def interpolate_full(self, img0, img1, interp=serp, x0=-0.5, x1=1.5, delta=1/32.0):
"""Return a visualization of an interpolation between img0 and img1,
using interpolation method interp. The interpolation starts
with parameter x0 and goes to x1, in increments of delta.
Note that img0 corresponds to parameter x0=0 and img1 to
parameter x1=1. The default is to start outside that range,
and so we do some extrapolation.
"""
z0 = self.get_lv(img0).reshape((100,))
z1 = self.get_lv(img1).reshape((100,))
ps = np.arange(x0, x1-0.000001, delta)
n = ps.size
arrays = [lerp(z0, z1, p) for p in ps]
z = np.stack(arrays); print z.shape
zmb = floatX(z[0 : n, :]); print zmb.shape
xmb = self.model._gen(zmb); print xmb.shape
samples = [xmb]
samples = np.concatenate(samples, axis=0)
samples = self.model.inverse_transform(
samples, npx=self.model.npx, nc=self.model.nc)
samples = (samples * 255).astype(np.uint8)
m = math.ceil(math.sqrt(n))
img_vis = utils.grid_vis(samples, m, m)
return img_vis
def update_vis(self):
ims = self.opt_engine.get_images(self.frame_id)
if ims is not None:
self.ims = ims
if self.ims is None:
return
ims_show = []
n_imgs = self.ims.shape[0]
for n in range(n_imgs):
# im = ims[n]
im_s = cv2.resize(self.ims[n], (self.width, self.width),
interpolation=cv2.INTER_CUBIC)
if n == self.select_id and self.topK > 1:
t = 3 # thickness
cv2.rectangle(im_s, (t, t), (self.width - t, self.width - t),
(0, 255, 0), t)
im_s = im_s[np.newaxis, ...]
ims_show.append(im_s)
if ims_show:
ims_show = np.concatenate(ims_show, axis=0)
g_tmp = utils.grid_vis(ims_show, self.grid_size[1],
self.grid_size[0]) # (nh, nw)
self.vis_results = g_tmp.copy()
self.update()
def interpolate(url0, url1, output_image):
model_class = locate('model_def.dcgan_theano')
model_file = './models/handbag_64.dcgan_theano'
model = model_class.Model(
model_name="handbag_64", model_file=model_file)
# save images
for j, url in enumerate([url0, url1]):
r = requests.get(url)
i = Image.open(StringIO(r.content))
i.save("pics/url"+str(j)+".jpg")
z0 = iGAN_predict.find_latent(url=url0).reshape((100,))
z1 = iGAN_predict.find_latent(url=url1).reshape((100,))
delta = 1.0/32.0
arrays = [p*z0+(1-p)*z1 for p in np.arange(-16*delta, 1+16*delta-0.0001, delta)]
z = np.stack(arrays)
print(z.shape)
zmb = floatX(z[0 : 64, :])
xmb = model._gen(zmb)
samples = [xmb]
samples = np.concatenate(samples, axis=0)
print(samples.shape)
samples = model.inverse_transform(samples, npx=model.npx, nc=model.nc)
samples = (samples * 255).astype(np.uint8)
# generate grid visualization
im_vis = utils.grid_vis(samples, 8, 8)
# write to the disk
im_vis = cv2.cvtColor(im_vis, cv2.COLOR_BGR2RGB)
cv2.imwrite(output_image, im_vis)
def imagify(self, z):
"""Return an image corresponding to the latent vector z."""
z = np.stack([z.reshape((100,))])
zmb = floatX(z[0 : 1, :]);
xmb = self.model._gen(zmb);
samples = np.concatenate([xmb], axis=0)
samples = self.model.inverse_transform(
samples, npx=self.model.npx, nc=self.model.nc)
samples = (samples * 255).astype(np.uint8)
img_vis = utils.grid_vis(samples, 1, 1)
return img_vis
def generate(z, output_image):
model_class = locate('model_def.dcgan_theano')
model_file = './models/handbag_64.dcgan_theano'
model = model_class.Model(
model_name="handbag_64", model_file=model_file)
samples = []
n = 1
batch_size = 1
z = z.reshape((1, 100))
zmb = floatX(z[0 : n, :])
xmb = model._gen(zmb)
samples.append(xmb)
samples = np.concatenate(samples, axis=0)
samples = model.inverse_transform(samples, npx=model.npx, nc=model.nc)
samples = (samples * 255).astype(np.uint8)
#samples = model.gen_samples(z0=None, n=196, batch_size=49, use_transform=True)
# generate grid visualization
im_vis = utils.grid_vis(samples, 1, 1)
# write to the disk
im_vis = cv2.cvtColor(im_vis, cv2.COLOR_BGR2RGB)
cv2.imwrite(output_image, im_vis)
def update_vis(self):
ims = self.opt_engine.get_images(self.frame_id)
if ims is not None:
self.ims = ims
if self.ims is None:
return
ims_show = []
n_imgs = self.ims.shape[0]
for n in range(n_imgs):
# im = ims[n]
im_s = cv2.resize(self.ims[n], (self.width, self.width), interpolation=cv2.INTER_CUBIC)
if n == self.select_id and self.topK > 1:
t = 3 # thickness
cv2.rectangle(im_s, (t, t), (self.width - t, self.width - t), (0, 255, 0), t)
im_s = im_s[np.newaxis, ...]
ims_show.append(im_s)
if ims_show:
ims_show = np.concatenate(ims_show, axis=0)
g_tmp = utils.grid_vis(ims_show, self.grid_size[1], self.grid_size[0]) # (nh, nw)
self.vis_results = g_tmp.copy()
self.update()
# load data from hdf5 file
tr_data, te_data, tr_stream, te_stream, ntrain, ntest = load.load_imgs(
ntrain=None,
ntest=None,
batch_size=args.batch_size,
data_file=args.data_file)
te_handle = te_data.open()
test_x, = te_data.get_data(te_handle, slice(0, ntest))
# generate real samples and test transform/inverse_transform
test_x = train_dcgan_utils.transform(test_x, nc=nc)
vis_idxs = py_rng.sample(np.arange(len(test_x)), n_vis)
vaX_vis = train_dcgan_utils.inverse_transform(test_x[vis_idxs], npx=npx, nc=nc)
# st()
n_grid = int(np.sqrt(n_vis))
grid_real = utils.grid_vis((vaX_vis * 255.0).astype(np.uint8), n_grid, n_grid)
train_dcgan_utils.save_image(grid_real,
os.path.join(sample_dir, 'real_samples.png'))
# define DCGAN model
disc_params = train_dcgan_utils.init_disc_params(n_f=n_f,
n_layers=n_layers,
nc=nc)
gen_params = train_dcgan_utils.init_gen_params(nz=nz,
n_f=n_f,
n_layers=n_layers,
nc=nc)
x = T.tensor4()
z = T.matrix()
gx = train_dcgan_utils.gen(z, gen_params, n_layers=n_layers, n_f=n_f, nc=nc)
parser.add_argument('--model_type', dest='model_type', help='the generative models and its deep learning framework', default='dcgan_theano', type=str)
parser.add_argument('--framework', dest='framework', help='deep learning framework', default='theano')
parser.add_argument('--model_file', dest='model_file', help='the file that stores the generative model', type=str, default=None)
parser.add_argument('--output_image', dest='output_image', help='the name of output image', type=str, default=None)
args = parser.parse_args()
return args
if __name__ == '__main__':
args = parse_args()
if not args.model_file: #if model directory is not specified
args.model_file = './models/%s.%s' % (args.model_name, args.model_type)
if not args.output_image:
args.output_image = '%s_%s_samples.png' % (args.model_name, args.model_type)
for arg in vars(args):
print('[%s] =' % arg, getattr(args, arg))
# initialize model and constrained optimization problem
model_class = locate('model_def.%s' % args.model_type)
model = model_class.Model(model_name=args.model_name, model_file=args.model_file)
# generate samples
samples = model.gen_samples(z0=None, n=196, batch_size=49, use_transform=True)
# generate grid visualization
im_vis = utils.grid_vis(samples, 14, 14)
# write to the disk
im_vis = cv2.cvtColor(im_vis, cv2.COLOR_BGR2RGB)
cv2.imwrite(args.output_image, im_vis)
print('samples_shape', samples.shape)
print('save image to %s' % args.output_image)
web_dir = os.path.join(args.cache_dir, 'web_dcgan') html = image_save.ImageSave(web_dir, expr_name, append=True) utils.mkdirs([sample_dir, model_dir, log_dir, web_dir]) # load data from hdf5 file tr_data, te_data, tr_stream, te_stream, ntrain, ntest = load.load_imgs(ntrain=None, ntest=None, batch_size=args.batch_size,data_file=args.data_file) te_handle = te_data.open() test_x, = te_data.get_data(te_handle, slice(0, ntest)) # generate real samples and test transform/inverse_transform test_x = train_dcgan_utils.transform(test_x, nc=nc) vis_idxs = py_rng.sample(np.arange(len(test_x)), n_vis) vaX_vis = train_dcgan_utils.inverse_transform(test_x[vis_idxs], npx=npx, nc=nc) # st() n_grid = int(np.sqrt(n_vis)) grid_real = utils.grid_vis((vaX_vis*255.0).astype(np.uint8), n_grid, n_grid) train_dcgan_utils.save_image(grid_real, os.path.join(sample_dir, 'real_samples.png')) # define DCGAN model disc_params = train_dcgan_utils.init_disc_params(n_f=n_f, n_layers=n_layers, nc=nc) gen_params = train_dcgan_utils.init_gen_params(nz=nz, n_f=n_f, n_layers=n_layers, nc=nc) x = T.tensor4() z = T.matrix() gx = train_dcgan_utils.gen(z, gen_params, n_layers=n_layers, n_f=n_f, nc=nc) p_real = train_dcgan_utils.discrim(x, disc_params, n_layers=n_layers) p_gen = train_dcgan_utils.discrim(gx, disc_params, n_layers=n_layers) d_cost_real = costs.bce(p_real, T.ones(p_real.shape)) d_cost_gen = costs.bce(p_gen, T.zeros(p_gen.shape))
return args
if __name__ == '__main__':
args = parse_args()
if not args.model_file: #if model directory is not specified
args.model_file = './models/%s.%s' % (args.model_name, args.model_type)
if not args.output_image:
args.output_image = '%s_%s_samples.png' % (args.model_name,
args.model_type)
for arg in vars(args):
print('[%s] =' % arg, getattr(args, arg))
# initialize model and constrained optimization problem
model_class = locate('model_def.%s' % args.model_type)
model = model_class.Model(model_name=args.model_name,
model_file=args.model_file)
# generate samples
samples = model.gen_samples(z0=None,
n=196,
batch_size=49,
use_transform=True)
# generate grid visualization
im_vis = utils.grid_vis(samples, 14, 14)
# write to the disk
im_vis = cv2.cvtColor(im_vis, cv2.COLOR_BGR2RGB)
cv2.imwrite(args.output_image, im_vis)
print('samples_shape', samples.shape)
print('save image to %s' % args.output_image)
