forked from NVlabs/alias-free-gan
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathindex.html
753 lines (691 loc) · 21.4 KB
/
index.html
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
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
<!DOCTYPE html>
<html lang="" xml:lang="" xmlns="http://www.w3.org/1999/xhtml">
<head>
<meta charset="utf-8"/>
<meta content="width=device-width, initial-scale=1" name="viewport"/>
<link href="https://nvlabs-fi-cdn.nvidia.com/_web/alias-free-gan/img/favicon.ico" rel="shortcut icon"/>
<title>
Alias-Free GAN
</title>
<meta content="Alias-Free GAN" property="og:title"/>
<meta content="We eliminate “texture sticking” in GANs through a comprehensive overhaul of all signal processing aspects of the generator, paving the way for better synthesis of video and animation." name="description" property="og:description"/>
<meta content="https://nvlabs-fi-cdn.nvidia.com/_web/alias-free-gan/img/alias-free-gan-teaser-1920x1006.png" property="og:image"/>
<meta content="https://nvlabs.github.io/alias-free-gan" property="og:url"/>
<style type="text/css">
:root {
--small-thumb-border-radius: 2px;
--larger-thumb-border-radius: 3px;
}
html {
font-size: 14px;
line-height: 1.6;
font-family: Inter, system-ui, -apple-system, BlinkMacSystemFont, "Segoe UI", Roboto, "Helvetica Neue", Arial, "Noto Sans", sans-serif, "Apple Color Emoji", "Segoe UI Emoji", "Segoe UI Symbol", "Noto Color Emoji";
text-size-adjust: 100%;
-ms-text-size-adjust: 100%;
-webkit-text-size-adjust: 100%;
}
@media(min-width: 768px) {
html {
font-size: 16px;
}
}
body {
margin: 0px;
padding: 0px;
}
.base-grid,
.n-header,
.n-byline,
.n-title,
.n-article,
.n-footer {
display: grid;
justify-items: stretch;
grid-template-columns: [screen-start] 8px [page-start kicker-start text-start gutter-start middle-start] 1fr 1fr 1fr 1fr 1fr 1fr 1fr 1fr [text-end page-end gutter-end kicker-end middle-end] 8px [screen-end];
grid-column-gap: 8px;
}
.grid {
display: grid;
grid-column-gap: 8px;
}
@media(min-width: 768px) {
.base-grid,
.n-header,
.n-byline,
.n-title,
.n-article,
.n-footer {
display: grid;
justify-items: stretch;
grid-template-columns: [screen-start] 1fr [page-start kicker-start middle-start text-start] 45px 45px 45px 45px 45px 45px 45px 45px [ kicker-end text-end gutter-start] 45px [middle-end] 45px [page-end gutter-end] 1fr [screen-end];
grid-column-gap: 16px;
}
.grid {
grid-column-gap: 16px;
}
}
@media(min-width: 1000px) {
.base-grid,
.n-header,
.n-byline,
.n-title,
.n-article,
.n-footer {
display: grid;
justify-items: stretch;
grid-template-columns: [screen-start] 1fr [page-start kicker-start] 50px [middle-start] 50px [text-start kicker-end] 50px 50px 50px 50px 50px 50px 50px 50px [text-end gutter-start] 50px [middle-end] 50px [page-end gutter-end] 1fr [screen-end];
grid-column-gap: 16px;
}
.grid {
grid-column-gap: 16px;
}
}
@media (min-width: 1180px) {
.base-grid,
.n-header,
.n-byline,
.n-title,
.n-article,
.n-footer {
display: grid;
justify-items: stretch;
grid-template-columns: [screen-start] 1fr [page-start kicker-start] 60px [middle-start] 60px [text-start kicker-end] 60px 60px 60px 60px 60px 60px 60px 60px [text-end gutter-start] 60px [middle-end] 60px [page-end gutter-end] 1fr [screen-end];
grid-column-gap: 32px;
}
.grid {
grid-column-gap: 32px;
}
}
.base-grid {
grid-column: screen;
}
/* default grid column assignments */
.n-title > * {
grid-column: text;
}
.n-article > * {
grid-column: text;
}
.n-header {
height: 0px;
}
.n-footer {
height: 60px;
}
.n-title {
padding: 4rem 0 1.5rem;
}
.l-page {
grid-column: page;
}
.l-article {
grid-column: text;
}
p {
margin-top: 0;
margin-bottom: 1em;
}
.pixelated {
image-rendering: pixelated;
}
strong {
font-weight: 600;
}
/*------------------------------------------------------------------*/
/* title */
.n-title h1 {
font-family: "Barlow",system-ui,Arial,sans-serif;
color:#082333;
grid-column: text;
font-size: 40px;
font-weight: 700;
line-height: 1.1em;
margin: 0 0 0.5rem;
text-align: center;
}
@media (min-width: 768px) {
.n-title h1 {
font-size: 50px;
}
}
/*------------------------------------------------------------------*/
/* article */
.n-article {
color: rgb(33, 40, 53);
border-top: 1px solid rgba(0, 0, 0, 0.1);
padding-top: 2rem;
}
.n-article h2 {
contain: layout style;
font-weight: 600;
font-size: 24px;
line-height: 1.25em;
margin: 2rem 0 1.5rem 0;
border-bottom: 1px solid rgba(0, 0, 0, 0.1);
padding-bottom: 1rem;
}
@media (min-width: 768px) {
.n-article {
line-height: 1.7;
}
.n-article h2 {
font-size: 36px;
}
}
/*------------------------------------------------------------------*/
/* byline */
.n-byline {
contain: style;
overflow: hidden;
border-top: 1px solid rgba(0, 0, 0, 0.1);
font-size: 0.8rem;
line-height: 1.8em;
padding: 1.5rem 0;
min-height: 1.8em;
}
.n-byline .byline {
grid-column: text;
}
.byline {
grid-template-columns: 1fr 1fr 1fr 1fr;
}
.grid {
display: grid;
grid-column-gap: 8px;
}
@media (min-width: 768px) {
.grid {
grid-column-gap: 16px;
}
}
.n-byline p {
margin: 0;
}
.n-byline h3 {
font-size: 0.6rem;
font-weight: 400;
color: rgba(0, 0, 0, 0.5);
margin: 0;
text-transform: uppercase;
}
.n-byline .authors-affiliations {
grid-column-end: span 2;
grid-template-columns: 1fr 1fr;
}
/*------------------------------------------------------------------*/
/* figures */
.figure {
margin-top: 1.5rem;
margin-bottom: 1rem;
}
figcaption, .figcaption {
color: rgba(0, 0, 0, 0.6);
font-size: 12px;
line-height: 1.5em;
}
ul.authors {
list-style-type: none;
padding: 0;
margin: 0;
text-align: center;
}
ul.authors li {
padding: 0 0.5rem;
display: inline-block;
}
ul.authors sup {
color: rgb(126,126,126);
}
ul.authors.affiliations {
margin-top: 0.5rem;
}
ul.authors.affiliations li {
color: rgb(126,126,126);
}
/* Download section columns. This switches between two layouts::after
- two columns on larger viewport sizes: side-by-side paper thumb and links
- single column: no thumb
*/
.download-section {
display: grid;
grid-template-areas: "links";
}
.download-section h4 {
margin-left: 2.5rem;
display: block;
}
.download-thumb {
grid-area: thumb;
display: none;
}
.download-links {
grid-area: links;
}
img.dropshadow {
box-shadow: 0 1px 10px rgba(0,0,0, 0.3);
}
@media(min-width: 1180px) {
.download-section {
display: grid;
grid-template-areas: "thumb links";
}
.download-thumb {
display: block;
}
}
/* For BibTeX */
pre {
font-size: 0.9em;
padding-left: 7px;
padding-right: 7px;
padding-top: 3px;
padding-bottom: 3px;
border-radius: 3px;
background-color: rgb(235, 235, 235);
overflow-x: auto;
}
/* video caption */
.video {
margin-top: 1.5rem;
margin-bottom: 1.5rem;
}
.videocaption {
display: flex;
font-size: 16px;
line-height: 1.5em;
margin-bottom: 1rem;
justify-content: center;
}
.disable-selection {
user-select: none;
-moz-user-select: none; /* Firefox */
-ms-user-select: none; /* Internet Explorer */
-khtml-user-select: none; /* KHTML browsers (e.g. Konqueror) */
-webkit-user-select: none; /* Chrome, Safari, and Opera */
-webkit-touch-callout: none; /* Disable Android and iOS callouts*/
}
.hidden {
display: none;
}
h3.figtitle {
margin-top: 0;
margin-bottom: 0;
}
.fig-title-line {
grid-template-columns: 2fr 0.75fr;
}
.fig-thumb-image-row {
grid-template-columns: 1fr 1fr;
grid-template-rows: 1fr;
}
.fig-thumb-image-row-item {
width: 100%;
min-height: auto;
border-radius: var(--small-thumb-border-radius);
}
.fig-dataset-button {
border-color: rgba(0,0,0,0);
border-width: 1px;
border-style: solid;
cursor: pointer;
opacity: 0.6;
}
.fig-dataset-button.active {
border-color: rgba(0,0,0,0.7);
border-width: 1px;
border-style: solid;
opacity: 1.0;
}
.grid {
display: grid;
grid-column-gap: 8px;
}
.fig-3-image-row {
margin-top: 1em;
grid-template-columns: 1fr 1.3fr 1fr;
grid-template-rows: 1fr;
}
.fig-3-image-item {
justify-self: center;
align-self: center;
width: 100%;
border-radius: var(--larger-thumb-border-radius);
}
/*---------------------------------------------------------------------*/
.fig-slider {
grid-template-columns: auto 2fr;
grid-template-rows: 1fr;
margin-top: 1em;
align-items: start;
justify-content: center;
}
.fig-slider img.play_button {
margin-right: 8px;
cursor: pointer;
justify-self: center;
}
.fig-slider svg {
touch-action: none;
}
.fig-preload {
display: none;
}
/*---------------------------------------------------------------------*/
</style>
<!-- inline stylesheet files into the above <style> element -->
<link href="https://fonts.googleapis.com/css?family=Montserrat|Segoe+UI" rel="stylesheet"/>
<link as="image" href="https://nvlabs-fi-cdn.nvidia.com/_web/alias-free-gan/img/alias-free-gan-teaser-1920x1006.png" rel="preload"/>
<link as="image" href="https://nvlabs-fi-cdn.nvidia.com/_web/alias-free-gan/img/paper-pdf-512.png" rel="preload"/>
</head>
<body>
<div class="n-header">
</div>
<div class="n-title">
<h1>
Alias-Free GAN
</h1>
</div>
<div class="n-byline">
<div class="byline">
<ul class="authors">
<li>
Tero Karras
<sup>
1
</sup>
</li>
<li>
Miika Aittala
<sup>
1
</sup>
</li>
<li>
Samuli Laine
<sup>
1
</sup>
</li>
<li>
Erik Härkönen
<sup>
2, 1
</sup>
</li>
<li>
Janne Hellsten
<sup>
1
</sup>
</li>
<li>
Jaakko Lehtinen
<sup>
1, 2
</sup>
</li>
<li>
Timo Aila
<sup>
1
</sup>
</li>
</ul>
<ul class="authors affiliations">
<li>
<sup>
1
</sup>
NVIDIA
</li>
<li>
<sup>
2
</sup>
Aalto University
</li>
</ul>
</div>
</div>
<div class="n-article">
<div class="l-article">
<img src="https://nvlabs-fi-cdn.nvidia.com/_web/alias-free-gan/img/alias-free-gan-teaser-1920x1006.png" width="100%"/>
</div>
<h2 id="abstract">
Abstract
</h2>
<p>
We observe that despite their hierarchical convolutional nature, the synthesis process of typical generative adversarial networks depends on absolute pixel coordinates in an unhealthy manner. This manifests itself as, e.g., detail appearing to be glued to image coordinates instead of the surfaces of depicted objects. We trace the root cause to careless signal processing that causes aliasing in the generator network. Interpreting all signals in the network as continuous, we derive generally applicable, small architectural changes that guarantee that unwanted information cannot leak into the hierarchical synthesis process. The resulting networks match the FID of StyleGAN2 but differ dramatically in their internal representations, and they are fully equivariant to translation and rotation even at subpixel scales. Our results pave the way for generative models better suited for video and animation.
</p>
<h2 id="links">
Links
</h2>
<div class="grid download-section">
<div class="download-thumb">
<a href="https://nvlabs-fi-cdn.nvidia.com/alias-free-gan/alias-free-gan-paper.pdf">
<img class="dropshadow" src="https://nvlabs-fi-cdn.nvidia.com/_web/alias-free-gan/img/paper-pdf-512.png"/>
</a>
</div>
<div class="download-links">
<ul>
<li>
<a href="https://nvlabs-fi-cdn.nvidia.com/alias-free-gan/alias-free-gan-paper.pdf">
Paper PDF
</a>
</li>
<li>
<a href="https://arxiv.org/abs/2106.12423">
arXiv
</a>
</li>
<li>
<a href="https://github.com/NVlabs/alias-free-gan">
Code release on GitHub
</a>
</li>
</ul>
</div>
</div>
<h2 id="videos">
Videos
</h2>
<p>
The first two videos demonstrate the “texture sticking” issue in in two “cinemagraphs” created using generators trained on the unaligned FFHQ-U dataset. The looping videos show small random walks around a central point in the latent space. Observe how the details (hairs, wrinkles, etc.) the StyleGAN2 result (left) appear to be glued to the screen coordinates while the face moves under it, while all details transform coherently in our result (right).
</p>
<div class="l-page video">
<video controls="" loop="" width="100%">
<!-- t=0.001 is a hack to make iPhone show video thumbnail -->
<source src="https://nvlabs-fi-cdn.nvidia.com/_web/alias-free-gan/videos/video_0_ffhq_cinemagraphs.mp4#t=0.001" type="video/mp4"/>
</video>
<div class="videocaption">
<div>
<strong>
Video 1a:
</strong>
FFHQ-U Cinemagraph
</div>
</div>
</div>
<div class="l-page video">
<video controls="" loop="" width="100%">
<!-- t=0.001 is a hack to make iPhone show video thumbnail -->
<source src="https://nvlabs-fi-cdn.nvidia.com/_web/alias-free-gan/videos/video_1_ffhq_cinemagraphs.mp4#t=0.001" type="video/mp4"/>
</video>
<div class="videocaption">
<div>
<strong>
Video 1b:
</strong>
FFHQ-U Cinemagraph
</div>
</div>
</div>
<p>
The following videos show interpolations between hand-picked latent points in several datasets. Observe again how the textural detail appears fixed in the StyleGAN2 result, but transforms smoothly with the rest of the scene in the alias-free generator.
</p>
<div class="l-page video">
<video controls="" loop="" width="100%">
<!-- t=0.001 is a hack to make iPhone show video thumbnail -->
<source src="https://nvlabs-fi-cdn.nvidia.com/_web/alias-free-gan/videos/video_2_metfaces_interpolations.mp4#t=0.001" type="video/mp4"/>
</video>
<div class="videocaption">
<div>
<strong>
Video 2:
</strong>
MetFaces interpolations
</div>
</div>
</div>
<div class="l-page video">
<video controls="" loop="" width="100%">
<source src="https://nvlabs-fi-cdn.nvidia.com/_web/alias-free-gan/videos/video_3_afhq_interpolations.mp4#t=0.001" type="video/mp4"/>
</video>
<div class="videocaption">
<div>
<strong>
Video 3:
</strong>
AFHQv2 interpolations
</div>
</div>
</div>
<p>
We note, in particular, how our alias-free generator appears to have learned to mimic camera motion in the Beaches dataset.
</p>
<div class="l-page video">
<video controls="" loop="" width="100%">
<source src="https://nvlabs-fi-cdn.nvidia.com/_web/alias-free-gan/videos/video_4_beaches_interpolations.mp4#t=0.001" type="video/mp4"/>
</video>
<div class="videocaption">
<div>
<strong>
Video 4:
</strong>
Beaches interpolations
</div>
</div>
</div>
<p>
The following video illustrates translational equivariance, or lack thereof, in several “bridge” configurations, and aims to visually demonstrate the meaning of EQ-T equivariance scores. In all panels, the first image is the result of running the corresponding generator with analytically translated Fourier input features. The second image has obtained from the first by “untransforming” the pixels using the inverse translation by an extremely high-quality resampling filter. For a perfectly equivariant generator, the first two images are the same, modulo image boundaries (not shown due to light cropping) and numerical noise from the resampling. The third image visualizes the difference of the first two images. As can be seen, EQ-T scores in the 60 dB range are essentially visually perfect. Please consult the Appendix for technical details.
</p>
<div class="l-page video">
<video controls="" loop="" width="100%">
<source src="https://nvlabs-fi-cdn.nvidia.com/_web/alias-free-gan/videos/video_5_figure_3_left_equivariance_quality.mp4#t=0.001" type="video/mp4"/>
</video>
<div class="videocaption">
<div>
<strong>
Video 5:
</strong>
Visualization of translation equivariance (Figure 3, left)
</div>
</div>
</div>
<p>
The following video illustrates rotation equivariance in a manner similar to the previous video. Our alias-free Config-T, which has only been designed for translation equivariance, fails completely, as expected. The following comparison method is a variant of Config-T that uses a p4 symmetric G-CNN for rotation equivariance. The model shows a cyclic behavior, where the rotation is exact at multiples of 90 degrees but breaks down at intermediate angles Our alias-free Config-R features high-quality, though not visually perfect rotation equivariance.
</p>
<div class="l-article video">
<video controls="" loop="" width="100%">
<source src="https://nvlabs-fi-cdn.nvidia.com/_web/alias-free-gan/videos/video_6_figure_5_right_g-cnn_comparison.mp4#t=0.001" type="video/mp4"/>
</video>
<div class="videocaption">
<div>
<strong>
Video 6:
</strong>
Visualization of rotation equivariance (Figure 5, right)
</div>
</div>
</div>
<p>
The following video illustrates the aliasing inherent to pointwise nonlinearities (here, ReLU), and our solution.
<strong>
Left column:
</strong>
The original bandlimited signal z. Its ideal version (top) is sampled (middle), and then reconstructed from the samples (bottom). As the sampling rate is high enough to capture the signal, no aliasing occurs.
<strong>
Middle column:
</strong>
applying a pointwise non-linearity in the continuous domain (top) yields a non-smooth function due to clipping at the zero crossings. Sampling this signal (middle) and reconstructing the function from the samples (bottom) yields an aliased result, as the high frequencies created by the clipping cannot be represented by the sample grid.
<strong>
Right column:
</strong>
applying a low-pass filter to the ReLUed function in the continuous domain (top) yields again a smooth function; sampling it (middle) allows a faithful reconstruction (bottom).
</p>
<div class="l-article video">
<video controls="" loop="" width="100%">
<source src="https://nvlabs-fi-cdn.nvidia.com/_web/alias-free-gan/videos/video_7_figure_2_right_filtered_nonlinearity.mp4#t=0.001" type="video/mp4"/>
</video>
<div class="videocaption">
<div>
<strong>
Video 7:
</strong>
Visualization of the filtered non-linearity
</div>
</div>
</div>
<p>
The below video compares alias-free generators’ internal activations to those of StyleGAN2 (top). Our alias-free translation (middle) and rotation (bottom) equivariant networks build the image in a radically different manner from what appear to be multi-scale phase signals that follow the features seen in the final image. Due to our alias-free construction, these signals must control both the appearance of as well as the relative positions of image features; we hypothesize that the local oriented oscillations form a basis that enables hierarchical localization. Our construction appears to make it natural for the network to construct them from the low-frequency input Fourier features.
</p>
<div class="l-page video">
<video controls="" loop="" width="100%">
<source src="https://nvlabs-fi-cdn.nvidia.com/_web/alias-free-gan/videos/fig6-video.mp4#t=0.001" type="video/mp4"/>
</video>
<div class="videocaption">
<div>
<strong>
Video 8:
</strong>
Internal activations
</div>
</div>
</div>
<p>
The following video clarifies the slice visualization of Figure 1, right.
</p>
<div class="l-article video">
<video controls="" loop="" width="100%">
<source src="https://nvlabs-fi-cdn.nvidia.com/_web/alias-free-gan/videos/slice-vid.mp4#t=0.001" type="video/mp4"/>
</video>
<div class="videocaption">
<div>
<strong>
Video 9:
</strong>
Slice visualization. Top row: alias-free Config-T, bottom row: StyleGAN2.
</div>
</div>
</div>
<h2 id="citation">
Citation
</h2>
<pre><code>@article{Karras2021,
author = {Tero Karras and Miika Aittala and Samuli Laine and Erik H\"{a}rk\"{o}nen and Janne Hellsten and Jaakko Lehtinen and Timo Aila},
journal = {CoRR},
title = {Alias-Free Generative Adversarial Networks},
volume = {abs/2106.12423},
year = {2021}
}</code></pre>
<h2 id="license">
License
</h2>
<p>
Images, text and video files on this site are made freely available for non-commercial use under the
<a href="https://nvlabs-fi-cdn.nvidia.com/_web/alias-free-gan/LICENSE.txt">
Creative Commons CC BY-NC 4.0 license
</a>
. Feel free to use any of the material in your own work, as long as you give us appropriate credit by mentioning the title and author list of our paper.
</p>
<h2 id="acknowledgments">
Acknowledgments
</h2>
<p>
We thank David Luebke, Ming-Yu Liu, Koki Nagano, Tuomas Kynkäänniemi, and Timo Viitanen for reviewing early drafts and helpful suggestions. Frédo Durand for early discussions. Tero Kuosmanen for maintaining our compute infrastructure. AFHQ authors for an updated version of their dataset. Getty Images for the training images in the BEACHES dataset.
</p>
</div>
<div class="n-footer">
</div>
</body>
</html>