def coco(mode="dev", n_captions=1, test_size=None):
"""loads coco data into train and test features and targets.
mode = 'dev' is used for development (quick load of subset)
"""
# train_fns
dataType='train2014'
train_fns = os.listdir("%s/features/%s"%(COCO_DIR, dataType))
# reduce it to a dev set
if mode == "dev":
train_fns = shuffle(train_fns)[:256]
trX, trY = loadFeaturesTargets(train_fns, dataType, n_captions)
# val_fns
dataType='val2014'
test_fns = os.listdir("%s/features/%s"%(COCO_DIR, dataType))
# reduce it to a dev set
if mode == "dev":
test_fns = shuffle(test_fns)[:128]
if test_size:
test_fns = shuffle(test_fns)[:test_size]
teX, teY = loadFeaturesTargets(test_fns, dataType, n_captions)
return trX, teX, trY, teY
def coco(mode="dev", n_captions=1, test_size=None):
"""loads coco data into train and test features and targets.
mode = 'dev' is used for development (quick load of subset)
"""
# train_fns
dataType = 'train2014'
train_fns = os.listdir("%s/features/%s" % (COCO_DIR, dataType))
# reduce it to a dev set
if mode == "dev":
train_fns = shuffle(train_fns)[:256]
trX, trY = loadFeaturesTargets(train_fns, dataType, n_captions)
# val_fns
dataType = 'val2014'
test_fns = os.listdir("%s/features/%s" % (COCO_DIR, dataType))
# reduce it to a dev set
if mode == "dev":
test_fns = shuffle(test_fns)[:128]
if test_size:
test_fns = shuffle(test_fns)[:test_size]
teX, teY = loadFeaturesTargets(test_fns, dataType, n_captions)
return trX, teX, trY, teY
def iterXY(self, X, Y):
if self.shuffle:
X, Y = shuffle(X, Y)
for xmb, ymb in iter_data(X, Y, size=self.size):
xmb = self.trXt(xmb)
ymb = self.trYt(ymb)
yield xmb, ymb
def iterXY(self, X, Y):
if self.shuffle:
X, Y = shuffle(X, Y)
for xmb, ymb in iter_data(X, Y, size=self.size):
xmb = self.trXt(xmb)
ymb = self.trYt(ymb)
yield xmb, ymb
def coco(mode="dev", batch_size=64, n_captions=1):
# train_fns
dataType='train2014'
train_fns = os.listdir("%s/features/%s"%(dataDir, dataType))
# reduce it to a dev set
if mode == "dev":
train_fns = shuffle(train_fns)[:batch_size*50]
trX, trY = loadFeaturesTargets(train_fns, dataType, n_captions)
# val_fns
dataType='val2014'
test_fns = os.listdir("%s/features/%s"%(dataDir, dataType))
# reduce it to a dev set
if mode == "dev":
test_fns = shuffle(test_fns)[:batch_size*25]
teX, teY = loadFeaturesTargets(test_fns, dataType, n_captions)
return trX, teX, trY, teY
def iterXY(self, X, Y):
if self.shuffle:
X, Y = shuffle(X, Y)
self.loader = Loader(X, self.train_load, self.train_transform, self.size)
self.proc = Process(target=self.loader.load)
self.proc.start()
for ymb in iter_data(Y, size=self.size):
xmb = self.loader.get()
yield xmb, floatX(ymb)
def iterXY(self, X, Y):
if self.shuffle:
X, Y = shuffle(X, Y)
self.loader = Loader(X, self.train_load, self.train_transform,
self.size)
self.proc = Process(target=self.loader.load)
self.proc.start()
for ymb in iter_data(Y, size=self.size):
xmb = self.loader.get()
yield xmb, floatX(ymb)
def iterXY(self, X, Y):
if self.shuffle:
X, Y = shuffle(X, Y)
for x_chunk, y_chunk in iter_data(X, Y, size=self.size*20):
sort = np.argsort([len(x) for x in x_chunk])
x_chunk = [x_chunk[idx] for idx in sort]
y_chunk = [y_chunk[idx] for idx in sort]
mb_chunks = [[x_chunk[idx:idx+self.size], y_chunk[idx:idx+self.size]] for idx in range(len(x_chunk))[::self.size]]
py_rng.shuffle(mb_chunks)
for xmb, ymb in mb_chunks:
xmb = self.trXt(xmb)
ymb = self.trYt(ymb)
yield xmb, ymb
def iterXY(self, X, Y):
if self.shuffle:
X, Y = shuffle(X, Y)
for x_chunk, y_chunk in iter_data(X, Y, size=self.size*20):
sort = np.argsort([len(x) for x in x_chunk])
x_chunk = [x_chunk[idx] for idx in sort]
y_chunk = [y_chunk[idx] for idx in sort]
mb_chunks = [[x_chunk[idx:idx+self.size], y_chunk[idx:idx+self.size]] for idx in range(len(x_chunk))[::self.size]]
py_rng.shuffle(mb_chunks)
for xmb, ymb in mb_chunks:
xmb = self.trXt(xmb)
ymb = self.trYt(ymb)
yield xmb, ymb
def loadFeaturesTargets(fns, dataType, n_captions=1):
"""
Note: filenames should come from the same type of dataType.
filenames from val2014, for example, should have dataType val2014
Parameters
----------
fns: filenames, strings
dataType: string folder, i.e. train2014, val2014
n_captions: int, number of captions for each image to load
Returns
-------
X: list of im_vects
1st list length = len(fns)
vectors are shape (4096, )
Y: list of list of captions.
1st list length = len(fns)
sublist length = n_captions
"""
annFile = '%s/annotations/captions_%s.json' % (COCO_DIR, dataType)
caps = COCO(annFile)
X = []
Y = []
for fn in fns:
# Features
x = np.load('%s/features/%s/%s' % (COCO_DIR, dataType, fn))
# Targets
annIds = caps.getAnnIds(imgIds=getImageId(fn))
anns = caps.loadAnns(annIds)
# sample n_captions per image
anns = shuffle(anns)
captions = [getCaption(anns[i]) for i in range(n_captions)]
X.append(x)
Y.append(captions)
return X, Y
def loadFeaturesTargets(fns, dataType, n_captions=1):
"""
Note: filenames should come from the same type of dataType.
filenames from val2014, for example, should have dataType val2014
Parameters
----------
fns: filenames, strings
dataType: string folder, i.e. train2014, val2014
n_captions: int, number of captions for each image to load
Returns
-------
X: list of im_vects
1st list length = len(fns)
vectors are shape (4096, )
Y: list of list of captions.
1st list length = len(fns)
sublist length = n_captions
"""
annFile = '%s/annotations/captions_%s.json'%(dataDir,dataType)
caps=COCO(annFile)
X = []
Y = []
for fn in fns:
# Features
x = np.load('%s/features/%s/%s'%(dataDir, dataType, fn))
# Targets
annIds = caps.getAnnIds(imgIds=getImageId(fn));
anns = caps.loadAnns(annIds)
# sample n_captions per image
anns = shuffle(anns)
captions = [getCaption(anns[i]) for i in range(n_captions)]
X.append(x)
Y.append(captions)
return X, Y
def cocoXYFilenames(n_captions=5, dataType='val2014'):
"""Helps when you are evaluating and want the filenames
associated with the features and target variables
Parameters
----------
n_captions: integer
how many captions to load for the image
dataType: 'val2014' or 'train2014'
Returns
-------
X: the features
Y: the targets
filenames: the filenames corresponding to each
"""
fns = os.listdir("%s/features/%s" % (COCO_DIR, dataType))
fns = shuffle(fns)
X, Y = loadFeaturesTargets(fns, dataType, n_captions)
return X, Y, fns
def cocoXYFilenames(n_captions=5, dataType='val2014'):
"""Helps when you are evaluating and want the filenames
associated with the features and target variables
Parameters
----------
n_captions: integer
how many captions to load for the image
dataType: 'val2014' or 'train2014'
Returns
-------
X: the features
Y: the targets
filenames: the filenames corresponding to each
"""
fns = os.listdir("%s/features/%s"%(dataDir, dataType))
fns = shuffle(fns)
X, Y = loadFeaturesTargets(fns, dataType, n_captions)
return X, Y, fns
def get_data(self, request=None):
if request is not None:
raise ValueError
data = next(self.child_epoch_iterator)
return shuffle(*data)
def get_data(self, request=None):
if request is not None:
raise ValueError
data = next(self.child_epoch_iterator)
return shuffle(*data)