-
Notifications
You must be signed in to change notification settings - Fork 2
/
index.html
497 lines (456 loc) · 20.6 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
<!doctype html>
<html lang="en">
<head>
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1">
<link href="bootstrap.min.css" rel="stylesheet" />
<meta name="twitter:card" content="summary_large_image" />
<!-- anon-on -->
<meta name="twitter:site" content="@GoogleAI" />
<!-- anon-off -->
<meta name="twitter:title" content="OVEN: Open-domain Visual Entity Recognition" />
<meta name="twitter:description" content="We formally present the task of <i>Open-domain Visual Entity recognitioN (OVEN)</i>, where a model need to link an image onto a Wikipedia entity with respect to a text query. We construct OVEN-Wiki by re-purposing 14 existing datasets with all labels grounded onto one single label space: Wikipedia entities. OVEN challenges models to select among six million possible Wikipedia entities, making it a general visual recognition benchmark with the largest number of labels." />
<link rel="preconnect" href="https://fonts.googleapis.com">
<link rel="preconnect" href="https://fonts.gstatic.com" crossorigin>
<link href="https://fonts.googleapis.com/css2?family=Roboto:wght@300&display=swap" rel="stylesheet">
<link href="https://fonts.googleapis.com/css2?family=Google+Sans:wght@400" rel="stylesheet">
<link rel="stylesheet" type="text/css" href="styles.css" />
<!-- Google tag (gtag.js) -->
<script async src="https://www.googletagmanager.com/gtag/js?id=G-55TFRQJMDH"></script>
<script>
window.dataLayer = window.dataLayer || [];
function gtag(){dataLayer.push(arguments);}
gtag('js', new Date());
gtag('config', 'G-55TFRQJMDH');
</script>
<script src="jquery.min.js"></script>
<script src="jquery.flip.min.js"></script>
<title>OVEN: Open-domain Visual Entity Recognition</title>
</head>
<body>
<div id="gallery0" style="padding-top: 50px; padding-bottom: 20px; background-color: rgb(245,245,245);">
<h2 style="text-align: center;">Open-domain Visual Entity Recognition</h2>
<h2 style="text-align: center;">Towards Recognizing Millions of Wikipedia Entities</h2>
<!-- anon-on -->
<div class="authors text-center">
<div class="row">
<div class="col">Hexiang Hu<sup>1</sup></div>
<div class="col">Yi Luan<sup>1</sup></div>
<div class="col">Yang Chen<sup>1,2*</sup></div>
<div class="col">Urvashi Khandelwal <sup>1</sup></div>
</div>
<div class="row">
<div class="col">Mandar Joshi<sup>1</sup></div>
<div class="col">Kenton Lee<sup>1</sup></div>
<div class="col">Kristina Toutanova<sup>1</sup></div>
<div class="col">Ming-Wei Chang<sup>1</sup></div>
</div>
<div class="row mt-3">
<div class="col">
<sup>1</sup><img src="assets/gdm-logo.svg" alt="Google Deepmind" height=35>
</div>
<div class="col">
<sup>2</sup><img src="https://brand.gatech.edu/sites/default/files/inline-images/extended-RGB.png" alt="Georgia Tech" height=45>
</div>
</div>
<div class="row">
<div class="col text-small">(*: Work done when author was interned at Google)</div>
</div>
<div class="row mt-0 nomobile">
<center><a href="https://arxiv.org/abs/2302.11154" target="_blank"><img class="thumbnail" src="assets/oven_thumbnail.png"></a></center>
</div>
</div>
<!-- anon-off -->
</div>
<div class="abstract">
<div class="inside">
<p class="text">
Large-scale multi-modal pre-training models such as CLIP and PaLI exhibit strong generalization on various visual domains and tasks. However, existing image classification benchmarks often evaluate recognition on a specific domain (e.g., outdoor images) or a specific task (e.g., classifying plant species), which falls short of evaluating whether pre-trained foundational models are universal visual recognizers. To address this, we formally present the task of <i>Open-domain Visual Entity recognitioN (OVEN)</i>, where a model need to link an image onto a Wikipedia entity with respect to a text query. We construct OVEN-Wiki by re-purposing 14 existing datasets with all labels grounded onto one single label space: Wikipedia entities. OVEN challenges models to select among six million possible Wikipedia entities, making it a general visual recognition benchmark with the largest number of labels. Our study on state-of-the-art pre-trained models reveals large headroom in generalizing to the massive-scale label space. We show that a PaLI-based auto-regressive visual recognition model performs surprisingly well, even on Wikipedia entities that have never been seen during fine-tuning. We also find existing pretrained models yield different strengths: while PaLI-based models obtain higher overall performance, CLIP-based models are better at recognizing tail entities.
</p>
<!-- anon-on -->
<a class="read-paper" href="https://arxiv.org/abs/2302.11154" target="_blank"><button>Research Paper</button></a>
<a class="read-paper" href="https://github.com/open-vision-language/oven", target="_blank"><button>Dataset</button></a>
<a class="read-paper" href="https://github.com/edchengg/oven_eval", target="_blank"><button>Contributed Code</button></a>
<a class="dataset" href="https://open-vision-language.github.io/infoseek/", target="_blank"><button>Next: InfoSeek</button></a>
<!-- anon-off -->
</div>
</div>
<div class="header_dark_gray text-center">
<h1>OVEN models recognize the <i>Visual Entity on the Wikipedia</i>, from images in the wild</h1>
</div>
<div class="image_8" id="gallery1">
<div class="flipcard" id="fig0">
<div class="front">
<figure>
<img>
<figcaption class="question"></figcaption>
<figcaption class="answer"></figcaption>
</figure>
</div>
<div class="back">
<figure>
<img>
<figcaption class="question"></figcaption>
<figcaption class="answer"></figcaption>
</figure>
</div>
</div>
<div class="flipcard nomobile" id="fig1">
<div class="front">
<figure>
<img>
<figcaption class="question"></figcaption>
<figcaption class="answer"></figcaption>
</figure>
</div>
<div class="back">
<figure>
<img />
<figcaption class="question"></figcaption>
<figcaption class="answer"></figcaption>
</figure>
</div>
</div>
<div class="flipcard nomobile" id="fig2">
<div class="front">
<figure>
<img>
<figcaption class="question"></figcaption>
<figcaption class="answer"></figcaption>
</figure>
</div>
<div class="back">
<figure>
<img>
<figcaption class="question"></figcaption>
<figcaption class="answer"></figcaption>
</figure>
</div>
</div>
<div class="flipcard nomobile" id="fig3">
<div class="front">
<figure>
<img>
<figcaption class="question"></figcaption>
<figcaption class="answer"></figcaption>
</figure>
</div>
<div class="back">
<figure>
<img>
<figcaption class="question"></figcaption>
<figcaption class="answer"></figcaption>
</figure>
</div>
</div>
<div class="flipcard nomobile" id="fig4">
<div class="front">
<figure>
<img>
<figcaption class="question"></figcaption>
<figcaption class="answer"></figcaption>
</figure>
</div>
<div class="back">
<figure>
<img>
<figcaption class="question"></figcaption>
<figcaption class="answer"></figcaption>
</figure>
</div>
</div>
<div class="flipcard nomobile" id="fig5">
<div class="front">
<figure>
<img>
<figcaption class="question"></figcaption>
<figcaption class="answer"></figcaption>
</figure>
</div>
<div class="back">
<figure>
<img>
<figcaption class="question"></figcaption>
<figcaption class="answer"></figcaption>
</figure>
</div>
</div>
<div class="flipcard nomobile" id="fig6">
<div class="front">
<figure>
<img>
<figcaption class="question"></figcaption>
<figcaption class="answer"></figcaption>
</figure>
</div>
<div class="back">
<figure>
<img>
<figcaption class="question"></figcaption>
<figcaption class="answer"></figcaption>
</figure>
</div>
</div>
<div class="flipcard nomobile" id="fig7">
<div class="front">
<figure>
<img>
<figcaption class="question"></figcaption>
<figcaption class="answer"></figcaption>
</figure>
</div>
<div class="back">
<figure>
<img>
<figcaption class="question"></figcaption>
<figcaption class="answer"></figcaption>
</figure>
</div>
</div>
</div>
<!-- anon-on -->
<div class="content">
<h1>Special Thanks</h1>
<p class="smaller" style="text-align: justify;"> We thank Boqing Gong, Soravit Changpinyo for reviewing on an early version of this paper in depth, with valuable comments and suggestions. We thank Xi Chen for providing different variants of PaLI pre-trained checkpoints. We thank Huiwen Chang and the Muse Team for providing their website template. We also thank Radu Soricut, Anelia Angelova, Alan Ritter, Chao-Yuan Wu, Jiacheng Chen for discussions and feedback on the project.
</p>
</div>
<!-- anon-off -->
<script type="text/javascript">
gallery_images = [
{
"src": "./assets/oven-images/fungi1.png",
"question": "<span style='color: #c994c7'>Query</span>: Which species of fungi is this?",
"answer": "<span style='color: #dd1c77'>Entity</span>: <a href='https://en.wikipedia.org/wiki/Omphalotus_illudens' target='_blank'>Omphalotus illudens (Q7090729)</a>"
},
{
"src": "./assets/oven-images/bird1.png",
"question": "<span style='color: #c994c7'>Query</span>: What is the species of this animal?",
"answer": "<span style='color: #dd1c77'>Entity</span>: <a href='https://en.wikipedia.org/wiki/Ash-throated_flycatcher' target='_blank'>Ash-throated flycatcher (Q650369)</a>",
},
{
"src": "./assets/oven-images/mountain1.png",
"question": "<span style='color: #c994c7'>Query</span>: What is this mountain called?",
"answer": "<span style='color: #dd1c77'>Entity</span>: <a href='https://en.wikipedia.org/wiki/Sawtooth_Range_(Idaho)' target='_blank'>Sawtooth Range (Q7428707)</a>",
},
{
"src": "./assets/oven-images/food1.png",
"question": "<span style='color: #c994c7'>Query</span>: What dish is pictured?",
"answer": "<span style='color: #dd1c77'>Entity</span>: <a href='https://en.wikipedia.org/wiki/Ceviche' target='_blank'>Ceviche (Q1057953)</a>",
},
{
"src": "./assets/oven-images/beer1.png",
"question": "<span style='color: #c994c7'>Query</span>: What brand of beer is in the bucket?",
"answer": "<span style='color: #dd1c77'>Entity</span>: <a href='https://en.wikipedia.org/wiki/Estrella_Galicia' target='_blank'>Estrella Galicia (Q3394030)</a>",
},
{
"src": "./assets/oven-images/aircraft1.png",
"question": "<span style='color: #c994c7'>Query</span>: What model of aircraft is shown?",
"answer": "<span style='color: #dd1c77'>Entity</span>: <a href='https://en.wikipedia.org/wiki/Cessna_172' target='_blank'>Cessna 172 (Q244479)</a>"
},
{
"src": "./assets/oven-images/shoe1.png",
"question": "<span style='color: #c994c7'>Query</span>: What kind of shoes is this person wearing?",
"answer": "<span style='color: #dd1c77'>Entity</span>: <a href='https://en.wikipedia.org/wiki/Flip-flops' target='_blank'>Flip-flops (Q380339)</a>",
},
{
"src": "./assets/oven-images/material1.png",
"question": "<span style='color: #c994c7'>Query</span>: What is that statue made out of?",
"answer": "<span style='color: #dd1c77'>Entity</span>: <a href='https://en.wikipedia.org/wiki/Cast_iron' target='_blank'>Cast iron (Q483269)</a>",
},
{
"src": "./assets/oven-images/car1.png",
"question": "<span style='color: #c994c7'>Query</span>: What is the make and model of this car?",
"answer": "<span style='color: #dd1c77'>Entity</span>: <a href='https://en.wikipedia.org/wiki/Nissan_Juke' target='_blank'>Nissan Juke (Q1415815)</a>"
},
{
"src": "./assets/oven-images/place1.png",
"question": "<span style='color: #c994c7'>Query</span>: What is the place these guys are fighting called?",
"answer": "<span style='color: #dd1c77'>Entity</span>: <a href='https://en.wikipedia.org/wiki/Wrestling_ring' target='_blank'> Wrestling ring (Q6763849)</a>",
},
{
"src": "./assets/oven-images/food2.png",
"question": "<span style='color: #c994c7'>Query</span>: What is the name of the dish shown?",
"answer": "<span style='color: #dd1c77'>Entity</span>: <a href='https://en.wikipedia.org/wiki/French_onion_soup' target='_blank'>French onion soup (Q244617)</a>"
},
{
"src": "./assets/oven-images/place2.png",
"question": "<span style='color: #c994c7'>Query</span>: What is the name of the location at the end of the street?",
"answer": "<span style='color: #dd1c77'>Entity</span>: <a href='https://en.wikipedia.org/wiki/Palais_Longchamp' target='_blank'>Palais Longchamp (Q1619084)</a>"
},
];
lastCardFlippedIdx = {
gallery1: -1,
};
lastCardFlippedLastSwapTime = {
gallery1: -1,
};
$(".flipcard").flip({
trigger: "manual",
reverse: false,
});
function checkIfImageIsVisible(name, image) {
numCards = $(name + " .flipcard:visible").size();
for (var cardIdx = 0; cardIdx < numCards; ++cardIdx) {
var isFlipped = $(name + " #fig" + cardIdx).data("flip-model").isFlipped;
if (!isFlipped) {
src = $(name + " #fig" + cardIdx + " div.front figure img").attr("src")
} else {
src = $(name + " #fig" + cardIdx + " div.back figure img").attr("src")
}
if (image.src == src) {
return true;
}
}
return false;
}
function getNextImage(name) {
// different galleries have different sources of imagery.
img_src = gallery_images;
while (true) {
randomImageIdx = Math.floor(Math.random() * img_src.length);
if (!checkIfImageIsVisible(name, img_src[randomImageIdx])) {
return img_src[randomImageIdx];
}
}
}
function randomSwapImageWithId(name, randomCardIdx) {
numCards = $(name + " .flipcard:visible").size();
image = getNextImage(name);
const d = new Date();
let currentTime = d.getTime();
lastCardFlippedIdx[name] = randomCardIdx;
lastCardFlippedLastSwapTime[name] = currentTime;
var isFlipped = $(name + " #fig" + randomCardIdx).data("flip-model").isFlipped;
if (!isFlipped) {
$(name + " #fig" + randomCardIdx + " div.back figure img").attr("src", image.src)
$(name + " #fig" + randomCardIdx + " div.back figure figcaption.question").html(image.question)
$(name + " #fig" + randomCardIdx + " div.back figure figcaption.answer").html(image.answer)
} else {
$(name + " #fig" + randomCardIdx + " div.front figure img").attr("src", image.src)
$(name + " #fig" + randomCardIdx + " div.front figure figcaption.question").html(image.question)
$(name + " #fig" + randomCardIdx + " div.front figure figcaption.answer").html(image.answer)
}
$(name + " #fig" + randomCardIdx).flip("toggle");
}
function randomSwapImage(name) {
numCards = $(name + " .flipcard:visible").size();
const d = new Date();
if (d.getTime() - lastCardFlippedIdx[name] < 5000) {
setTimeout(randomSwapImage, 1000, name);
return;
}
do {
randomCardIdx = Math.floor(Math.random() * numCards);
if (numCards == 1) {
break;
}
} while (randomCardIdx == lastCardFlippedIdx[name]);
randomSwapImageWithId(name, randomCardIdx);
setTimeout(randomSwapImage, 3000, name);
}
function setupGallery1(id) {
randomSwapImageWithId("#gallery1", id);
$("#gallery1 #fig" + id).click(
function() {
randomSwapImageWithId("#gallery1", id);
}
);
}
for (id = 0; id < 8; id ++) {
setupGallery1(id);
}
setTimeout(randomSwapImage, 1000, "#gallery1");
</script>
<script type="text/javascript">
compositionalAnimalsManualClicked = false;
compositionalMadeOfManualClicked = false;
comp_files = {
"Original (with mask)": "./images/inpaint/boston_masked.jpg",
"\"New York in the background\"": "./images/inpaint/boston_nyc.jpg",
"\"Paris in the background\"": "./images/inpaint/boston_paris.jpg",
"\"San Francisco in the background\"": "./images/inpaint/boston_sf.jpg"
}
function updateComposition(name) {
var compositionSpans = $(name + " p span.selected").map(function() {
return $(this).text();
}).get();
var compositionText = "";
for (var compositionSpansIdx = 0; compositionSpansIdx < compositionSpans.length; ++compositionSpansIdx) {
if (compositionSpansIdx > 0) {
compositionText += " ";
}
compositionText += compositionSpans[compositionSpansIdx];
}
if (name == "#compositional_madeof") {
if (compositionText == "Original") {
$(name + "_img").attr("src", "./images/edit_opt/orig.gif")
} else {
// Generate a random index that is not the one already selected
randImgIdx = "/"; // something guaranteed to be in the path
cur_src = $(name + "_img").attr("src");
while (cur_src.search(randImgIdx) != -1) {
randImgIdx = Math.floor(Math.random() * 4);
}
first_space = compositionText.search(" ");
file = compositionText.slice(first_space + 1, -2).replaceAll(" ", "-");
$(name + "_img").attr("src", "./images/edit_opt/" + file + "_" + randImgIdx + ".gif")
}
} else if (name == "#compositional_animals") {
randImgIdx = Math.floor(Math.random() * 4);
$(name + "_img").attr("src", comp_files[compositionText]);
}
}
function randomComposition(name) {
if (name == "#compositional_animals" && compositionalAnimalsManualClicked == true) {
compositionalAnimalsManualClicked = false;
setTimeout(randomComposition, 4000, name);
return;
} else if (name == "#compositional_madeof" &&
compositionalMadeOfManualClicked == true) {
compositionalMadeOfManualClicked = false;
setTimeout(randomComposition, 4000, name);
return;
}
var compositionPs = $(name + " p.selectable").toArray();
var compositionOptions = [];
for (var i = 0; i < compositionPs.length; ++i) {
var compositionSpanSelected = $(compositionPs[i]).children("span.selected");
var compositionSpans = $(compositionPs[i]).children("span").not("span.selected").toArray();
for (var j = 0; j < compositionSpans.length; ++j) {
compositionOption = {
"p": compositionPs[i],
"span_selected": compositionSpanSelected,
"span_option": $(compositionSpans[j])
};
compositionOptions.push(compositionOption);
}
}
randomCompositionOptionIdx = Math.floor(Math.random() * compositionOptions.length);
randomCompositionOption = compositionOptions[randomCompositionOptionIdx];
randomCompositionOption.span_selected.toggleClass("selected");
randomCompositionOption.span_option.toggleClass("selected");
updateComposition(name);
setTimeout(randomComposition, 4000, name);
}
$("#compositional_animals p.selectable span").click(function() {
$(this).siblings().removeClass("selected");
$(this).addClass("selected");
updateComposition("#compositional_animals");
compositionalAnimalsManualClicked = true;
});
$("#compositional_madeof_img").click(function() {
updateComposition("#compositional_madeof");
compositionalAnimalsManualClicked = true;
});
$("#compositional_madeof p.selectable span").click(function() {
$(this).siblings().removeClass("selected");
$(this).addClass("selected");
updateComposition("#compositional_madeof");
compositionalMadeOfManualClicked = true;
});
<!-- setTimeout(randomComposition, 2000, "#compositional_animals"); -->
<!-- setTimeout(randomComposition, 4000, "#compositional_madeof"); -->
</script>
<script src="bootstrap.bundle.min.js"></script>
</body>
</html>