):
class_ = filename.split('/')[0]
data_path = join(root, filename)
if _is_video_file(data_path):
video_meta = ffprobe(data_path)
start_idx = frames_counter
frames = int(video_meta['video'].get('@nb_frames'))
frames_per_video.append(frames)
frames_counter += frames
item = ((frames_counter - 1, start_idx),
(filename, class_to_idx[class_]))
videos.append(item)
sleep(0.5) # allows for progress bar completion
# update the attributes with the altered sequence
self.video_paths = video_paths
self.videos = videos
self.frames = frames_counter
self.frames_per_video = frames_per_video
def source_data(data_info_file, img_dir, rand_val=False, gender=None):
"""Read sample information, get split train- and test-dataset."""
# config sample number per class
#all_sample_num = 1500
#train_sample_num = 1350
all_sample_num = 1000
train_sample_num = 900
# read sample info
all_info = open(data_info_file).readlines()
all_info.pop(0)
all_info = [line.strip().split(',') for line in all_info]
# select specific gender samples
if gender == 'm':
all_info = [line for line in all_info if int(line[1][16]) % 2 == 1]
elif gender == 'f':
all_info = [line for line in all_info if int(line[1][16]) % 2 == 0]
else:
pass
imgs = [os.path.join(img_dir, line[2]) for line in all_info]
ages = []
for line in all_info:
birth_year = int(line[1][6:10])
birth_month = int(line[1][10:12])
a = 2008 - birth_year + (12 - birth_month) * 1.0 / 12
ages.append(a)
vals = [float(line[3]) for line in all_info]
# sort the IQs, and split dataset into high and low parts
sorted_idx = np.argsort(vals)
low_part = sorted_idx[0:all_sample_num]
high_part = sorted_idx[(-1 * all_sample_num):]
low_imgs = [imgs[i] for i in low_part]
high_imgs = [imgs[i] for i in high_part]
low_ages = [ages[i] for i in low_part]
high_ages = [ages[i] for i in high_part]
# shuffle the sample parts
rand_low_idx = range(len(low_imgs))
list_shuffle(rand_low_idx)
low_imgs = [low_imgs[i] for i in rand_low_idx]
low_ages = [low_ages[i] for i in rand_low_idx]
rand_high_idx = range(len(high_imgs))
list_shuffle(rand_high_idx)
high_imgs = [high_imgs[i] for i in rand_high_idx]
high_ages = [high_ages[i] for i in rand_high_idx]
train_imgs = low_imgs[:train_sample_num] + high_imgs[:train_sample_num]
train_ages = low_ages[:train_sample_num] + high_ages[:train_sample_num]
val_imgs = low_imgs[train_sample_num:] + high_imgs[train_sample_num:]
val_ages = low_ages[train_sample_num:] + high_ages[train_sample_num:]
train_labels = [0] * train_sample_num + [1] * train_sample_num
val_labels = [0]*(all_sample_num-train_sample_num) + \
[1]*(all_sample_num-train_sample_num)
if rand_val:
list_shuffle(val_labels)
return train_imgs, train_ages, train_labels, val_imgs, val_ages, val_labels
def source_mbti_data(data_info_file, img_dir, rand_val=False, gender=None):
"""Read sample information, get split train- and test-dataset."""
# config sample selection criteria for each class
# for E-I factor, <8 : 2971 subjects, >13: 3070 subjects
low_thresh = 8
high_thresh = 13
# select 200 subjects from each class for validation
val_sample_num = 200
# read sample info
all_info = open(data_info_file).readlines()
all_info.pop(0)
all_info = [line.strip().split(',') for line in all_info]
# select samples of specific gender
if gender=='m':
all_info = [line for line in all_info if line[1]=='male']
elif gender=='f':
all_info = [line for line in all_info if line[1]=='female']
else:
pass
imgs = [os.path.join(img_dir, line[9]) for line in all_info]
vals = [float(line[5]) for line in all_info]
# select samples based on specific criteria
assert len(imgs)==len(vals)
high_imgs = []
low_imgs = []
for i in range(len(vals)):
if vals[i]<low_thresh:
low_imgs.append(imgs[i])
elif vals[i]>high_thresh:
high_imgs.append(imgs[i])
# shuffle the samples and split training/validation dataset
list_shuffle(low_imgs)
list_shuffle(high_imgs)
train_imgs = low_imgs[:(-1*val_sample_num)] + high_imgs[:(-1*val_sample_num)]
val_imgs = low_imgs[(-1*val_sample_num):] + high_imgs[(-1*val_sample_num):]
train_labels = [0]*(len(low_imgs)-val_sample_num) + [1]*(len(high_imgs)-val_sample_num)
val_labels = [0]*val_sample_num + [1]*val_sample_num
if rand_val:
list_shuffle(val_labels)
print('Training samples %s - %s'%(len(train_imgs), len(train_labels)))
print('Validation samples %s - %s'%(len(val_imgs), len(val_labels)))
return train_imgs, train_labels, val_imgs, val_labels
def source_data(data_info_file, img_dir, rand_val=False, gender=None):
"""Read sample information, get split train- and test-dataset."""
# config sample number per class
#all_sample_num = 1500
#train_sample_num = 1350
all_sample_num = 1000
train_sample_num = 900
# read sample info
all_info = open(data_info_file).readlines()
all_info.pop(0)
all_info = [line.strip().split(',') for line in all_info]
# select specific gender samples
if gender == 'm':
all_info = [line for line in all_info if int(line[1][16]) % 2 == 1]
elif gender == 'f':
all_info = [line for line in all_info if int(line[1][16]) % 2 == 0]
else:
pass
imgs = [os.path.join(img_dir, line[2]) for line in all_info]
vals = [float(line[3]) for line in all_info]
sorted_idx = np.argsort(vals)
low_part = sorted_idx[0:all_sample_num]
high_part = sorted_idx[(-1 * all_sample_num):]
low_imgs = [imgs[i] for i in low_part]
high_imgs = [imgs[i] for i in high_part]
list_shuffle(low_imgs)
list_shuffle(high_imgs)
train_imgs = low_imgs[:train_sample_num] + high_imgs[:train_sample_num]
val_imgs = low_imgs[train_sample_num:] + high_imgs[train_sample_num:]
train_labels = [0] * train_sample_num + [1] * train_sample_num
val_labels = [0]*(all_sample_num-train_sample_num) + \
[1]*(all_sample_num-train_sample_num)
if rand_val:
list_shuffle(val_labels)
return train_imgs, train_labels, val_imgs, val_labels
def source_landmark_with_age_sampling(data_file, rand_val=False, gender=None):
"""Read sample information, get split train- and test-dataset."""
# config sample number per class
sample_num = 100
small_sample_num = 75
# read sample info
all_info = open(data_file).readlines()
all_info = [line.strip().split(',') for line in all_info]
# select specific gender samples
if gender=='m':
all_info = [line for line in all_info
if int(line[0].split('_')[0][16])%2==1]
elif gender=='f':
all_info = [line for line in all_info
if int(line[0].split('_')[0][16])%2==0]
else:
pass
# get landmarks and IQs
landmarks = [[float(line[2+i]) for i in range(144)] for line in all_info]
vals = [float(line[1]) for line in all_info]
ages = []
for line in all_info:
birth_year = int(line[0].split('_')[0][6:10])
birth_month = int(line[0].split('_')[0][10:12])
a = 2008 - birth_year + (12-birth_month)*1.0/12
ages.append(a)
# select samples within each age group
landmark_list = []
label_list = []
for a in np.unique(ages):
tmp_landmarks = [landmarks[i] for i in range(len(ages)) if ages[i]==a]
tmp_vals = [vals[i] for i in range(len(ages)) if ages[i]==a]
if len(tmp_landmarks)<200:
snum = small_sample_num
else:
snum = sample_num
# select top- and bottom- part of samples
sorted_idx = np.argsort(tmp_vals)
low_part = sorted_idx[:snum]
high_part = sorted_idx[-snum:]
sel_idx = np.concatenate((low_part, high_part))
tmp_landmarks = [tmp_landmarks[i] for i in sel_idx]
tmp_labels = [0]*snum + [1]*snum
# shuffle the sample parts
rand_idx = range(len(tmp_landmarks))
list_shuffle(rand_idx)
tmp_landmarks = [tmp_landmarks[i] for i in rand_idx]
tmp_labels = [tmp_labels[i] for i in rand_idx]
landmark_list.append(tmp_landmarks)
label_list.append(tmp_labels)
# select two subsets of the age groups as validation dataset
group_idx = range(len(landmark_list))
list_shuffle(group_idx)
landmark_list = [landmark_list[i] for i in group_idx]
label_list = [label_list[i] for i in group_idx]
val_landmarks = []
val_labels = []
val_landmarks = [item for line in landmark_list[:2] for item in line]
val_labels = [item for line in label_list[:2] for item in line]
train_landmarks = [item for line in landmark_list[2:] for item in line]
train_labels = [item for line in label_list[2:] for item in line]
if rand_val:
list_shuffle(val_labels)
print('Training samples %s'%(len(train_landmarks)))
print('Validation samples %s'%(len(val_landmarks)))
return train_landmarks, train_labels, val_landmarks, val_labels
def source_data_with_age_sampling(data_info_file, img_dir, sample_num,
small_sample_num, val_set_idx,
rand_val=False, gender=None):
"""Read sample information, get split train- and test-dataset."""
# config sample number per class
#sample_num = 100
#small_sample_num = 75
# read sample info
all_info = open(data_info_file).readlines()
all_info.pop(0)
all_info = [line.strip().split(',') for line in all_info]
# select specific gender samples
if gender=='m':
all_info = [line for line in all_info if int(line[0][16])%2==1]
elif gender=='f':
all_info = [line for line in all_info if int(line[0][16])%2==0]
else:
pass
# remove samples with glasses
all_info = [line for line in all_info if not int(line[1])]
# remove 2 2006-students
all_info = [line for line in all_info if not int(line[0][6:10])==2006]
print('%s samples selected'%(len(all_info)))
imgs = []
for line in all_info:
tmp = []
for item in line[6:]:
tmp.append(os.path.join(img_dir, item))
imgs.append(tmp)
# index 1: graphic reasoning, 2: expanding graph
vals = [float(line[2]) for line in all_info]
ages = []
for line in all_info:
birth_year = int(line[0][6:10])
birth_month = int(line[0][10:12])
a = 2008 - birth_year + (12-birth_month)*1.0/12
ages.append(a)
# select samples within each age group
img_list = []
label_list = []
unique_ages = np.unique(ages)
for a in unique_ages:
tmp_imgs = [imgs[i] for i in range(len(ages)) if ages[i]==a]
tmp_vals = [vals[i] for i in range(len(ages)) if ages[i]==a]
if len(tmp_imgs)<(2*sample_num):
snum = small_sample_num
else:
snum = sample_num
# select top- and bottom- part of samples
sorted_idx = np.argsort(tmp_vals)
low_part = sorted_idx[:snum]
high_part = sorted_idx[-snum:]
sel_idx = np.concatenate((low_part, high_part))
tmp_imgs = [tmp_imgs[i] for i in sel_idx]
tmp_labels = [0]*snum + [1]*snum
# shuffle the sample parts
rand_idx = list(range(len(tmp_imgs)))
list_shuffle(rand_idx)
tmp_imgs = [tmp_imgs[i] for i in rand_idx]
tmp_labels = [tmp_labels[i] for i in rand_idx]
img_list.append(tmp_imgs)
label_list.append(tmp_labels)
# select two subsets of the age groups as validation dataset
val_set_ids = []
#for i in range(len(img_list)):
# for j in range(i+1, len(img_list)):
# val_set_ids.append([i, j])
#val_set_ids = val_set_ids[val_set_idx]
#set_ids = range(len(img_list))
#train_set_ids = [item for item in set_ids if not item in val_set_ids]
#print 'Age groups for validation: %s and %s'%(unique_ages[val_set_ids[0]],
# unique_ages[val_set_ids[1]])
for i in range(len(img_list)):
val_set_ids.append([i])
val_set_ids = val_set_ids[val_set_idx]
set_ids = range(len(img_list))
train_set_ids = [item for item in set_ids if not item in val_set_ids]
print('Age groups for validation: %s'%(unique_ages[val_set_ids[0]]))
val_imgs = [line for i in val_set_ids
for line in img_list[i]]
val_labels = [item for i in val_set_ids
for item in label_list[i]]
train_imgs = [line for i in train_set_ids
for line in img_list[i]]
train_labels = [item for i in train_set_ids
for item in label_list[i]]
# make a full size image array
train_mlen = max([len(line) for line in train_imgs])
train_imgs = [line+['null']*(train_mlen-len(line)) for line in train_imgs]
val_mlen = max([len(line) for line in val_imgs])
val_imgs = [line+['null']*(val_mlen-len(line)) for line in val_imgs]
if rand_val:
list_shuffle(val_labels)
print('Training samples %s'%(len(train_imgs)))
print('Validation samples %s'%(len(val_imgs)))
return train_imgs, train_labels, val_imgs, val_labels
def shuffle(self):
list_shuffle(self.cards)