def __init__(self, params):
self.batch_size = params['batch_size']
self.outshape = params['shape']
self.classes_per_batch = params['classes_per_batch']
self.img_lmdb = lmdbs(params['img_source'])
if params['skt_source'].endswith('.pkl'):
self.skt_lmdb = svgs(params['skt_source'])
else:
self.skt_lmdb = lmdbs(params['skt_source'])
self.img_labels = self.img_lmdb.get_label_list()
self.skt_labels = self.skt_lmdb.get_label_list()
self.img_mean = biproto2py(params['mean_file']).squeeze()
self.num_classes = len(set(self.skt_labels))
assert len(self.skt_labels)%self.num_classes==0, \
'Unequal sketch training samples for each class'
self.skt_per_class = len(self.skt_labels) / self.num_classes
if 'hard_pos' in params:
self.hard_sel = 'hard_pos'
self.hard_sel_file = params['hard_pos']
elif 'hard_neg' in params:
self.hard_sel = 'hard_neg'
self.hard_sel_file = params['hard_neg']
elif 'hard_pn' in params:
self.hard_sel = 'hard_pn'
self.hard_sel_file = params['hard_pn']
else:
assert False, 'Hard selection must be on'
self.watchChange = WatchDog(
self.hard_sel_file) #check if file has been updated
self.hardsel_tab = np.load(self.hard_sel_file)
self.hard_pos = self.hardsel_tab['pos']
self.hard_neg = self.hardsel_tab['neg']
self.img_labels_dict, self.classes = vec2dic(self.img_labels)
self.NSKTS = len(self.skt_labels)
self.indexlist = range(self.NSKTS)
self.shuffle_keeping_min_classes_per_batch()
self._cur = 0 # current image
# this class does some simple data-manipulations
self.img_augment = SimpleAugment(mean=self.img_mean,
shape=params['shape'],
scale=params['scale'],
rot=params['rot'])
if 'verbose' not in params:
print "BatchLoader initialized with {} sketches, {} images of {} classes".format(
len(self.skt_labels), len(self.img_labels), self.num_classes)
print('Hard selection: {}'.format(self.hard_sel))
#create threadpools for parallel augmentation
self.pool = ThreadPool() #4
def __init__(self, params):
self.batch_size = params['batch_size']
self.outshape = params['shape']
self.classes_per_batch = params['classes_per_batch']
self.img_lmdb = lmdbs(params['img_source'])
if params['skt_source'].endswith('.pkl'):
self.skt_lmdb = svgs(params['skt_source'])
else:
self.skt_lmdb = lmdbs(params['skt_source'])
self.img_labels = self.img_lmdb.get_label_list()
self.skt_labels = self.skt_lmdb.get_label_list()
self.img_mean = biproto2py(params['mean_file']).squeeze()
self.num_classes = len(set(self.skt_labels))
assert len(self.skt_labels)%self.num_classes==0, \
'Unequal sketch training samples for each class'
self.skt_per_class = len(self.skt_labels) / self.num_classes
self.hard_sel = 0
self.img_labels_dict, self.classes = vec2dic(self.img_labels)
self.indexlist = range(len(self.skt_labels))
self.shuffle_keeping_min_classes_per_batch()
self._cur = 0 # current image
# this class does some simple data-manipulations
self.img_augment = SimpleAugment(mean=self.img_mean,
shape=params['shape'],
scale=params['scale'],
rot=params['rot'])
if 'verbose' not in params:
print "BatchLoader initialized with {} sketches, {} images of {} classes".format(
len(self.skt_labels), len(self.img_labels), self.num_classes)
#create threadpools for parallel augmentation
self.pool = ThreadPool() #4
def __init__(self, img_dir='', img_lst='', lmdb=''):
self.have_img_lst = self.have_img_dir = self.have_lmdb = False
self.hardsel = None
if img_lst:
helps = helper()
self.img_lst = helps.read_list(img_lst,
delimeter=',',
keep_original=False)
self.have_img_lst = True
if img_dir:
assert os.path.isdir(img_dir), 'Opps. img_dir is not a dir'
self.img_dir = img_dir
self.have_img_dir = True
if lmdb:
self.lmdb = lmdbs(lmdb)
self.have_lmdb = True
def __init__(self, params):
self.batch_size = params['batch_size']
self.outshape = params['shape']
self.lmdb = lmdbs(params['source'])
self.labels = self.lmdb.get_label_list()
self.img_mean = biproto2py(params['mean_file']).squeeze()
self.NIMGS = len(self.labels)
self.num_batches = int(np.ceil(self.NIMGS/float(self.batch_size)))
self._cur = 0 # current batch
# this class does some simple data-manipulations
self.img_augment = SimpleAugment(mean=self.img_mean,shape=params['shape'],
scale = params['scale'])
def __init__(self, params):
self.batch_size = params['batch_size']
self.outshape = params['shape']
self.lmdb = lmdbs(params['source'])
self.labels = self.lmdb.get_label_list()
self.img_mean = biproto2py(params['mean_file']).squeeze()
self.NIMGS = len(self.labels)
assert self.NIMGS%self.batch_size==0,'NIMGS {} not dividible by batchsize {}'.format(
self.NIMGS,self.batch_size)
self.num_batches = self.NIMGS/self.batch_size
self._cur = 0 # current batch
self.labels_tab = self.labels.reshape((self.num_batches,self.batch_size))
# this class does some simple data-manipulations
self.img_augment = SimpleAugment(mean=self.img_mean,shape=params['shape'],
scale = params['scale'])
def lmdb_get_info(DB):
"""
get information about an lmdb
IN: DB can be either a true lmdb or a python pickle
includes: number of classes, number of sample per class, number of samples
"""
if DB.endswith('.pkl'):
helps = helper()
data = helps.load(DB, 1) #load the first variable which is the labels
labels = data['labels']
else:
lmdb_ = lmdbs(DB)
labels = lmdb_.get_label_list()
out = {}
out['num_classes'] = len(set(labels))
out['num_samples'] = len(labels)
if out['num_samples'] % out['num_classes'] != 0:
print 'We got an unbalance lmdb having {} samples of {} classes'.format(\
out['num_samples'],out['num_classes'])
out['samples_per_class'] = out['num_samples'] / out['num_classes']
return out
def __init__(self, params):
self.batch_size = params['batch_size']
self.img_shape = params['shape']
self.classes_per_batch = params['classes_per_batch']
self.img_lmdb = lmdbs(params['img_source'])
if params['skt_source'].endswith('.pkl'):
self.skt_lmdb = svgs(params['skt_source'])
else:
self.skt_lmdb = lmdbs(params['skt_source'])
self.img_labels = self.img_lmdb.get_label_list()
self.skt_labels = self.skt_lmdb.get_label_list()
label_ids = list(set(self.img_labels))
NCATS = len(label_ids)
if label_ids[0] != 0 or label_ids[-1] != NCATS - 1:
if 'verbose' not in params:
print 'Your data labels are not [0:{}]. Converting label ...'.format(
NCATS - 1)
self.img_labels = [
label_ids.index(label) for label in self.img_labels
]
self.skt_labels = [
label_ids.index(label) for label in self.skt_labels
]
self.img_mean = biproto2py(params['mean_file']).squeeze()
#self.skt_mean = biproto2py(params['skt_mean']).squeeze()
self.num_classes = len(set(self.skt_labels))
assert self.num_classes == NCATS, 'XX!!Sketch & image datasets unequal #categories'
assert len(self.skt_labels)%self.num_classes==0, \
'Unequal sketch training samples for each class'
self.skt_per_class = len(self.skt_labels) / self.num_classes
if 'hard_pos' in params:
self.hard_sel = 1
self.hard_pos = np.load(params['hard_pos'])['pos']
elif 'hard_neg' in params:
self.hard_sel = 2
self.hard_neg = np.load(params['hard_neg'])['neg']
elif 'hard_pn' in params:
self.hard_sel = 3
tmp = np.load(params['hard_pn'])
self.hard_pos = tmp['pos']
self.hard_neg = tmp['neg']
else: #hard selection turn off
self.hard_sel = 0
#self.img_labels_dict, self.classes = vec2dic(self.img_labels)
self.indexlist = range(len(self.skt_labels))
self.indexlist_img = range(len(self.img_labels))
#self.shuffle_keeping_min_classes_per_batch()
shuffle(self.indexlist)
shuffle(self.indexlist_img)
self._cur = 0 # current image
self._cur_img = 0
# this class does some simple data-manipulations
self.img_augment = SimpleAugment(mean=self.img_mean,
shape=self.img_shape,
scale=params['scale'],
rot=params['rot'])
print "BatchLoader initialized with {} sketches, {} images of {} classes".format(
len(self.skt_labels), len(self.img_labels), self.num_classes)
#create threadpools for parallel augmentation
self.pool = ThreadPool() #4
def extract_cnn_feat(net_params, DB, OUT, layer=0, verbose=True):
"""
extract features from CNN
DB: lmdb data you want to extract feature
net_params: dictionary with keys "DEPLOY_PRO","data_mean",
"WEIGHTS","scale_factor", batch_size
OUT: save output in mat file
layer: 0 for last layer, -1: one before the last layer, -2: ...
"""
assert layer <= 0, 'layer should be a non-positive integer'
DEPLOY_PRO = net_params['DEPLOY_PRO']
WEIGHTS = net_params['WEIGHTS']
scale_factor = net_params['scale_factor']
data_mean = net_params['data_mean']
batch_size = net_params['batch_size']
net = caffe.Net(DEPLOY_PRO, WEIGHTS, caffe.TEST)
if verbose:
print 'Extracting cnn feats...'
print ' Model def: {}\n Weights: {}'.format(DEPLOY_PRO, WEIGHTS)
start_t = time.time()
db = lmdbs(DB)
labels = db.get_label_list()
NIMGS = labels.size
img_mean = biproto2py(data_mean)
inblob = net.inputs[0]
in_dim = net.blobs[inblob].data.shape[1:]
prep = SimpleAugment(mean=img_mean, shape=in_dim[-2:])
feat_l = net.blobs.keys()[layer - 1]
out_dim = net.blobs[feat_l].data.squeeze().shape[-1]
feats = np.zeros((NIMGS, out_dim), dtype=np.float32)
for i in xrange(0, NIMGS, batch_size):
batch = range(i, min(i + batch_size, NIMGS))
if verbose:
print('Processing sample #{} - {}'.format(batch[0], batch[-1]))
new_shape = (len(batch), ) + in_dim
net.blobs[inblob].reshape(*new_shape)
chunk = db.get_data(batch)
net.blobs[inblob].data[...] = prep.augment_deploy(chunk)
temp = net.forward()
feats[batch] = net.blobs[feat_l].data.squeeze()
#apply scale factor
feats *= scale_factor
if OUT.endswith('.mat'):
py2mat(dict(feats=feats, labels=labels), OUT)
elif OUT.endswith('.npz'):
np.savez(OUT, feats=feats, labels=labels)
else: #assume it is pickle
helps = helper()
helps.save(OUT, feats=feats, labels=labels)
net = None
if verbose:
end_t = time.time()
print 'Save features to {}.'.format(OUT)
print 'Time: {}\n'.format(timedelta(seconds=int(end_t - start_t)))
