-
Notifications
You must be signed in to change notification settings - Fork 4
/
Copy pathindex.html
657 lines (584 loc) · 30.2 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
<!DOCTYPE html>
<html>
<head>
<meta charset="utf-8">
<!-- Meta tags for social media banners, these should be filled in appropriatly as they are your "business card" -->
<!-- Replace the content tag with appropriate information -->
<!-- <meta name="description" content="DESCRIPTION META TAG"> -->
<meta property='og:title' content='Diffusion2GAN for One-step Text-to-Image Synthesis. arXiv2024'/>
<meta property='og:url' content='https://mingukkang.github.io/Diffusion2GAN/'/>
<meta property="og:image" content="static/images/banner.png" />
<meta property="og:image:width" content="1200" />
<meta property="og:image:height" content="630" />
<!-- <meta name="twitter:title" content="TWITTER BANNER TITLE META TAG">
<meta name="twitter:description" content="TWITTER BANNER DESCRIPTION META TAG"> -->
<!-- Path to banner image, should be in the path listed below. Optimal dimenssions are 1200X600-->
<!-- <meta name="twitter:image" content="static/images/your_twitter_banner_image.png"> -->
<!-- <meta name="twitter:card" content="summary_large_image"> -->
<!-- Keywords for your paper to be indexed by-->
<!-- <meta name="keywords" content="KEYWORDS SHOULD BE PLACED HERE"> -->
<meta name="viewport" content="width=device-width, initial-scale=1">
<title>Distilling Diffusion Models into Conditional GANs</title>
<link href="https://fonts.googleapis.com/css?family=Google+Sans|Noto+Sans|Castoro"
rel="stylesheet">
<link rel="stylesheet" href="./static/css/bulma.min.css">
<link rel="stylesheet" href="./static/css/bulma-carousel.min.css">
<link rel="stylesheet" href="./static/css/bulma-slider.min.css">
<link rel="stylesheet" href="./static/css/tab_gallery.css">
<link rel="stylesheet" href="./static/css/fontawesome.all.min.css">
<link rel="stylesheet"
href="https://cdn.jsdelivr.net/gh/jpswalsh/academicons@1/css/academicons.min.css">
<link rel="stylesheet" href="./static/css/index.css">
<link rel="icon" href="./static/images/favicon.svg">
<link rel="stylesheet" href="juxtapose/css/juxtapose.css">
<script src="https://ajax.googleapis.com/ajax/libs/jquery/3.5.1/jquery.min.js"></script>
<script defer src="./static/js/fontawesome.all.min.js"></script>
<script src="./static/js/bulma-carousel.min.js"></script>
<script src="./static/js/bulma-slider.min.js"></script>
<script src="./static/js/index.js"></script>
<script src="./static/js/magnifier.js"></script>
<link href="https://fonts.cdnfonts.com/css/menlo" rel="stylesheet">
<link rel="stylesheet" href="./static/css/image_card_fader.css">
<link rel="stylesheet" href="./static/css/image_card_slider.css">
</head>
<body>
<section class="hero banner">
<div class="hero-body">
<div class="container is-max-desktop">
<div class="columns is-centered">
<div class="column has-text-centered">
<h1 class="title is-2 publication-title">Diffusion2GAN:<br> Distilling Diffusion Models into Conditional GANs</h1>
<div class="is-size-5 publication-authors">
<span class="author-block">
<a href="https://mingukkang.github.io/">Minguk Kang</a><sup>1,2</sup>,</span>
</span>
<span class="author-block">
<a href="https://richzhang.github.io/">Richard Zhang</a><sup>2</sup>,
</span>
<span class="author-block">
<a href="https://www.connellybarnes.com/work/">Connelly Barnes</a><sup>2</sup>,
</span>
<span class="author-block">
<a href="https://research.adobe.com/person/sylvain-paris/">Sylvain Paris</a><sup>2</sup>,
</span>
<span class="author-block">
<a href="https://suhakwak.github.io/">Suha Kwak</a><sup>1</sup>,
</span>
<span class="author-block">
<a href="https://jaesik.info/">Jaesik Park</a><sup>3</sup>,
</span>
<span class="author-block">
<a href="https://research.adobe.com/person/eli-shechtman/">Eli Shechtman</a><sup>2</sup>,
</span>
<span class="author-block">
<a href="https://www.cs.cmu.edu/~junyanz/">Jun-Yan Zhu</a><sup>4</sup>,</span>
</span>
<span class="author-block">
<a href="https://taesung.me/">Taesung Park</a><sup>2</sup>
</span>
</div>
<div class="is-size-5 publication-authors">
<span class="author-block"><sup>1</sup>POSTECH,</span>
<span class="author-block"><sup>2</sup>Adobe Research,</span>
<span class="author-block"><sup>3</sup>Seoul National University,</span>
<span class="author-block"><sup>4</sup>Carnegie Mellon University</span>
</div>
<div class="is-size-5 publication-venue">
in ECCV 2024
</div>
<div class="column has-text-centered">
<div class="publication-links">
<span class="link-block">
<a href="static/paper/diffusion2gan_arxiv_v2.pdf"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="fas fa-file-pdf"></i>
</span>
<span>Paper</span>
</a>
</span>
<span class="link-block">
<a href="https://arxiv.org/abs/2405.05967"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="ai ai-arxiv"></i>
</span>
<span>arXiv</span>
</a>
</span>
<!-- Code Link. -->
<span class="link-block">
<a href="https://github.com/mingukkang/elatentlpips"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="fab fa-github"></i>
</span>
<span>E-LatentLPIPS</span>
</a>
</span>
<!-- Video Link. -->
<span class="link-block">
<a href="https://www.youtube.com/watch?v=nHav5H4uBgA"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="fab fa-youtube"></i>
</span>
<span>Video</span>
</a>
</span>
<span class="link-block">
<a href="static/slides/[PPT]Diffusion2GAN.pdf"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="fas fa-file-pdf"></i>
</span>
<span>Slides</span>
</a>
</span>
<span class="link-block">
<a href="static/slides/[poster]Diffusion2GAN.pdf"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="fas fa-file-pdf"></i>
</span>
<span>Poster</span>
</a>
</span>
<span class="link-block">
<a href="#bibtex"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="ai ai-obp"></i>
</span>
<span>BibTex</span>
</a>
</span>
</div>
</div>
</div>
</div>
</div>
</div>
</section>
<section class="hero is-light is-small">
<div class="hero-body">
<div class="container is-max-desktop has-text-centered">
<h3 class="title is-4">Diffusion2GAN can learn noise-to-image mapping of a target diffusion model</h3>
<div class="content has-text-justified">
<p>
<!--<b>A large 1B-parameter GAN model for text-to-image, with a competitive FID, disentangled latent space, and a fast upsampler for generating 4K outputs.</b>-->
</p>
</div>
<div id="results-carousel" class="carousel results-carousel">
<div class="item item-t2i0">
<img id="myt2i0" src="./static/images/ode_learning0.png"
class="interpolation-image"/>
</div>
<div class="item item-t2i0">
<img id="myt2i0" src="./static/images/ode_learning1.png"
class="interpolation-image"/>
</div>
<div class="item item-t2i0">
<img id="myt2i0" src="./static/images/ode_learning2.png"
class="interpolation-image"/>
</div>
<div class="item item-t2i1">
<img id="myt2i1" src="./static/images/ode_learning3.png"
class="interpolation-image"/>
</div>
</div>
</div>
</div>
<div class="container is-max-desktop has-text-centered">
<h3 class="title is-4">One-step image synthesis using Diffusion2GAN</h3>
<div class="content has-text-justified">
<p>
<!--<b>A large 1B-parameter GAN model for text-to-image, with a competitive FID, disentangled latent space, and a fast upsampler for generating 4K outputs.</b>-->
<b>Diffusion2GAN can generate a 512px/1024px image at an interactive speed of 0.09/0.16 second.</b> By learning the direct mapping from Gaussian noises to their corresponding images, Diffusion2GAN enables one-step image synthesis.
<br>
</p>
</div>
<div id="results-carousel" class="carousel results-carousel">
<div class="item item-t2i1">
<img id="myt2i0" src="./static/images/Text2img.png"
class="interpolation-image"/>
</div>
<div class="item item-t2i1">
<img id="myt2i1" src="./static/images/Text2img1.png"
class="interpolation-image"/>
</div>
<div class="item item-t2i1">
<img id="myt2i1" src="./static/images/Text2img2.png"
class="interpolation-image"/>
</div>
</div>
</div>
</div>
</section>
<section class="hero is-light is-small">
<div class="hero-body">
<div class="container is-max-desktop has-text-centered">
<h2 class="title is-4">Two-step 4K image generation using Diffusion2GAN and GigaGAN</h2>
<div class="content has-text-justified">
<p>
<!--<b>Diffusion2GAN is employed to produce low-resolution base images (128px) at an interactive speed of 0.16~0.17 seconds. Optionally, the GigaGAN upsampler can be introduced to enhance these images into high-quality 4k resolution.</b>-->
<b>The images generated by Diffusion2GAN can be seamlessly upsampled to 4k res using GigaGAN upsampler.</b> This indicates that we can generate low-resolution preview images using Diffusion2GAN and then enhance some preferred images to 4k resolution using the GigaGAN upsampler.
<br>
</p>
</div>
<!-- The expanding image container -->
<div class="tab_container">
<!-- Close the image -->
<!-- <span onclick="this.parentElement.style.display='none'" class="closebtn">×</span> -->
<!-- Expanded image -->
<div id="juxtapose-embed" data-startingposition="30%" data-animate="true">
</div>
<div>
<div id="juxtapose-hidden"></div>
</div>
<!-- Image text -->
<div id="imgtext"></div>
</div>
<!-- The grid: four columns -->
<div class="tab_row">
<div class="tab_column">
<img src="./static/images/corgi_input.jpeg" onclick="tab_gallery_click('corgi');">
</div>
<div class="tab_column">
<img src="./static/images/waterfall_input.jpeg" onclick="tab_gallery_click('waterfall');">
</div>
<div class="tab_column">
<img src="./static/images/jelly_input.jpeg" onclick="tab_gallery_click('jelly');">
</div>
<div class="tab_column">
<img src="./static/images/monet_input.jpeg" onclick="tab_gallery_click('monet');">
</div>
<div class="tab_column">
<img src="./static/images/unicorn_input.jpeg" onclick="tab_gallery_click('unicorn');">
</div>
<div class="tab_column">
<img src="./static/images/fire_input.jpeg" onclick="tab_gallery_click('fire');">
</div>
<div class="tab_column">
<img src="./static/images/canon_input.jpeg" onclick="tab_gallery_click('canon');">
</div>
<div class="tab_column">
<img src="./static/images/mushroom_input.jpeg" onclick="tab_gallery_click('mushroom');">
</div>
<div class="tab_column">
<img src="./static/images/castle_input.jpeg" onclick="tab_gallery_click('castle');">
</div>
<div class="tab_column">
<img src="./static/images/handsome_input.jpeg" onclick="tab_gallery_click('handsome');">
</div>
</div>
</div>
</div>
</section>
<section class="section">
<div class="container is-max-desktop">
<!-- Abstract. -->
<div class="columns is-centered has-text-centered">
<div class="column is-four-fifths">
<h2 class="title is-3">Abstract</h2>
<div class="content has-text-justified">
<p>
We propose a method to distill a complex multistep diffusion model into a single-step conditional GAN student model, dramatically accelerating inference while preserving image quality.
Our approach interprets the diffusion distillation as a paired image-to-image translation task, using noise-to-image pairs of the diffusion model's ODE trajectory. For efficient regression loss computation, we propose E-LatentLPIPS, a Perceptual loss in the latent space of the diffusion model with an ensemble of augmentations.
Despite dataset construction costs, E-LatentLPIPS converges more efficiently than many existing distillation methods.
Furthermore, we adapt a diffusion model to construct a multi-scale discriminator with a text alignment loss to build an effective conditional GAN-based formulation.
We demonstrate that our one-step generator outperforms cutting-edge one-step diffusion distillation models, DMD, SDXL-Turbo, and SDXL-Lightning, on the zero-shot COCO benchmark.
</p>
</div>
</div>
</div>
<!--/ Abstract. -->
</div>
</section>
<section class="section">
<!--/ Matting. -->
<div class="container is-max-desktop">
<!-- Latent space editing applications -->
<div class="columns is-centered">
<div class="column is-full-width">
<h2 class="title is-3 has-text-centered">Diffusion2GAN Framework</h2>
<!-- Prompt Interpolation image -->
<h3 class="title is-4">Diffusion2GAN one-step generator</h3>
<div class="content has-text-centered">
<img src="./static/images/G_architecture.png" width="90%">
</div>
<div class="content has-text-justified">
<p>
We collect diffusion model's <b>output latents</b> along with <b>input noises</b> and <b>prompts</b>. Then, the generator is trained to map <b>noise</b> and <b>prompt</b> to the <b>target latent</b> using our proposed <b>E-LatentLPIPS regression loss</b> and the <b>GAN loss</b>. While the output of the
generator can be decoded by the SD latent decoder into RGB pixels, it is a compute intensive operation that is never performed during training.
</p>
<br>
</div>
<!-- Prompt Interpolation image -->
<h3 class="title is-4">E-LatentLPIPS for latent space distillation</h3>
<div class="content has-text-centered">
<img src="./static/images/elatentlpips_v3.png" width="90%">
</div>
<div class="content has-text-justified">
<p>
Training a single iteration with <b>LPIPS</b> takes <b>117ms</b> and <b>15.0GB</b> extra memory on NVIDIA A100, whereas our <b>E-LatentLPIPS</b> requires <b>12.1ms</b> and <b>0.6GB</b> on the same device. Consequently, <b>E-latentLPIPS</b> accelerates the <b>perceptual loss computation time</b> by <b>9.7×</b> compared to <b>LPIPS</b>, while simultaneously reducing memory consumption.
</p>
<br>
</div>
<!-- Prompt Interpolation image -->
<h3 class="title is-4">Diffusion2GAN multi-scale conditional discriminator</h3>
<div class="content has-text-centered">
<img src="./static/images/D_architecture.png" width="90%">
</div>
<div class="content has-text-justified">
<p>
We reuse the <b>pre-trained weights</b> from the teacher model and augment it with <b>multi-scale input and output branches</b>. Concretely, we feed the resized version of input latents to each downsampling block of the encoder. For the decoder part, we enforce the <b>discriminator</b> to make <b>real/fake predictions</b> at <b>three places</b> at each scale: <b>before, at, and after</b> the <b>skip connection</b>. This multi-scale adversarial training further improves image quality.
</p>
<br>
</div>
</div>
</div>
<!-- Latent space editing applications -->
<div class="columns is-centered">
<div class="column is-full-width">
<h2 class="title is-3 has-text-centered">Experimental Results</h2>
<!-- Prompt Interpolation image -->
<h3 class="title is-4">Single image reconstruction using LPIPS variants</h3>
<div class="content has-text-centered">
<img src="./static/images/LatentLPIPS_recon.png" width="100%">
</div>
<div class="content has-text-justified">
<p>
We conduct an image reconstruction experiment by directly optimizing a single latent with different loss functions. Reconstruction with <b>LPIPS</b> roughly reproduces the target image, but at the cost of needing to decode into pixels. <b>LatentLPIPS</b> alone cannot precisely reconstruct the image. However, our ensembled augmentation, <b>E-LatentLPIPS</b>, can more precisely reconstruct the target while operating directly in the latent space.
</p>
</div>
<!-- Prompt Interpolation image -->
<h3 class="title is-4">Comparison with cutting-edge distillation models</h3>
<div class="content has-text-centered">
<img src="./static/images/Radar_chart_sdxl.png" width="80%">
</div>
<div class="content has-text-justified">
<p>
We compare <b>Diffusion2GAN</b> with two concurrent one-step diffusion distillation generators: <b>SDXL-turbo</b> and <b>SDXL-Lightning</b>, as well as their teacher model, <b>SDXL-Base-1.0</b>. Although <b>SDXL-Lightning</b> produces more diverse images compared to <b>Diffusion2GAN</b> and <b>SDXL-Turbo</b>, it does so at the expense of FID and CLIP-score. On the other hand, <b>Diffusion2GAN</b> demonstrates better FID and CLIP-score than <b>SDXL-Turbo</b> and <b>SDXL-Lightning</b> but generates more diverse images compared to <b>SDXL-Turbo</b>.
</p>
<br>
</div>
<!-- Prompt Interpolation image -->
<h3 class="title is-4">COCO2014 benchmark (Stable Diffusion 1.5)</h3>
<div class="content has-text-centered">
<img src="./static/images/COCO2014_benchmark.png" width="70%">
</div>
<div class="content has-text-justified">
</div>
<!-- Prompt Interpolation image -->
<h3 class="title is-4">COCO2017 benchmark (SDXL-Base-1.0)</h3>
<div class="content has-text-centered">
<img src="./static/images/COCO2017_benchmark.png" width="87%">
</div>
<div class="content has-text-justified">
</div>
<!-- Prompt Interpolation image -->
<h3 class="title is-4">Human preference study</h3>
<div class="content has-text-centered">
<img src="./static/images/human_evaluation1.png" width="100%">
</div>
<div class="content has-text-justified">
<p>
We conduct a human preference study. <b>SD1.5-Diffusion2GAN</b> produces more realistic images with better text-to-image alignment compared to <b>InstaFlow-0.9B</b>. In SDXL distillation, <b>SDXL-Diffusion2GAN</b> shows superior realism and alignment compared to <b>SDXL-Turbo</b> and <b>SDXL-Lightning</b>. However, unlike COCO2014 and COCO2017, human preference favors teacher diffusion models. We leave developing a better evaluation metric for future work.
</p>
<br>
</div>
</div>
</div>
<div class="container is-max-desktop has-text-centered">
<h2 class="title is-3">Diverse Image Generation using Diffusion2GAN</h2>
<div class="content has-text-justified">
<p>
<!--<b>A large 1B-parameter GAN model for text-to-image, with a competitive FID, disentangled latent space, and a fast upsampler for generating 4K outputs.</b>-->
<b>Diversity and text-to-image alignment of generated images from Diffusion2GAN.</b> Diffusion2GAN can generate more diverse images than SDXL-Turbo while exhibiting better text-to-image alignment than SDXL-Lightning.
<br>
</p>
</div>
<div id="results-carousel" class="carousel results-carousel">
<div class="item item-t2i1">
<img id="myt2i0" src="./static/images/Diversity_d2g0.png"
class="interpolation-image"/>
</div>
<div class="item item-t2i1">
<img id="myt2i1" src="./static/images/Diversity_turbo0.png"
class="interpolation-image"/>
</div>
<div class="item item-t2i1">
<img id="myt2i1" src="./static/images/Diversity_lightning0.png"
class="interpolation-image"/>
</div>
<div class="item item-t2i1">
<img id="myt2i1" src="./static/images/Diversity_d2g1.png"
class="interpolation-image"/>
</div>
<div class="item item-t2i1">
<img id="myt2i1" src="./static/images/Diversity_turbo1.png"
class="interpolation-image"/>
</div>
<div class="item item-t2i1">
<img id="myt2i1" src="./static/images/Diversity_lightning1.png"
class="interpolation-image"/>
</div>
<div class="item item-t2i1">
<img id="myt2i1" src="./static/images/Diversity_d2g2.png"
class="interpolation-image"/>
</div>
<div class="item item-t2i1">
<img id="myt2i1" src="./static/images/Diversity_turbo2.png"
class="interpolation-image"/>
</div>
<div class="item item-t2i1">
<img id="myt2i1" src="./static/images/Diversity_lightning2.png"
class="interpolation-image"/>
</div>
</div>
</div>
</div>
<br>
<!--/ Matting. -->
<div class="container is-max-desktop">
<!-- Latent space editing applications -->
<div class="columns is-centered">
<div class="column is-full-width">
<h2 class="title is-3 has-text-centered">Related Works</h2>
<div class="content has-text-justified">
<p>
<li>Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Bjorn Ommer. <a href="https://arxiv.org/abs/2112.10752">High-resolution image synthesis with latent diffusion models.</a> In IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2022.</li><br>
<li>Dustin Podell, Zion English, Kyle Lacey, Andreas Blattmann, Tim Dockhorn, Jonas Müller, Joe Penna, and Robin Rombach. <a href="https://arxiv.org/abs/2307.01952">SDXL: Improving Latent Diffusion Models for High-Resolution Image Synthesis.</a> In International Conference on Learning Representation (ICLR), 2024.</li><br>
<li>Markus Kettunen, Erik Härkönen, and Jaakko Lehtinen. <a href="https://arxiv.org/abs/1906.03973">E-LPIPS: Robust Perceptual Image Similarity via Random Transformation Ensembles.</a> In arXiv preprint arXiv:1906.03973, 2019.</li><br>
<li>Tianwei Yin, Michael Gharbi, Richard Zhang, Eli Shechtman, Fredo Durand, William T. Freeman, and Taesung Park. <a href="https://arxiv.org/abs/2311.18828">One-step Diffusion with Distribution Matching Distillation.</a> In IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2024.</li><br>
<li>Axel Sauer, Dominik Lorenz, Andreas Blattmann, and Robin Rombach. <a href="https://arxiv.org/abs/2311.17042">Adversarial Diffusion Distillation.</a> In arXiv preprint arXiv:2311.17042, 2023.</li><br>
<li>Shanchuan Lin, Anran Wang, and Xiao Yang. <a href="https://arxiv.org/abs/2402.13929">SDXL-Lightning: Progressive Adversarial Diffusion Distillation.</a> In arXiv preprint arXiv:2402.13929, 2024.</li><br>
</p>
</div>
<!-- Prompt Interpolation image -->
</div>
</div>
<!-- Concurrent Work. -->
<div class="columns is-centered">
<div class="column is-full-width">
<h2 class="title is-3 has-text-centered">Acknowledgements</h2>
<div class="content has-text-justified">
<p>
We would like to thank Tianwei Yin, Seungwook Kim, Sungyeon Kim for their valuable feedback and comments. Part of this work was done while Minguk Kang was an intern at Adobe Research. Minguk Kang and Suha Kwak were supported by the NRF grant and IITP grant funded by Ministry of Science and ICT, Korea (NRF-2021R1A2C3012728, AI Graduate School (POSTECH): IITP-2019-0-01906). Jaesik Park was supported by the IITP grant funded by the government of South Korea (MSIT) (AI Graduate School (SNU): 2021-0-01343 and AI Innovation Hub: 2021-0-02068). Jun-Yan Zhu was supported by the Packard Fellowship.
</p>
</div>
</div>
</div>
<!--/ Concurrent Work. -->
</div>
</section>
<section class="section" id="BibTeX">
<div class="container is-max-desktop content">
<h2 class="title"><a id="bibtex">BibTeX</a></h2>
<pre><code>@inproceedings{kang2024diffusion2gan,
author = {Kang, Minguk and Zhang, Richard and Barnes, Connelly and Paris, Sylvain and Kwak, Suha and Park, Jaesik and Shechtman, Eli and Zhu, Jun-Yan and Park, Taesung},
title = {{Distilling Diffusion Models into Conditional GANs}},
booktitle = {European Conference on Computer Vision (ECCV)},
year = {2024},
}</code></pre>
</div>
</section>
<footer class="footer">
<div class="columns is-centered">
<div class="column is-8">
<div class="content">
<p>
This website is licensed under a <a rel="license"
href="http://creativecommons.org/licenses/by-sa/4.0/">Creative
Commons Attribution-ShareAlike 4.0 International License</a>.
</p>
<p>
Website adapted from the following <a href="https://github.com/nerfies/nerfies.github.io">source code</a>.
</p>
</div>
</div>
</div>
</div>
</footer>
<script src="juxtapose/js/juxtapose.js"></script>
<script>
var slider;
let origImages = [
{"src": "./static/images/corgi_input.jpeg", "label": "Generated by SDXL-Diffusion2GAN (512px)",},
{"src": "./static/images/corgi_output.jpeg", "label": "8 x Upsampled by GigaGAN (4K)",}
];
let origOptions = {
"makeResponsive": true,
"showLabels": true,
"mode": "horizontal",
"showCredits": true,
"animate": true,
"startingPosition": "50"
};
const juxtaposeSelector = "#juxtapose-embed";
const transientSelector = "#juxtapose-hidden";
function tab_gallery_click(name) {
// Get the expanded image
let inputImage = {
label: "Generated by Diffusion2GAN (512px)",
};
let outputImage = {
label: "8 x Upsampled by GigaGAN (4K)",
};
inputImage.src = "./static/images/".concat(name, "_input.jpeg")
outputImage.src = "./static/images/".concat(name, "_output.jpeg")
let images = [inputImage, outputImage];
let options = slider.options;
options.callback = function(obj) {
var newNode = document.getElementById(obj.selector.substring(1));
var oldNode = document.getElementById(juxtaposeSelector.substring(1));
console.log(obj.selector.substring(1));
console.log(newNode.children[0]);
oldNode.replaceChild(newNode.children[0], oldNode.children[0]);
//newNode.removeChild(newNode.children[0]);
};
slider = new juxtapose.JXSlider(transientSelector, images, options);
};
(function() {
slider = new juxtapose.JXSlider(
juxtaposeSelector, origImages, origOptions);
//document.getElementById("left-button").onclick = replaceLeft;
//document.getElementById("right-button").onclick = replaceRight;
})();
// Get the image text
var imgText = document.getElementById("imgtext");
// Use the same src in the expanded image as the image being clicked on from the grid
// expandImg.src = imgs.src;
// Use the value of the alt attribute of the clickable image as text inside the expanded image
imgText.innerHTML = name;
// Show the container element (hidden with CSS)
// expandImg.parentElement.style.display = "block";
$(".flip-card").click(function() {
console.log("fading in")
div_back = $(this).children().children()[1]
div_front = $(this).children().children()[0]
// console.log($(this).children("div.flip-card-back"))
console.log(div_back)
$(div_front).addClass("out");
$(div_front).removeClass("in");
$(div_back).addClass("in");
$(div_back).removeClass("out");
});
$(".flip-card").mouseleave(function() {
console.log("fading in")
div_back = $(this).children().children()[1]
div_front = $(this).children().children()[0]
// console.log($(this).children("div.flip-card-back"))
console.log(div_back)
$(div_front).addClass("in");
$(div_front).removeClass("out");
$(div_back).addClass("out");
$(div_back).removeClass("in");
});
</script>
<!-- <script src="https://ajax.googleapis.com/ajax/libs/jquery/1.7.1/jquery.min.js" type="text/javascript"></script> -->
<script src="https://cdn.jsdelivr.net/npm/[email protected]/dist/umd/popper.min.js" integrity="sha384-ApNbgh9B+Y1QKtv3Rn7W3mgPxhU9K/ScQsAP7hUibX39j7fakFPskvXusvfa0b4Q" crossorigin="anonymous"></script>
<script src="https://cdn.jsdelivr.net/npm/[email protected]/dist/js/bootstrap.min.js"></script>
</body>
</html>