def gen_output_to_enc_input(gX):
gX = gX.reshape(-1, nc, crop, crop)
gX = np.round(rescale(gX, dataset.native_range, (0, 255)))
gX[gX < 0] = 0
gX[gX > 255] = 255
return np.array(gX, dtype=np.uint8)
def gen_transform(gX):
# X: float tensor in [0, 1] (e.g., output by generator's sigmoid)
# returns float tensor in native_range
# ([-1, 1] for ImageNet, [0, 1] for MNIST)
return rescale(gX, (0, 1), dataset.native_range)
def transform(X, crop=args.crop_resize):
# X: uint8-type ndarray [0, 255] (possibly flattened)
# returns NCHW float array in [0, 1]
X = floatX(X).reshape(-1, nc, crop, crop)
return rescale(X, (0, 255), dataset.native_range)
def input_transform(X):
# X: uint8-type tensor [0, 255]
# returns a float tensor in native_range
# ([-1, 1] for ImageNet, [0, 1] for MNIST)
X = T.cast(X, theano.config.floatX)
return rescale(X, (0, 255), dataset.native_range)
import sys import data # parameters f_in=sys.argv[1] f_out=sys.argv[2] # get parameters scale min_val,max_val=data.get_parameters_scale(f_in) print "minimal value",min_val print "maximal value",max_val # scale the dataset data.rescale(f_in,f_out,min_val,max_val)