1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
|
import click
from app.utils import click_utils
from app.settings import app_cfg
import os
from os.path import join
import time
import numpy as np
import random
from subprocess import call
import cv2 as cv
from PIL import Image
from glob import glob
import tensorflow as tf
import tensorflow_hub as hub
import shutil
import h5py
from app.search.json import save_params_latent, save_params_dense
from app.search.image import image_to_uint8, imconvert_uint8, imconvert_float32, \
imread, imwrite, imgrid
from app.search.vector import truncated_z_sample, truncated_z_single, create_labels
@click.command('')
@click.option('-i', '--input', 'opt_fp_in', required=True,
help='Path to input image')
@click.option('-s', '--dims', 'opt_dims', default=128, type=int,
help='Dimensions of BigGAN network (128, 256, 512)')
@click.option('-v', '--video', 'opt_video', is_flag=True,
help='Export a video for each dataset')
@click.option('-t', '--tag', 'opt_tag', default='inverse_' + str(int(time.time() * 1000)),
help='Tag this dataset')
# @click.option('-r', '--recursive', 'opt_recursive', is_flag=True)
@click.pass_context
def cli(ctx, opt_fp_in, opt_dims, opt_video, opt_tag):
"""
Search for an image (class vector) in BigGAN using gradient descent
"""
generator = hub.Module('https://tfhub.dev/deepmind/biggan-' + str(opt_dims) + '/2')
inputs = {k: tf.compat.v1.placeholder(v.dtype, v.get_shape().as_list(), k)
for k, v in generator.get_input_info_dict().items()}
input_z = inputs['z']
input_y = inputs['y']
input_trunc = inputs['truncation']
output = generator(inputs)
vocab_size = input_y.shape.as_list()[1]
sess = tf.compat.v1.Session()
sess.run(tf.compat.v1.global_variables_initializer())
sess.run(tf.compat.v1.tables_initializer())
if os.path.isdir(opt_fp_in):
paths = glob(os.path.join(opt_fp_in, '*.jpg')) + \
glob(os.path.join(opt_fp_in, '*.jpeg')) + \
glob(os.path.join(opt_fp_in, '*.png'))
else:
paths = [opt_fp_in]
fp_inverses = os.path.join(app_cfg.DIR_INVERSES, opt_tag)
os.makedirs(fp_inverses, exist_ok=True)
save_params_latent(fp_inverses, opt_tag)
save_params_dense(fp_inverses, opt_tag)
out_file = h5py.File(join(fp_inverses, 'dataset.hdf5'), 'w')
out_images = out_file.create_dataset('xtrain', (len(paths), 3, 512, 512,), dtype='float32')
out_labels = out_file.create_dataset('ytrain', (len(paths), vocab_size,), dtype='float32')
out_fns = out_file.create_dataset('fn', (len(paths),), dtype=h5py.string_dtype())
for path, index in enumerate(paths):
out_labels[index] = os.path.basename(path)
fp_frames = find_nearest_vector(generator, sess, input_z, input_y, input_trunc, output, path, opt_dims, out_images, out_labels, index)
if opt_video:
export_video(fp_frames)
def find_nearest_vector(generator, sess, input_z, input_y, input_trunc, output, opt_fp_in, opt_dims, out_images, out_labels, index):
"""
Find the closest latent and class vectors for an image. Store the class vector in an HDF5.
"""
z_dim = input_z.shape.as_list()[1]
vocab_size = input_y.shape.as_list()[1]
# scalar truncation value in [0.02, 1.0]
batch_size = 25
truncation = 1.0
z = truncated_z_sample(batch_size, z_dim, truncation/2)
num_classes = 1
y = create_labels(batch_size, vocab_size, num_classes)
if opt_fp_in:
print("Processing {}".format(opt_fp_in))
fn = os.path.basename(opt_fp_in)
fbase, ext = os.path.splitext(fn)
fp_frames = "frames_{}_{}".format(fbase, int(time.time() * 1000))
os.makedirs(join(app_cfg.DIR_OUTPUTS, fp_frames), exist_ok=True)
target_im = imread(opt_fp_in)
# crop image to 512 and save for later processing
phi_target_for_inversion = resize_and_crop_image(target_im, 512)
b = np.dsplit(phi_target_for_inversion, 3)
phi_target_for_inversion = np.stack(b).reshape((3, 512, 512))
out_images[index] = phi_target_for_inversion
# crop image to 128 to find vectors
phi_target = resize_and_crop_image(target_im, opt_dims)
phi_target = np.expand_dims(phi_target, 0)
phi_target = np.repeat(phi_target, batch_size, axis=0)
else:
print("Processing random vector")
fp_frames = "frames_{}".format(int(time.time() * 1000))
os.makedirs(join(app_cfg.DIR_OUTPUTS, fp_frames), exist_ok=True)
z_target = np.repeat(truncated_z_single(z_dim, truncation), batch_size, axis=0)
y_target = np.repeat(create_labels(1, vocab_size, 1), batch_size, axis=0)
feed_dict = {input_z: z_target, input_y: y_target, input_trunc: truncation}
phi_target = sess.run(output, feed_dict=feed_dict)
target_im = imgrid(imconvert_uint8(phi_target), cols=5)
imwrite(join(app_cfg.DIR_OUTPUTS, fp_frames, 'frame_0000_target.png'), target_im)
#dy_dx = g.gradient(y, x)
cost = tf.reduce_sum(tf.pow(output - phi_target, 2))
dc_dz, dc_dy, = tf.gradients(cost, [input_z, input_y])
#dc_dy, = tf.gradients(cost, [input_y])
lr_z = 0.0001
lr_y = 0.000001
#z = truncated_z_sample(batch_size, z_dim, truncation/2)
feed_dict = {input_z: z, input_y: y, input_trunc: truncation}
phi_start = sess.run(output, feed_dict=feed_dict)
start_im = imgrid(imconvert_uint8(phi_start), cols=5)
imwrite(join(app_cfg.DIR_OUTPUTS, fp_frames, 'frame_0000_start.png'), start_im)
try:
for i in range(500):
feed_dict = {input_z: z, input_y: y, input_trunc: truncation}
grad_z, grad_y = sess.run([dc_dz, dc_dy], feed_dict=feed_dict)
z -= grad_z * lr_z
y -= grad_y * lr_y
lr_z *= 0.997
lr_y *= 0.999
if i % 30 == 0:
# lr_y *= 1.002
y = np.clip(y, 0, 1)
# for j in range(batch_size):
# y[j] /= y[j].sum()
if i > 200 and i % 100 == 0:
mean = np.mean(y, axis=0)
y = y * 3 / 4 + mean / 4
indices = np.logical_or(z <= -2*truncation, z >= +2*truncation)
z[indices] = np.random.randn(np.count_nonzero(indices))
feed_dict = {input_z: z, input_y: y, input_trunc: truncation}
phi_guess = sess.run(output, feed_dict=feed_dict)
guess_im = imgrid(imconvert_uint8(phi_guess), cols=5)
imwrite(join(app_cfg.DIR_OUTPUTS, fp_frames, 'frame_{:04d}.png'.format(i)), guess_im)
if i % 20 == 0:
print('lr: {}, iter: {}, grad_z: {}, grad_y: {}'.format(lr_z, i, np.std(grad_z), np.std(grad_y)))
except KeyboardInterrupt:
pass
out_labels[index] = y
return fp_frames
def export_video(fp_frames):
print("Exporting video...")
cmd = [
'/home/lens/bin/ffmpeg',
'-y', # '-v', 'quiet',
'-r', '30',
'-i', join(app_cfg.DIR_OUTPUTS, fp_frames, 'frame_%04d.png'),
'-pix_fmt', 'yuv420p',
join(app_cfg.DIR_OUTPUTS, fp_frames + '.mp4')
]
# print(' '.join(cmd))
call(cmd)
shutil.rmtree(join(app_cfg.DIR_OUTPUTS, fp_frames))
|
