-
Notifications
You must be signed in to change notification settings - Fork 1
/
index.html
652 lines (585 loc) · 25.7 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
<!DOCTYPE html>
<html>
<head>
<meta charset="utf-8">
<meta name="description"
content="DiffusionFeatures">
<meta name="keywords" content="DiffusionFeatures">
<meta name="viewport" content="width=device-width, initial-scale=1">
<title>DiffusionFeatures</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/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">
<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>
</head>
<body>
<section class="hero">
<div class="hero-body publication-header">
<div class="container is-max-desktop">
<div class="columns is-centered">
<div class="column has-text-centered">
<h1 class="title is-1">Emergent Correspondence <br> from Image Diffusion</h1>
<div class="is-size-5 publication-authors">
<span class="author-block">
<a href="https://lumingtang.info/">Luming Tang</a><sup>*</sup>, </span>
<span class="author-block">
<a href="https://kmnp.github.io/">Menglin Jia</a><sup>*</sup>, </span>
<span class="author-block">
<a href="https://www.cs.cornell.edu/~qqw/">Qianqian Wang</a><sup>*</sup>, </span>
<span class="author-block">
<a href="https://www.cs.cornell.edu/~cpphoo/">Cheng Perng Phoo</a>,
</span>
<span class="author-block">
<a href="https://www.cs.cornell.edu/~bharathh/">Bharath Hariharan</a>
</span>
</div>
<div class="is-size-5 publication-authors">
<span class="author-block">Cornell University</span>
</div>
<span class="is-size-7">(*Equal contribution)</span>
<br>
<br>
<!-- <div class="is-size-10 publication-authors">
(*Equal contribution)
</div> -->
<h1 style="font-size:24px;font-weight:bold">NeurIPS 2023</h1>
<div class="column has-text-centered">
<div class="publication-links">
<!-- PDF Link. -->
<span class="link-block">
<a href="https://arxiv.org/abs/2306.03881"
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/Tsingularity/dift"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="fab fa-github"></i>
</span>
<span>Code</span>
</a>
</span>
<span class="link-block">
<a href="https://colab.research.google.com/drive/1km6MGafhAvbPOouD3oo64aUXgLlWM6L1?usp=sharing"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<img src = "static/images/colab-logo.svg" alt="colab-logo"/>
</span>
<span>Colab</span>
</a>
</span>
</div>
</div>
</div>
</div>
</div>
</div>
</section>
<div align="center" style="margin-top:0px; margin-bottom:0px;">
<div class="carousel-teaser-padding">
<div class="hero-body">
<div class="container">
<div id="results-carousel" class="carousel results-carousel">
<div class="item item-toby">
<video poster="" autoplay playsinline muted loop>
<source type="video/mp4" src="static/videos/teaser_video_0_compressed.mp4" />
</video>
</div>
<div class="item item-toby">
<video poster="" autoplay playsinline muted loop>
<source type="video/mp4" src="static/videos/teaser_video_4_compressed.mp4" />
</video>
</div>
<div class="item item-toby">
<video poster="" autoplay playsinline muted loop>
<source type="video/mp4" src="static/videos/teaser_video_2_compressed.mp4" />
</video>
</div>
<div class="item item-toby">
<video poster="" autoplay playsinline muted loop>
<source type="video/mp4" src="static/videos/teaser_video_3_compressed.mp4" />
</video>
</div>
<div class="item item-toby">
<video poster="" autoplay playsinline muted loop>
<source type="video/mp4" src="static/videos/teaser_video_1_compressed.mp4" />
</video>
</div>
</div>
</div>
<h3 class="subtitle has-text-centered">
<font size="4">
<span class="dnerf"></span>
Without any supervision, Diffusion Features can find correspondences on real images across instances, categories, and even domains.
</font>
</h3>
</div>
</div>
</div>
<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>
Finding correspondences between images is a fundamental problem in computer vision.
In this paper, we show that correspondence emerges in image diffusion models
<i>without any explicit supervision</i>. We propose a simple strategy to extract this implicit
knowledge out of diffusion networks as image features,
namely <strong>DI</strong>ffusion <strong>F</strong>ea<strong>T</strong>ures (<strong>DIFT</strong>),
and use them to establish correspondences between real images. Without any additional fine-tuning or
supervision on the task-specific data or annotations, DIFT is able to outperform both
weakly-supervised methods and competitive off-the-shelf features in identifying semantic,
geometric, and temporal correspondences. Particularly for semantic correspondence,
DIFT from Stable Diffusion is able to outperform DINO and OpenCLIP by 19 and 14 accuracy points
respectively on the challenging SPair-71k benchmark. It even outperforms the state-of-the-art
supervised methods on 9 out of 18 categories while remaining on par for the overall performance.
</p>
</div>
</div>
</div>
<!--/ Abstract. -->
</div>
</section>
<br>
<br>
<section class="section grey">
<div class="columns is-centered has-text-centered">
<h2 class="title is-3">Semantic Correspondence</h2>
</div>
<div align="center" style="margin-top:0px;" style="margin-bottom:0px;"></div>
<div class="columns is-centered has-text-centered">
<div class="column is-four-fifths">
<div class="content has-text-justified">
<p>
Without any fine-tuning or correspondence supervision, DIFT is able to establish reasonable and accurate semantic correspondence, outperform previous
weakly-supervised methods with a large margin, and even on par with the state-of-the-art supervised methods.
</p>
</div>
</div>
</div>
<br>
<br>
<div class="columns is-centered has-text-centered">
<h3 class="title is-3 margin-bottom-8">Comparison on SPair-71k</h3>
</div>
<div class="columns is-centered has-text-centered">
<div class="column is-four-fifths">
<div class="content has-text-justified">
<p>
Compared to popular off-the-shelf features that (pre-)trained on similar data,
DIFT identifies better correspondences under occlusion, clustered scenes, viewpoint change, pose variants,
and instance-level appearance change,
outperforming its self-supervised learning counterpart feature (DIFT_sd vs. OpenCLIP; DIFT_adm vs. DINO) over 14 PCK points.
<br>
Below we visualize correspondence predictions using different features.
The leftmost image is the source image with a set of keypoints;
the rightmost image contains the ground truth correspondence for a target image
whereas any images in between contain keypoints found using feature matching
with various features. We use different colors to indicate different keypoints.
Circles indicate correctly-predicted keypoints and crosses for incorrect matches.
</p>
</div>
</div>
</div>
<div class="carousel-spair-padding">
<div class="hero-body carousel-body-vert-padding">
<div class="container">
<div id="results-carousel" class="carousel results-carousel">
<div class="item item-toby">
<img style='height: auto; width: 100%; object-fit: contain' src="static/images/web_spair_0.jpg" alt="edit_propagation">
</div>
<div class="item item-toby">
<img style='height: auto; width: 100%; object-fit: contain' src="static/images/web_spair_1.jpg" alt="edit_propagation">
</div>
<div class="item item-toby">
<img style='height: auto; width: 100%; object-fit: contain' src="static/images/web_spair_2.jpg" alt="edit_propagation">
</div>
<div class="item item-toby">
<img style='height: auto; width: 100%; object-fit: contain' src="static/images/web_spair_3.jpg" alt="edit_propagation">
</div>
<div class="item item-toby">
<img style='height: auto; width: 100%; object-fit: contain' src="static/images/web_spair_4.jpg" alt="edit_propagation">
</div>
</div>
</div>
</div>
</div>
<br>
<div class="columns is-centered has-text-centered">
<h3 class="title is-3">Image Editing with DIFT</h3>
</div>
<div class="columns is-centered has-text-centered">
<div class="column is-four-fifths">
<div class="content has-text-justified">
<p> DIFT can easily propagate edits from one image to others across different instances, categories, and domains, without any correspondence supervision. </p>
</div>
</div>
</div>
<!-- <div class="carousel-extra-padding"> -->
<div class="carousel-spair-padding">
<div class="hero-body carousel-body-vert-padding">
<div class="container">
<div id="results-carousel" class="carousel results-carousel">
<div class="item item-toby">
<video poster="" autoplay playsinline controls muted loop>
<source type="video/mp4" src="static/videos/image_edit_dog_compressed.mp4" />
</video>
</div>
<div class="item item-toby">
<video poster="" autoplay playsinline controls muted loop>
<source type="video/mp4" src="static/videos/image_edit_cat_compressed.mp4" />
</video>
</div>
<div class="item item-toby">
<video poster="" autoplay playsinline controls muted loop>
<source type="video/mp4" src="static/videos/image_edit_bird_compressed.mp4" />
</video>
</div>
</div>
</div>
</div>
</div>
</section>
<br>
<br>
<section class="section hero is-small">
<div class="columns is-centered has-text-centered">
<h2 class="title is-3">Geometric Correspondence</h2>
</div>
<div class="columns is-centered has-text-centered">
<div class="column is-four-fifths">
<div class="content has-text-justified">
<p> By using a small time step t, DIFT also shows competitive performance on geometric correspondence without any such supervision.
Below we show the sparse matching results on HPatches using DIFT after removing outliers. We can see it works well under challenging viewpoint and
illumination changes.</p>
</div>
</div>
</div>
<!-- <div class="carousel-homography-padding"> -->
<div class="carousel-spair-padding">
<div class="hero-body carousel-body-vert-padding">
<div class="container">
<div id="results-carousel" class="carousel results-carousel">
<div class="item item-toby">
<img style='height: auto; width: 100%; object-fit: contain' src="static/images/homography.jpg" alt="edit_propagation">
</div>
<div class="item item-toby">
<img style='height: auto; width: 100%; object-fit: contain' src="static/images/homography1.jpg" alt="edit_propagation">
</div>
<div class="item item-toby">
<img style='height: auto; width: 100%; object-fit: contain' src="static/images/homography2.jpg" alt="edit_propagation">
</div>
</div>
</div>
</div>
</div>
</section>
<br>
<br>
<section class="hero is-small section grey">
<div class="columns is-centered has-text-centered">
<h2 class="title is-3">Temporal Correspondence</h2>
</div>
<div class="columns is-centered has-text-centered">
<div class="column is-four-fifths">
<div class="content has-text-justified">
<p> DIFT also demonstrates strong performance on temporal
correspondence tasks, although never trained or fine-tuned on video data.
</p>
<p>
Here are the video segementation propagation results on DAVIS using DIFT.
</p>
</div>
</div>
</div>
<!-- <div class="carousel-extra-padding"> -->
<div class="carousel-spair-padding">
<div class="hero-body carousel-body-vert-padding">
<div class="container">
<div id="results-carousel" class="carousel results-carousel">
<div class="item item-toby">
<video poster="" autoplay playsinline muted loop>
<source type="video/mp4" src="static/videos/davis/blackswan_adm_compressed.mp4" />
</video>
</div>
<div class="item item-toby">
<video poster="" autoplay playsinline muted loop>
<source type="video/mp4" src="static/videos/davis/car-shadow_adm_compressed.mp4" />
</video>
</div>
<div class="item item-toby">
<video poster="" autoplay playsinline muted loop>
<source type="video/mp4" src="static/videos/davis/cows_adm_compressed.mp4" />
</video>
</div>
<div class="item item-toby">
<video poster="" autoplay playsinline muted loop>
<source type="video/mp4" src="static/videos/davis/dog_adm_compressed.mp4" />
</video>
</div>
<div class="item item-toby">
<video poster="" autoplay playsinline muted loop>
<source type="video/mp4" src="static/videos/davis/drift-chicane_adm_compressed.mp4" />
</video>
</div>
<div class="item item-toby">
<video poster="" autoplay playsinline muted loop>
<source type="video/mp4" src="static/videos/davis/drift-straight_adm_compressed.mp4" />
</video>
</div>
<div class="item item-toby">
<video poster="" autoplay playsinline muted loop>
<source type="video/mp4" src="static/videos/davis/goat_adm_compressed.mp4" />
</video>
</div>
<div class="item item-toby">
<video poster="" autoplay playsinline muted loop>
<source type="video/mp4" src="static/videos/davis/dance-twirl_adm_compressed.mp4" />
</video>
</div>
<div class="item item-toby">
<video poster="" autoplay playsinline muted loop>
<source type="video/mp4" src="static/videos/davis/gold-fish_adm_compressed.mp4" />
</video>
</div>
<div class="item item-toby">
<video poster="" autoplay playsinline muted loop>
<source type="video/mp4" src="static/videos/davis/horsejump-high_adm_compressed.mp4" />
</video>
</div>
<div class="item item-toby">
<video poster="" autoplay playsinline muted loop>
<source type="video/mp4" src="static/videos/davis/india_adm_compressed.mp4" />
</video>
</div>
<div class="item item-toby">
<video poster="" autoplay playsinline muted loop>
<source type="video/mp4" src="static/videos/davis/libby_adm_compressed.mp4" />
</video>
</div>
<div class="item item-toby">
<video poster="" autoplay playsinline muted loop>
<source type="video/mp4" src="static/videos/davis/parkour_adm_compressed.mp4" />
</video>
</div>
<div class="item item-toby">
<video poster="" autoplay playsinline muted loop>
<source type="video/mp4" src="static/videos/davis/pigs_adm_compressed.mp4" />
</video>
</div>
</div>
</div>
</div>
</div>
<div class="columns is-centered has-text-centered">
<div class="column is-four-fifths">
<div class="content has-text-justified">
<p>
Here are the human pose keypoints tracking results on JHMDB using DIFT.
</p>
</div>
</div>
</div>
<div class="carousel-jhmdb-padding">
<div class="hero-body carousel-body-vert-padding">
<div class="container">
<div id="results-carousel" class="scroll_carousel results-carousel">
<div class="item item-toby">
<video poster="" autoplay playsinline muted loop>
<source type="video/mp4" src="static/videos/jhmdb/0.mp4" />
</video>
</div>
<div class="item item-toby">
<video poster="" autoplay playsinline muted loop>
<source type="video/mp4" src="static/videos/jhmdb/1.mp4" />
</video>
</div>
<div class="item item-toby">
<video poster="" autoplay playsinline muted loop>
<source type="video/mp4" src="static/videos/jhmdb/2.mp4" />
</video>
</div>
<div class="item item-toby">
<video poster="" autoplay playsinline muted loop>
<source type="video/mp4" src="static/videos/jhmdb/3.mp4" />
</video>
</div>
<div class="item item-toby">
<video poster="" autoplay playsinline muted loop>
<source type="video/mp4" src="static/videos/jhmdb/4.mp4" />
</video>
</div>
<div class="item item-toby">
<video poster="" autoplay playsinline muted loop>
<source type="video/mp4" src="static/videos/jhmdb/5.mp4" />
</video>
</div>
<div class="item item-toby">
<video poster="" autoplay playsinline muted loop>
<source type="video/mp4" src="static/videos/jhmdb/6.mp4" />
</video>
</div>
<div class="item item-toby">
<video poster="" autoplay playsinline muted loop>
<source type="video/mp4" src="static/videos/jhmdb/7.mp4" />
</video>
</div>
<div class="item item-toby">
<video poster="" autoplay playsinline muted loop>
<source type="video/mp4" src="static/videos/jhmdb/8.mp4" />
</video>
</div>
<div class="item item-toby">
<video poster="" autoplay playsinline muted loop>
<source type="video/mp4" src="static/videos/jhmdb/9.mp4" />
</video>
</div>
<div class="item item-toby">
<video poster="" autoplay playsinline muted loop>
<source type="video/mp4" src="static/videos/jhmdb/10.mp4" />
</video>
</div>
<div class="item item-toby">
<video poster="" autoplay playsinline muted loop>
<source type="video/mp4" src="static/videos/jhmdb/11.mp4" />
</video>
</div>
<div class="item item-toby">
<video poster="" autoplay playsinline muted loop>
<source type="video/mp4" src="static/videos/jhmdb/12.mp4" />
</video>
</div>
<div class="item item-toby">
<video poster="" autoplay playsinline muted loop>
<source type="video/mp4" src="static/videos/jhmdb/13.mp4" />
</video>
</div>
<div class="item item-toby">
<video poster="" autoplay playsinline muted loop>
<source type="video/mp4" src="static/videos/jhmdb/14.mp4" />
</video>
</div>
</div>
</div>
</div>
</div>
</section>
<br>
<section class="section">
<div class="columns is-centered">
<div class="column is-four-fifths">
<h3 class="title is-3">Interactive Demo: Give it a Try!</h3>
<div class="content has-text-justified">
We make an interative demo to demonstrate the semantic correspondence established by DIFT and you could try on your
own images. You could use either the provided <a href="https://colab.research.google.com/drive/1km6MGafhAvbPOouD3oo64aUXgLlWM6L1?usp=sharing">Colab</a>
or the <a href="https://github.com/Tsingularity/dift/blob/main/demo.ipynb">jupyter notebook</a> in our github repository.
After loading two images, you could left-click on an interesting point of the source image on the left,
then after 1 or 2 seconds, the corresponding point on the target image will be displayed as a red point on the right,
together with a heatmap showing the per-pixel cosine distance calculated using DIFT. Here're two examples on cat and guitar:
</div>
</div>
</div>
<br>
<div class="columns is-centered">
<div class="demo-section-videos-desktop demo-section-videos">
<div>
<video id="demo_cat" class="demo-section-video" autoplay muted loop playsinline>
<source src="static/videos/demo_cat.mp4" type="video/mp4">
</video>
</div>
<div>
<video id="demo_guitar" class="demo-section-video" autoplay muted loop playsinline>
<source src="static/videos/demo_guitar.mp4" type="video/mp4">
</video>
</div>
</div>
</div>
</section>
<section class="section">
<div class="columns is-centered">
<div class="column is-four-fifths">
<h3 class="title is-3">Concurrent Works</h3>
<div class="content has-text-justified">
<ul>
<li>
<a href="https://diffusion-hyperfeatures.github.io/">Diffusion Hyperfeatures</a> extracts
feature maps varying across timesteps and layers from the diffusion process
and trains a lightweight neural network to aggregate them together for semantic correspondence.
</li>
<li>
<a href="https://sd-complements-dino.github.io/">A Tale of Two Features</a> fuses the Stable Diffusion
features with DINOv2 features to build dense semantic correspondence.
</li>
<li>
<a href="https://arxiv.org/abs/2305.15581">Unsupervised Semantic Correspondence Using Stable Diffusion</a>
first optimizes the prompt embedding to maximize attention on the region of interest,
then uses it for semantic correspondence.
</li>
</ul>
</div>
</div>
</div>
</section>
<section class="section">
<div class="columns is-centered">
<div class="column is-four-fifths">
<h3 class="title is-3">Acknowledgements</h3>
<div class="content has-text-justified">
<p>
This work was partially funded by NSF 2144117 and the DARPA Learning with Less Labels program (HR001118S0044).
We would like to thank Zeya Peng for her help on the edit propagation section and the project page,
thank Kamal Gupta for sharing the evaluation details in the ASIC paper,
and thank Aaron Gokaslan, Utkarsh Mall, Jonathan Moon, Boyang Deng for valuable discussion and feedback.
</p>
</div>
</div>
</div>
</section>
<section class="section" id="BibTeX">
<div class="columns is-centered">
<div class="column is-four-fifths">
<h3 class="title is-3">BibTeX</h3>
</div>
</div>
<div class="container is-max-desktop content">
<pre><code>
@inproceedings{
tang2023emergent,
title={Emergent Correspondence from Image Diffusion},
author={Luming Tang and Menglin Jia and Qianqian Wang and Cheng Perng Phoo and Bharath Hariharan},
booktitle={Thirty-seventh Conference on Neural Information Processing Systems},
year={2023},
url={https://openreview.net/forum?id=ypOiXjdfnU}
}</code></pre>
</div>
</section>
<footer class="footer">
<div align="center" class="container">
<div class="columns is-centered">
<div class="content">
This website is modified from <a href="https://3d-moments.github.io">3D Moments</a>, <a href="https://infinite-nature-zero.github.io">InfiniteNature-Zero</a>, and <a href="https://github.com/nerfies/nerfies.github.io">Nerfies</a>.
</div>
</div>
</div>
</footer>
</body>
</html>