print('loaded VGG!')
# ############学習の高速化のためにパラメータを事前に整理しておく###################
if args.usevgg == 1:
print('preprocessing vgg data')
vgg_img_param = []
start = time.time()
if os.path.isfile('features/vggparam_yahoo.pickle'):
print('vgg checkpoint pickle is exist! loading pickle')
with open('features/vggparam_yahoo.pickle', mode='rb') as f:
vgg_img_param = pickle.load(f)
else:
for filename in filenames:
vgg_img_param.append(vggparamater(filename, args.gpu, vgg)[0])
with open('features/vggparam_yahoo.pickle', mode='wb') as f:
pickle.dump(vgg_img_param, f)
calctime = time.time() - start
print('VGG time: ' + str(calctime), '[sec]')
print('vgg end')
#############################################################################
print('training start')
dataset = []
if os.path.isfile('style.npy'):
print('the file is exists! load...')
styleg = np.load('styleg.npy')
style = np.load('style.npy')
else:
model = args.model
model_path = 'models/yahoo100m_gan/epoch_{}.model'.format(model)
vgg = VGGNet()
serializers.load_hdf5('/tmp/VGG.model', vgg)
if args.usevgg == 1:
tinynet = Generator()
else:
tinynet = Generator()
serializers.load_npz(model_path, tinynet)
tinynet.to_gpu()
words = ['outdoor', 'black', 'urban', 'white',
'indoor', 'new', 'serene', 'red', 'light', 'tree', 'country', 'old', 'dark']
for word in words:
for count in range(28):
filename = 'images/valid/{}.jpg'.format(count)
vgg_param = vggparamater(filename, 0, vgg)[0]
if args.usevgg == 1:
concatted = concatData(word, vgg_param)
else:
concatted = w2v(word)
print('moto:')
# print(concatted[0][400:420])
# print(concatted[0][0:10])
if args.usevgg == 0:
concatted = np.reshape(concatted, (1, 200))
print(concatted.shape)
concatted_g = Variable(cuda.to_gpu(concatted))
style_params = tinynet(concatted_g, train=False)
# print(style_params.data.shape,concatted_g.data.shape)
print('params:')
print(style_params.data[0][0:10])