File size: 10,355 Bytes
29cdbe6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Copyright 2020 Erik Härkönen. All rights reserved.\n",
    "# This file is licensed to you under the Apache License, Version 2.0 (the \"License\");\n",
    "# you may not use this file except in compliance with the License. You may obtain a copy\n",
    "# of the License at http://www.apache.org/licenses/LICENSE-2.0\n",
    "\n",
    "# Unless required by applicable law or agreed to in writing, software distributed under\n",
    "# the License is distributed on an \"AS IS\" BASIS, WITHOUT WARRANTIES OR REPRESENTATIONS\n",
    "# OF ANY KIND, either express or implied. See the License for the specific language\n",
    "# governing permissions and limitations under the License.\n",
    "\n",
    "# Teaser: sequence of 3 interesting edits\n",
    "%matplotlib inline\n",
    "from notebook_init import *\n",
    "\n",
    "rand = lambda : np.random.randint(np.iinfo(np.int32).max)\n",
    "outdir = Path('out/figures/teaser')\n",
    "makedirs(outdir, exist_ok=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "def setup_model(model_name, class_name, layer_name):\n",
    "    global inst, model, lat_comp, lat_mean, lat_std\n",
    "\n",
    "    use_w = 'StyleGAN' in model_name\n",
    "    inst = get_instrumented_model(model_name, class_name, layer_name, device, use_w=use_w, inst=inst)\n",
    "    model = inst.model\n",
    "\n",
    "    pc_config = Config(components=80, n=1_000_000, batch_size=200,\n",
    "        layer=layer_name, model=model_name, output_class=class_name, use_w=use_w)\n",
    "    dump_name = get_or_compute(pc_config, inst)\n",
    "\n",
    "    with np.load(dump_name) as data:\n",
    "        lat_comp = data['lat_comp']\n",
    "        lat_mean = data['lat_mean']\n",
    "        lat_std = data['lat_stdev']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "def perform_edit(seeds, edit_sequence, save_images=False, crop=None):\n",
    "    max_figs = 1000 if save_images else 10\n",
    "\n",
    "    for seed in seeds[:max_figs]:\n",
    "        w = model.sample_latent(1, seed=seed).cpu().numpy()\n",
    "        w = [w]*model.get_max_latents()\n",
    "        imgs = []\n",
    "        \n",
    "        # Starting point\n",
    "        imgs.append(model.sample_np(w))\n",
    "        \n",
    "        # Perform edits in order\n",
    "        for edit in edit_sequence:\n",
    "            (idx, start, end, strength, invert) = configs[edit]\n",
    "            \n",
    "            # Find out coordinate of w along PC\n",
    "            w_centered = w[0] - lat_mean\n",
    "            w_coord = np.sum(w_centered.reshape(-1)*lat_comp[idx].reshape(-1)) / lat_std[idx]\n",
    "            \n",
    "            # Invert property if desired (e.g. flip rotation)\n",
    "            # Otherwise reinforce existing\n",
    "            if invert:\n",
    "                sign = w_coord / np.abs(w_coord)\n",
    "                target = -sign*strength # opposite side of mean\n",
    "            else:\n",
    "                target = strength\n",
    "                \n",
    "            delta = target - w_coord # offset vector\n",
    "            \n",
    "            for l in range(start, end):\n",
    "                w[l] = w[l] + lat_comp[idx]*lat_std[idx]*delta\n",
    "            imgs.append(model.sample_np(w))\n",
    "        \n",
    "        # Crop away black borders\n",
    "        if crop:\n",
    "            imgs = [img[crop[0]:-crop[1], crop[2]:-crop[3], :] for img in imgs]\n",
    "\n",
    "        if save_images:\n",
    "            # Save to disk\n",
    "            for i, img in enumerate(imgs):\n",
    "                Image.fromarray((img*255).astype(np.uint8)).save(outdir / f'teaser_{seed}_{i}.png')\n",
    "        \n",
    "        # Show in notebook\n",
    "        strip = np.hstack(imgs)\n",
    "        #strip = strip[::2, ::2, :] # 2x downscale for preview\n",
    "        plt.figure(figsize=(30,5))\n",
    "        plt.imshow(strip, interpolation='bilinear')\n",
    "        plt.axis('off')\n",
    "        plt.show()\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# (idx, edit_start, edit_end, strength, invert)\n",
    "configs = {\n",
    "    # StyleGAN2 cars W\n",
    "    'Redness':          (22,  9, 11,   -8, False),\n",
    "    'Horizontal flip':  ( 0,  0,  5,  2.0, True),\n",
    "    'Add grass':        (41,  9, 11,  -18, False),\n",
    "    'Blocky shape':     (16,  3,  6,   25, False),\n",
    "\n",
    "    # BigGAN-512 irish_setter\n",
    "    'Move right':       ( 0,  0, 15, -1.5, False),\n",
    "    'Rotate':           ( 3,  0, 15,  -0.5, False),\n",
    "    'Move back':        ( 4,  0, 15,  2.5, False),\n",
    "    'Zoom in':          ( 6,  0, 15, -2.0, False),\n",
    "    'Zoom out':         (12,  0, 15, -4.0, False),\n",
    "    'Sharpen BG':       (13,  6,  9, 20.0, False),\n",
    "    'Camera down':      (15,  1,  6, -4.0, False),\n",
    "    'Light right':      (28,  7,  8,  30, False),\n",
    "    'Pixelate':         (46, 10, 11,  -25, False),\n",
    "    'Reeds':            (61,  4,  8,  -15, False),\n",
    "    'Dry BG':           (65,  6,  8,  -30, False),\n",
    "    'Grass length':     (69,  5,  8,   15, False),\n",
    "\n",
    "    # StyleGAN2 ffhq\n",
    "    'frizzy_hair':             (31,  2,  6,  20, False),\n",
    "    'background_blur':         (49,  6,  9,  20, False),\n",
    "    'bald':                    (21,  2,  5,  20, False),\n",
    "    'big_smile':               (19,  4,  5,  20, False),\n",
    "    'caricature_smile':        (26,  3,  8,  13, False),\n",
    "    'scary_eyes':              (33,  6,  8,  20, False),\n",
    "    'curly_hair':              (47,  3,  6,  20, False),\n",
    "    'dark_bg_shiny_hair':      (13,  8,  9,  20, False),\n",
    "    'dark_hair_and_light_pos': (14,  8,  9,  20, False),\n",
    "    'dark_hair':               (16,  8,  9,  20, False),\n",
    "    'disgusted':               (43,  6,  8, -30, False),\n",
    "    'displeased':              (36,  4,  7,  20, False),\n",
    "    'eye_openness':            (54,  7,  8,  20, False),\n",
    "    'eye_wrinkles':            (28,  6,  8,  20, False),\n",
    "    'eyebrow_thickness':       (37,  8,  9,  20, False),\n",
    "    'face_roundness':          (37,  0,  5,  20, False),\n",
    "    'fearful_eyes':            (54,  4, 10,  20, False),\n",
    "    'hairline':                (21,  4,  5, -20, False),\n",
    "    'happy_frizzy_hair':       (30,  0,  8,  20, False),\n",
    "    'happy_elderly_lady':      (27,  4,  7,  20, False),\n",
    "    'head_angle_up':           (11,  1,  4,  20, False),\n",
    "    'huge_grin':               (28,  4,  6,  20, False),\n",
    "    'in_awe':                  (23,  3,  6, -15, False),\n",
    "    'wide_smile':              (23,  3,  6,  20, False),\n",
    "    'large_jaw':               (22,  3,  6,  20, False),\n",
    "    'light_lr':                (15,  8,  9,  10, False),\n",
    "    'lipstick_and_age':        (34,  6, 11,  20, False),\n",
    "    'lipstick':                (34, 10, 11,  20, False),\n",
    "    'mascara_vs_beard':        (41,  6,  9,  20, False),\n",
    "    'nose_length':             (51,  4,  5, -20, False),\n",
    "    'elderly_woman':           (34,  6,  7,  20, False),\n",
    "    'overexposed':             (27,  8, 18,  15, False),\n",
    "    'screaming':               (35,  3,  7, -15, False),\n",
    "    'short_face':              (32,  2,  6, -20, False),\n",
    "    'show_front_teeth':        (59,  4,  5,  40, False),\n",
    "    'smile':                   (46,  4,  5, -20, False),\n",
    "    'straight_bowl_cut':       (20,  4,  5, -20, False),\n",
    "    'sunlight_in_face':        (10,  8,  9,  10, False),\n",
    "    'trimmed_beard':           (58,  7,  9,  20, False),\n",
    "    'white_hair':              (57,  7, 10, -24, False),\n",
    "    'wrinkles':                (20,  6,  7, -18, False),\n",
    "    'boyishness':              (8,   2,  5,  20, False),\n",
    "}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# StyleGAN2 faces - emphasis on novel edits\n",
    "setup_model('StyleGAN2', 'ffhq', 'style')\n",
    "model.truncation = 0.7\n",
    "model.use_w()\n",
    "\n",
    "seeds = [6293435, 2105448342] # + [rand() for _ in range(1)]\n",
    "print(seeds)\n",
    "edits = ['wrinkles', 'white_hair', 'in_awe', 'overexposed']\n",
    "perform_edit(seeds, edits, True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# StyleGAN2 cars\n",
    "setup_model('StyleGAN2', 'car', 'style')\n",
    "model.truncation = 0.6\n",
    "model.use_w()\n",
    "\n",
    "seeds = [440749230] # + [rand() for _ in range(10)]\n",
    "edits = ['Redness', 'Horizontal flip', 'Add grass', 'Blocky shape']\n",
    "perform_edit(seeds, edits, True, crop=[64, 64, 1, 1])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# BigGAN-512 irish setter\n",
    "setup_model('BigGAN-512', 'husky', 'generator.gen_z')\n",
    "model.set_output_class('irish_setter')\n",
    "model.truncation = 0.6\n",
    "\n",
    "seeds = [489408325]# + [rand() for _ in range(10)]\n",
    "edits = ['Rotate', 'Zoom out', 'Camera down', 'Reeds']\n",
    "perform_edit(seeds, edits, True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.7"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}