-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathindex.html
580 lines (509 loc) · 23.8 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
<!DOCTYPE html>
<html>
<head>
<meta charset="utf-8">
<meta name="description"
content="Evaluation Agent: Efficient and Promptable Evaluation Framework for Visual Generative Models">
<meta name="keywords" content="Evaluation, Benchmark, Prompts, Dataset, Video Generation, Text-to-Video, Stable Diffusion, AIGC, Image Generation, MMLab, Shanghai AI Lab">
<meta name="viewport" content="width=device-width, initial-scale=1">
<title>Evaluation Agent: Efficient and Promptable Evaluation Framework for Visual Generative Models</title>
<!-- Global site tag (gtag.js) - Google Analytics-->
<script async src="https://www.googletagmanager.com/gtag/js?id=G-Y5ZVQZ7NHC"></script>
<script>
window.dataLayer = window.dataLayer || [];
function gtag() {
dataLayer.push(arguments);
}
gtag('js', new Date());
gtag('config', 'G-Y5ZVQZ7NHC');
</script>
<link href="https://fonts.googleapis.com/css?family=Google+Sans|Noto+Sans|Castoro"
rel="stylesheet">
<link rel="stylesheet" href="https://cdn.datatables.net/1.13.6/css/jquery.dataTables.min.css">
<link rel="stylesheet" href="./assets/css/bulma.min.css">
<link rel="stylesheet" href="./assets/css/bulma-carousel.min.css">
<link rel="stylesheet" href="./assets/css/bulma-slider.min.css">
<link rel="stylesheet" href="./assets/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="./assets/css/index.css">
<script src="https://ajax.googleapis.com/ajax/libs/jquery/3.5.1/jquery.min.js"></script>
<script defer src="./assets/js/fontawesome.all.min.js"></script>
<script src="./assets/js/bulma-carousel.min.js"></script>
<script src="./assets/js/bulma-slider.min.js"></script>
<script src="./assets/js/index.js"></script>
</head>
<body>
<!-- title -->
<section class="hero">
<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-1 publication-title"><span style="color:#71a10a; font-weight: bold; font-style: italic">Evaluation Agent</span><br>
<span style="font-size: 0.91em;">Efficient and Promptable Evaluation Framework for Visual Generative Models</span>
</h1>
<!-- <h1 class="title is-1 publication-title"><span style="color:#71a10a; font-weight: bold; font-style: italic">Evaluation Agent</span> : Efficient and Promptable<br>Evaluation Framework<br>for Visual Generative Models
</h1> -->
<div class="is-size-5 publication-authors">
<span class="author-block">
<a href="https://github.com/zhangfan-p" target="_blank">Fan Zhang</a><sup>1*</sup>,
</span>
<span class="author-block">
<a href="https://shulin16.github.io/" target="_blank">Shulin Tian</a><sup>2*</sup>,
</span>
<span class="author-block">
<a href="https://ziqihuangg.github.io" target="_blank">Ziqi Huang</a><sup>2*</sup>,
</span>
<span class="author-block">
<a href="http://mmlab.siat.ac.cn/yuqiao/index.html" target="_blank">Yu Qiao</a><sup>1†</sup>,
</span>
<span class="author-block">
<a href="https://liuziwei7.github.io/" target="_blank">Ziwei Liu</a><sup>2†</sup>
</span>
</div>
<!-- <br> -->
<div class="is-size-5 publication-authors">
<span class="author-block">(* equal contributions, † corresponding authors) </span>
</div>
<div class="is-size-5 publication-authors">
<span class="author-block">
<sup>1</sup>
Shanghai Artificial Intelligence Laboratory
<sup>2</sup>
S-Lab, Nanyang Technological University
</span>
</div>
<div class="column has-text-centered">
<div class="publication-links">
<!-- Paper Link. -->
<span class="link-block">
<a href="https://arxiv.org/abs/2412.09645" target="_blank"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="ai ai-arxiv"></i>
</span>
<span>Paper</span>
</a>
</span>
<!-- Video Link. -->
<span class="link-block">
<a href="https://youtu.be/RyAS23dsgP8" target="_blank"
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>
<!-- Code Link. -->
<span class="link-block">
<a href="https://github.com/Vchitect/Evaluation-Agent" target="_blank"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="fab fa-github"></i>
</span>
<span>GitHub</span>
</a>
</span>
<!-- <span class="link-block">
<a href="https://github.com/Vchitect/Evaluation-Agent" class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="far fa-images"></i>
</span>
<span>Dataset</span>
</a> -->
</div>
</div>
</div>
</div>
</div>
</div>
</section>
<!-- teaser -->
<section class="hero teaser">
<div class="container is-max-desktop"style="text-align: center;">
<div class="hero-body">
<img src="./assets/images/fig_teaser.png" style="width:600px; margin-bottom:10px" alt="Teaser."/>
<p style="margin-top: 0;">
<b> An Example of Evaluation Agent.</b> Existing evaluation methods typically assess visual generative models by extensively sampling from a fixed benchmark. In contrast, our Evaluation Agent framework requires only a small number of sampled images or videos, tailored to the user's specific evaluation request. Additionally, it goes beyond providing a simple numerical score by offering detailed explanations to the evaluation conclusions.
</p>
</div>
</div>
</section>
<!-- Paper video. -->
<section class="hero is-light">
<div class="columns is-centered has-text-centered" style="margin-top: 0px; margin-bottom: 10px;">
<div class="column is-three-fifths">
<h2 class="title is-3">Video</h2>
<div class="publication-video">
<iframe width="560" height="315" src="https://www.youtube.com/embed/RyAS23dsgP8" title="YouTube video player" frameborder="0" allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share" allowfullscreen></iframe>
</div>
</div>
</div>
</section>
<!-- / Paper video. -->
<!-- Abstract. -->
<section>
<div class="container is-max-desktop">
<div class="columns is-centered has-text-centered" style="margin-top: 10px; margin-bottom: 0px;">
<div class="column is-four-fifths">
<h2 class="title is-3 is-centered">Abstract</h2>
<div class="content has-text-justified">
<p style="margin-top: 0; margin-bottom:-20px;">
Recent advancements in visual generative models have enabled high-quality image and video generation, opening diverse applications. However, evaluating these models often demands sampling hundreds or thousands of images or videos, making the process computationally expensive, especially for diffusion-based models with inherently slow sampling. Moreover, existing evaluation methods rely on rigid pipelines that overlook specific user needs and provide numerical results without clear explanations. In contrast, humans can quickly form impressions of a model's capabilities by observing only a few samples. To mimic this, we propose the <b>Evaluation Agent</b> framework, which employs human-like strategies for efficient, dynamic, multi-round evaluations using only a few samples per round, while offering detailed, user-tailored analyses. It offers four key advantages: 1) <b>efficiency</b>, 2) <b>promptable evaluation</b> tailored to diverse user needs, 3) <b>explainability</b> beyond single numerical scores, and 4) <b>scalability</b> across various models and tools. Experiments show that Evaluation Agent reduces evaluation time to 10% of traditional methods while delivering comparable results. The Evaluation Agent framework is fully open-sourced to advance research in visual generative models and their efficient evaluation.
</p>
</div>
</div>
</div>
</section>
<!--/ Abstract. -->
<!-- style="width: 50%; max-width: none;" -->
<!-- Evaluation Agent Framework. -->
<section class="section" style="margin-top:0px; margin-bottom:50px;">
<div class="container is-max-desktop">
<div class="hero-body">
<div class="section-title">
<h2 class="title is-3 is-centered">Pipeline of Evaluation Agent Framework</h2>
</div>
<div class="columns is-centered has-text-centered">
<div class="column" style="text-align: center;">
<div class="publication-img">
<img id="architecture" src="./assets/images/fig_framework.jpg" style="width:100%; max-width:100%; margin-top:30px;margin-bottom:10px;"/>
</div>
</div>
</div>
<p style="margin-top: 0; margin-bottom:-100px;">
<b>Overview of Evaluation Agent Framework.</b> This framework leverages LLM-powered agents for efficient and flexible visual model assessments. As shown, it consists of two stages: (a) the Proposal Stage, where user queries are decomposed into sub-aspects, and prompts are generated, and (b) the Execution Stage, where visual content is generated and evaluated using an Evaluation Toolkit. The two stages interact iteratively to dynamically assess models based on user queries.
</p>
</div>
</div>
</section>
<!-- /Evaluation Agent Framework. -->
<style>
.table-container {
overflow-x: auto;
text-align: center;
}
table {
width: 100%;
max-width: 100%;
border-collapse: collapse;
margin: 0 auto;
}
th, td {
padding: 8px;
text-align: center;
border: 1px solid #ddd;
}
th {
background-color: #f2f2f2;
font-weight: bold;
border-bottom: 2px solid #ddd;
}
td {
border: 1px solid #ddd;
}
.nowrap-table th, td {
white-space: nowrap;
}
.nowrap-table-td td {
white-space: nowrap;
}
</style>
<!-- style="width: 43%; max-width: none;" -->
<!-- Comparison between the Evaluation Agent Framework and other T2I/T2V Benchmarks. -->
<section class="section" style="margin-top:-50px; margin-bottom:-80px;">
<div class="container is-max-desktop">
<div class="hero-body">
<div class="section-title">
<h2 class="title is-3 is-centered">Comparing with Other T2I/T2V Benchmarks</h2>
</div>
</div>
<div class="table-container" style="text-align: center; margin-top:-30px;">
<table style="width:100%; text-align:center; border-collapse:collapse; margin: 0 auto;">
<thead>
<tr>
<th>Benchmark</th>
<th>Analysis</th>
<th>Customized Queries</th>
<th>Supported Models</th>
<th># Required Samples</th>
<th>Open Evaluation Request Support</th>
<th>Dynamic Evaluation</th>
<th>Open Tool-Use</th>
</tr>
</thead>
<tbody>
<tr>
<td>FID / FVD</td>
<td>❌</td>
<td>❌</td>
<td>T2I / T2V</td>
<td>2,048</td>
<td>❌ (Fixed-Form)</td>
<td>❌</td>
<td>❌</td>
</tr>
<tr>
<td>T2I-CompBench</td>
<td>❌</td>
<td>❌</td>
<td>T2I</td>
<td>18,000</td>
<td>❌ (Pre-Defined)</td>
<td>❌</td>
<td>❌</td>
</tr>
<tr>
<td>VBench</td>
<td>❌</td>
<td>❌</td>
<td>T2V</td>
<td>4,730</td>
<td>❌ (Pre-Defined)</td>
<td>❌</td>
<td>❌</td>
</tr>
<tr>
<td><b>Evaluation Agent (Ours)</b></td>
<td style="color: green;">✔</td>
<td style="color: green;">✔</td>
<td>T2I & T2V</td>
<td>400</td>
<td style="color: green;">✔ (Open-Ended)</td>
<td style="color: green;">✔</td>
<td style="color: green;">✔</td>
</tr>
</tbody>
<!-- <caption>Comparison between the Evaluation Agent Framework and other T2I/T2V Benchmarks.</caption> -->
</table>
</div>
<p style="margin-top: 0;">
<b>Comparison of the Evaluation Agent Framework with Traditional T2I and T2V Benchmarks.</b> The Evaluation Agent framework supports customized user queries in natural language and works with both T2I and T2V models. Unlike traditional benchmarks, it dynamically updates the evaluation process using multiple tools, providing comprehensive and explainable results with detailed textual analysis.
</p>
</div>
</section>
<!-- /Comparison between the Evaluation Agent Framework and other T2I/T2V Benchmarks -->
<!-- Experimental Results of Evaluation Agent.-->
<section class="section" style="margin-top:50px; margin-bottom:-100px;">
<div class="container is-max-desktop">
<div class="hero-body">
<div class="section-title">
<h2 class="title is-3 is-centered">Experimental Results of the Evaluation Agent</h2>
</div>
</div>
<div class="section-title" style="margin-top:-20px;">
<h4 class="title is-5 is-centered">Compare Time Consumption on the VBench</h5>
</div>
<div class="table-container" style="text-align: center;">
<table class="nowrap-table" style="width:50%; text-align:center; border-collapse:collapse; margin: 0 auto;">
<thead>
<tr>
<th>Models</th>
<th>VBench (Total Cost) ↓</th>
<th>VBench (Avg. Cost per Dimension) ↓</th>
<th>Evaluation Agent (Ours) ↓</th>
</tr>
</thead>
<tbody>
<tr>
<td>Latte-1</td>
<td>2557 min, 4355 samples</td>
<td>170 min, 290 samples</td>
<td>15 min, 25 samples</td>
</tr>
<tr>
<td>ModelScope</td>
<td>1160 min, 4355 samples</td>
<td>77 min, 290 samples</td>
<td>6 min, 23 samples</td>
</tr>
<tr>
<td>VideoCrafter-0.9</td>
<td>1459 min, 4355 samples</td>
<td>97 min, 290 samples</td>
<td>8 min, 24 samples</td>
</tr>
<tr>
<td>VideoCrafter-2</td>
<td>4261 min, 4355 samples</td>
<td>284 min, 290 samples</td>
<td>23 min, 23 samples</td>
</tr>
</tbody>
</table>
</div>
<p style="margin-top: 0;">
<b>Time Cost Comparison Across Models for VBench Dimensions.</b> This table compares the evaluation time of four different models using the original VBench pipelines versus the Evaluation Agent. The Evaluation Agent significantly reduces the overall evaluation time.
</p>
<br>
<div class="section-title" style="margin-top:10px;">
<h4 class="title is-5 is-centered">Compare Time Consumption on the T2I-CompBench</h5>
</div>
<div class="table-container" style="text-align: center; ">
<table class="nowrap-table" style="width:50%; text-align:center; border-collapse:collapse; margin: 0 auto;">
<thead>
<tr>
<th>Models</th>
<th>T2I-Comp (Total Cost) ↓</th>
<th>T2I-Comp (Avg. Cost per Dimension) ↓</th>
<th>Evaluation Agent (Ours) ↓</th>
</tr>
</thead>
<tbody>
<tr>
<td>SD-1.4</td>
<td>563 min, 12000 samples</td>
<td>141 min, 3000 samples</td>
<td>5 min, 26 samples</td>
</tr>
<tr>
<td>SD-2.1</td>
<td>782 min, 12000 samples</td>
<td>195 min, 3000 samples</td>
<td>5 min, 26 samples</td>
</tr>
<tr>
<td>SDXL</td>
<td>1543 min, 12000 samples</td>
<td>386 min, 3000 samples</td>
<td>8 min, 26 samples</td>
</tr>
<tr>
<td>SD-3.0</td>
<td>1410 min, 12000 samples</td>
<td>352 min, 3000 samples</td>
<td>7 min, 25 samples</td>
</tr>
</tbody>
</table>
</div>
<p style="margin-top: 0;">
<b>Time Cost Comparison Across Models for T2I-CompBench Dimensions.</b> This table compares the evaluation costs for assessing four models across T2I-CompBench dimensions using both the original T2I-CompBench pipelines and our Evaluation Agent. The Evaluation Agent achieves a substantial reduction in evaluation time compared to the traditional pipelines.
</p>
<br>
</div>
<div class="section-title" style="margin-top:30px;">
<h4 class="title is-5 is-centered">Performance across VBench Dimensions with Different Base Models</h5>
</div>
<!-- style="text-align: center;" -->
<div class="container is-max-desktop" >
<div class="publication-img">
<img id="architecture" src="./assets/images/fig_performance_ind.png" style="width:100%; margin-top:10px;margin-bottom:10px;"/>
</div>
<p style="margin-top: 0; margin-bottom:60px;">
<b>Performance across VBench Dimensions for Different Base Models.</b> This visualization highlights the performance of all backbone models, including GPT-4o and Claude models, providing a comprehensive comparison in each dimension for different backbone models. Hatched portions indicate predictions within the exact range, and solid portions within an error margin of one range.
</p>
<br>
</div>
</section>
<!-- Open Queries Set -->
<section class="section" style="margin-top:-95px; margin-bottom:30px;">
<div class="container is-max-desktop">
<div class="hero-body">
<div class="section-title">
<h2 class="title is-3 is-centered">Statistics of the Open-Ended User Query Dataset</h2>
</div>
<div class="columns is-centered has-text-centered">
<div class="column">
<div class="publication-img">
<img id="architecture" src="./assets/images/open_dataset_stats.png" style="width:100%; margin-top:30px;margin-bottom:10px;"/>
</div>
</div>
</div>
<p style="margin-top: 0;">
<b>Data Distribution of Open-Ended User Query Dataset.</b> We analyze the constructed open-ended user query dataset from three aspects: General/Specific, Ability, and Specific Domain. The results indicate that our dataset exhibits a relatively balanced distribution across these dimensions.
The three graphs give an overview of the distributions and types of our curated open-ended user queries dataset. <b>Left: </b>the distribution of query types, which are categorized as <i>General</i> or <i>Specific</i>. <b>Middle: </b>the distribution of the ability types, which are categorized as <i>Prompt Following</i>, <i>Visual Quality</i>, <i>Creativity</i>, <i>Knowledge</i> and <i>Others</i>. <b>Right: </b>the distribution of the content categories, which are categorized as
<i>History and Culture</i>,
<i>Film and Entertainment</i>,
<i>Science and Education</i>,
<i>Fashion</i>,
<i>Medical</i>,
<i>Game Design</i>,
<i>Architecture and Interior Design</i>, <i>Law</i>.
</p>
</div>
</div>
</section>
<section class="section" style="margin-top:-130px; margin-bottom:-80px;">
<div class="hero-body">
<div class="container is-max-desktop">
<div class="section-title">
<h2 class="title is-3" style="margin-bottom: 25pt;">Evaluation Cases for Open-Ended User Queries</h2>
<h2 class="subtitle has-text-centered">(Display of Comprehensive Evaluation Results for Open-Ended User Queries)</h2>
</div>
<div id="results-carousel-face" class="carousel results-carousel">
<div class="item item-puppet">
<div class="carousel-content">
<div class="publication-img">
<video id="slideshow" style="width:100%; margin-top:10px;margin-bottom:10px;" autoplay="" muted="" loop="" playsinline="" controls>
<source src="./assets/images/open-3.mov" type="video/mp4">
Your browser does not support the video tag.
</div>
</div>
</div>
<div class="item item-puppet">
<div class="carousel-content">
<div class="publication-img">
<video id="slideshow" style="width:100%; margin-top:10px;margin-bottom:10px;" autoplay="" muted="" loop="" playsinline="" controls>
<source src="./assets/images/open-1.mov" type="video/mp4">
Your browser does not support the video tag.
</div>
</div>
</div>
<div class="item item-round_bird">
<div class="carousel-content">
<div class="publication-img">
<video id="slideshow" style="width:100%; margin-top:10px;margin-bottom:10px;" autoplay="" muted="" loop="" playsinline="" controls>
<source src="./assets/images/open-2.mp4" type="video/mp4">
Your browser does not support the video tag.
</div>
</div>
</div>
</div>
</div>
</section>
<section class="hero is-light is-small" id="BibTeX" >
<div class="container is-max-desktop content" style="margin-top: 40px; margin-bottom: 20px;">
<h2 class="title">BibTeX</h2>
<p>If you find our work useful, please consider citing our paper:</p>
<pre><code>@article{zhang2024evaluationagent,
title = {Evaluation Agent: Efficient and Promptable Evaluation Framework for Visual Generative Models},
author = {Zhang, Fan and Tian, Shulin and Huang, Ziqi and Qiao, Yu and Liu, Ziwei},
journal={arXiv preprint arXiv:2412.09645},
year = {2024}
}</code></pre>
</div>
</section>
<!-- footer -->
<footer class="footer">
<div class="container">
<div class="content has-text-centered">
<a class="icon-link"
href="https://arxiv.org/abs/2412.09645">
<i class="fas fa-file-pdf"></i>
</a>
<a class="icon-link" href="https://github.com/Vchitect/Evaluation-Agent" class="external-link" disabled>
<i class="fab fa-github"></i>
</a>
</div>
<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 source code based on the <a href="https://nerfies.github.io/"> Nerfies</a> project page. If you want to reuse their <a
href="https://github.com/nerfies/nerfies.github.io">source code</a>, please credit them appropriately.
</p>
</div>
</div>
</div>
</div>
</footer>
</body>
</html>