# Set this
transformer_name = args['transformer'] # transformer to use
N = args['n_img'] # number of transformations
# Special for CPAB
if transformer_name == 'CPAB': s = setup_CPAB_transformer()
# Get transformer and random transformation
transformer = get_transformer_layer(transformer_name)
theta = get_random_theta(N, transformer_name)
# Load im and create a batch of imgs
im = get_cat()
im = np.tile(im, (N, 1, 1, 1))
# Cast to tensorflow
im_tf = tf.cast(im, tf.float32)
theta_tf = tf.cast(theta, tf.float32)
# Transformer imgs
trans_im, theta_fg = transformer(im_tf, theta_tf, (1200, 1600))
# Run computations
run_options = tf.RunOptions(report_tensor_allocations_upon_oom=True)
sess = tf.Session()
out_im, theta_gr = sess.run([trans_im, theta_fg], options=run_options)
print(theta_gr)
# Show the transformed images
show_images(out_im, title=transformer_name + ' transformations')
V = ST_TPS_transformer(input_repeated, thetas, out_size, tps_size)
return V
#%%
if __name__ == '__main__':
from ddtn.helper.utility import get_cat, show_images
# Load im and create a batch of imgs
N = 15
im = get_cat()
im = np.tile(im, (N, 1, 1, 1))
# Create transformation vector
theta = np.tile(np.array([1, 0, 0, 0, 1, 0], np.float32), (N, 1))
theta[:, 2] = np.random.normal(scale=0.5, size=N)
theta[:, 5] = np.random.normal(scale=0.5, size=N)
# Cast to tensorflow and normalize values
im_tf = tf.cast(im, tf.float32)
theta_tf = tf.cast(theta, tf.float32)
# Transformer imgs
trans_im = ST_Affine_transformer(im_tf, theta_tf, (1200, 1600))
# Run computations
sess = tf.Session()
out_im = sess.run(trans_im)
show_images(out_im, cols=3)