-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathindex.html
More file actions
1857 lines (1654 loc) · 97.3 KB
/
index.html
File metadata and controls
1857 lines (1654 loc) · 97.3 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
<!doctype html>
<html lang="en">
<head>
<style>
pre, code {
font-size: 16px;
}
</style>
<title>Thinking in Space: How Multimodal Large Language Models See, Remember and Recall Spaces</title>
<link rel="icon" type="image/x-icon" href="./static/img/icons/brain.ico">
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1">
<!-- Open Graph -->
<meta property="og:url" content="https://vision-x-nyu.github.io/thinking-in-space.github.io/" />
<meta property="og:image" content="./static/img/preview.png" />
<meta property="og:title" content="Thinking in Space: How Multimodal Large Language Models See, Remember and Recall Spaces" />
<!-- Twitter -->
<meta name="twitter:url" content="https://vision-x-nyu.github.io/thinking-in-space.github.io/" />
<meta name="twitter:card" content="summary_large_image" />
<meta name="twitter:image" content="./static/img/preview.png" />
<meta name="twitter:title" content="Thinking in Space: How Multimodal Large Language Models See, Remember and Recall Spaces" />
<meta name="twitter:description" content="We introduce VSI-Bench, a novel benchmark of over 5,000 video-based visual-spatial intelligence questions, to evaluate and probe MLLMs, which revealed that their emerging spatial reasoning and local world modeling capabilities remain subhuman but promising." />
<!-- JS Libraries -->
<script src="./static/js/distill_template.v2.js"></script>
<script src="https://polyfill.io/v3/polyfill.min.js?features=es6"></script>
<script id="MathJax-script" async src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-mml-chtml.js"></script>
<script src="https://d3js.org/d3.v5.min.js"></script>
<script src="https://d3js.org/d3-collection.v1.min.js"></script>
<script src="https://rawgit.com/nstrayer/slid3r/master/dist/slid3r.js"></script>
<script defer="" src="./static/js/hider.js"></script>
<script defer src="./static/js/image_interact.js"></script>
<script defer src="./static/js/switch_videos.js"></script>
<script defer src="./static/js/video-speed.js"></script>
<!-- Stylesheets -->
<link rel="stylesheet" href="./static/css/style.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">
<!-- KaTeX -->
<link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/katex@0.10.2/dist/katex.min.css" integrity="sha384-yFRtMMDnQtDRO8rLpMIKrtPCD5jdktao2TV19YiZYWMDkUR5GQZR/NOVTdquEx1j" crossorigin="anonymous">
<script defer src="https://cdn.jsdelivr.net/npm/katex@0.10.2/dist/katex.min.js" integrity="sha384-9Nhn55MVVN0/4OFx7EE5kpFBPsEMZxKTCnA+4fqDmg12eCTqGi6+BB2LjY8brQxJ" crossorigin="anonymous"></script>
<script defer src="https://cdn.jsdelivr.net/npm/katex@0.10.2/dist/contrib/auto-render.min.js" integrity="sha384-kWPLUVMOks5AQFrykwIup5lo0m3iMkkHrD0uJ4H5cjeGihAutqP0yW0J6dpFiVkI" crossorigin="anonymous" onload="renderMathInElement(document.body);"></script>
<!-- FontAwesome -->
<script defer src="./static/js/fontawesome.all.min.js"></script>
<!-- Medium Zoom -->
<script src="https://cdn.jsdelivr.net/npm/jquery@3.7.1/dist/jquery.min.js"></script>
<script defer src="./static/js/medium-zoom.min.js"></script>
<script defer src="./static/js/zoom.js"></script>
<!-- Leaderboard Styles -->
<style>
.leaderboard-table {
width: 140%;
margin-left: -20%;
border-collapse: separate;
border-spacing: 0;
background: #ffffff;
border-radius: 8px;
overflow: hidden;
box-shadow: 0 4px 6px -1px rgba(0, 0, 0, 0.1), 0 2px 4px -1px rgba(0, 0, 0, 0.06);
font-family: -apple-system, BlinkMacSystemFont, 'Segoe UI', Roboto, Oxygen, Ubuntu, sans-serif;
font-size: 14px;
margin-bottom: 40px;
}
.leaderboard-container {
position: relative;
}
.leaderboard-table thead {
background: #4a5568;
color: white;
}
.leaderboard-table th {
padding: 14px 12px;
text-align: center;
font-weight: 600;
font-size: 13px;
letter-spacing: 0.025em;
border-bottom: 1px solid rgba(255, 255, 255, 0.1);
position: relative;
}
.sortable-header {
cursor: pointer;
user-select: none;
}
.sortable-header .sort-indicator {
position: absolute;
right: 8px;
top: 50%;
font-size: 14px;
line-height: 1;
transform: translateY(-50%);
color: #a0aec0;
}
.sortable-header.active .sort-indicator {
color: white;
}
.section-header {
background-color: #deeef8;
font-weight: 700;
color: #2d3748;
letter-spacing: 1px;
font-size: 18px;
text-align: center;
padding: 12px !important;
border-bottom: 2px solid #e2e8f0;
}
.leaderboard-table td {
padding: 12px;
border-bottom: 1px solid #e2e8f0;
color: #4a5568;
text-align: center;
}
.leaderboard-table tbody tr {
transition: background-color 0.15s ease;
}
.leaderboard-table tbody tr:hover {
background-color: #f7fafc;
}
.leaderboard-table td:first-child {
font-weight: 600;
text-align: left;
color: #2d3748;
padding-left: 20px;
}
.leaderboard-table td:nth-child(5) {
font-weight: 700;
font-size: 15px;
color: #2d3748;
}
.human-level-row {
background: #fffbeb85 !important;
font-weight: 600;
}
.top-performer {
background: #f0fdf494 !important;
}
.top-performer td:first-child::before {
content: '🥇 ';
}
.model-badge {
display: inline-block;
padding: 2px 6px;
border-radius: 4px;
font-size: 11px;
font-weight: 500;
margin-left: 8px;
text-transform: uppercase;
letter-spacing: 0.025em;
}
.proprietary-badge {
background: #e2e8f0;
color: #475569;
border: 1px solid #cbd5e0;
}
.open-badge {
background: #e6fffa;
color: #047857;
border: 1px solid #a7f3d0;
}
.leaderboard-table tbody tr:nth-child(even) {
background-color: #fafafa;
}
.baseline-row {
font-style: italic;
opacity: 0.8;
background-color: #f9fafb;
}
.best-score {
font-weight: 700;
text-decoration: underline;
text-decoration-color: #cbd5e0;
text-underline-offset: 2px;
}
.section-divider {
border-top: 2px solid #e2e8f0;
}
@media (max-width: 768px) {
.leaderboard-container {
overflow-x: auto;
-webkit-overflow-scrolling: touch;
scroll-behavior: smooth;
border-radius: 8px;
box-shadow: 0 4px 6px -1px rgba(0, 0, 0, 0.1), 0 2px 4px -1px rgba(0, 0, 0, 0.06);
}
.leaderboard-table {
width: 1200px; /* Fixed width to ensure horizontal scrolling */
margin-left: 0;
font-size: 12px;
min-width: 1200px;
}
.leaderboard-table th,
.leaderboard-table td {
padding: 8px;
white-space: nowrap;
min-width: 80px;
}
.leaderboard-table td:first-child {
min-width: 180px;
max-width: 180px;
overflow: hidden;
text-overflow: ellipsis;
padding-left: 12px;
}
/* Add scroll hint for mobile users */
.leaderboard-container::after {
content: "← Swipe to scroll →";
position: absolute;
bottom: -25px;
left: 50%;
transform: translateX(-50%);
font-size: 11px;
color: #666;
font-style: italic;
pointer-events: none;
}
}
</style>
</head>
<body>
<div class="header-wrapper">
<div class="header-container" id="header-container">
<div class="header-content">
<h1 style="margin-top: 0px"><i>Thinking in Space</i></h1>
<div class="responsive-header">
<h2>How Multimodal Large Language Models See, Remember and Recall Spaces</h2>
</div>
<div class="icon-container">
<div class="icon-item">
<img src="./static/img/icons/bench.svg" alt="Visual Representation Icon">
<div><strong>VSI-Bench</strong>: We introduce a high-quality benchmark for the evaluation of 3D, video-based, visual-spatial intelligence of MLLMs.</div>
</div>
<div class="icon-item">
<img src="./static/img/icons/evaluation.svg" alt="Connector Design Icon">
<div><strong>Evaluation</strong>: We evaluate VSI-Bench on open- and closed-source MLLMs and find that MLLMs exhibit competitive—though subhuman—visual-spatial intelligence.</div>
</div>
<div class="icon-item">
<img src="./static/img/icons/speech.svg" alt="Speech Logo" class="icon">
<div><strong>Linguistic Analysis</strong>: We attribute VSI-Bench performance to spatial intelligence capabilities and show the differences between spatial and linguistic intelligence.</div>
</div>
<div class="icon-item">
<img src="./static/img/icons/vision.svg" alt="Vision Logo" class="icon">
<div><strong>Visual Analysis</strong>: We illuminate how MLLMs remember spaces via cognitive maps and show how explicitly remembering spaces improves spatial capabilities.</div>
</div>
</div>
<div class="button-container">
<a href="https://arxiv.org/abs/2412.14171" class="button paper-link" target="_blank">
<span class="icon is-small">
<i class="ai ai-arxiv"></i>
</span>
arXiv
</a>
<a href="https://arxiv.org/pdf/2412.14171" class="button paper-link" target="_blank">
<span class="icon is-small">
<i class="fas fa-file-pdf"></i>
</span>
<span>PDF</span>
</a>
<a href="https://github.com/vision-x-nyu/thinking-in-space" class="button" target="_blank">
<span class="icon is-small">
<i class="fab fa-github"></i>
</span>
<span>Code</span>
</a>
<a href="https://huggingface.co/datasets/nyu-visionx/VSI-Bench" class="button" target="_blank">
<span class="icon is-small">
<img src="https://huggingface.co/front/assets/huggingface_logo-noborder.svg" alt="Hugging Face logo" style="height: 1em;">
</span>
<span>VSI-Bench</span>
</a>
<a href="#vsi-leaderboard" class="button">
<span class="icon is-small">
<img src="./static/img/leaderboard-star-svgrepo-com.svg" alt="Leaderboard logo" style="height: 1em;">
</span>
<span>Leaderboard</span>
</a>
</div>
</div>
<div class="header-image">
<img draggable="false" src="static/img/preview.png" alt="Teaser Image" class="teaser-image">
</div>
</div>
</div>
<d-article>
<div class="byline">
<div class="byline-container">
<div class="byline-column" style="line-height: 1.2;">
<h3 style="font-size: 1.2em; margin-bottom: 0.2em;">Authors</h3>
<p style="margin: 0;">
<a href="https://jihanyang.github.io/" class="author-link" target="_blank" style="font-size: 0.9em;">Jihan Yang</a>
<sup style="font-size: 0.9em;">△*</sup>
</p>
<p style="margin: 0;">
<a href="https://github.com/vealocia" class="author-link" target="_blank" style="font-size: 0.9em;">Shusheng Yang</a>
<sup style="font-size: 0.9em;">△*</sup>
</p>
<p style="margin: 0;">
<a href="https://anjaliwgupta.com" class="author-link" target="_blank" style="font-size: 0.9em;">Anjali W. Gupta</a>
<sup style="font-size: 0.9em;">△*</sup>
</p>
<p style="margin: 0;">
<a href="https://rilynhan.github.io" class="author-link" target="_blank" style="font-size: 0.9em;">Rilyn Han</a>
<sup style="font-size: 0.9em;">▲*</sup>
</p>
<p style="margin: 0;">
<a href="https://profiles.stanford.edu/fei-fei-li" class="author-link" target="_blank" style="font-size: 0.9em;">Li Fei-Fei</a>
<sup style="font-size: 0.9em;">★</sup>
</p>
<p style="margin: 0;">
<a href="https://www.sainingxie.com/" class="author-link" target="_blank" style="font-size: 0.9em;">Saining Xie</a>
<sup style="font-size: 0.9em;">△</sup>
</p>
</div>
<div class="byline-column" style="line-height: 1.2;">
<h3 style="font-size: 1.2em; margin-bottom: 0.2em;">Affiliations</h3>
<p style="margin: 0;">
<sup style="font-size: 0.8em;">△</sup>
<a href="https://cs.nyu.edu/home/index.html" class="affiliation-link" target="_blank" style="font-size: 0.9em;">New York University</a>
</p>
<p style="margin: 0;">
<sup style="font-size: 0.8em;">▲</sup>
<a href="https://www.yale.edu" class="affiliation-link" target="_blank" style="font-size: 0.9em;">Yale University</a>
</p>
<p style="margin: 0;">
<sup style="font-size: 0.8em;">★</sup>
<a href="https://www.stanford.edu/" class="affiliation-link" target="_blank" style="font-size: 0.9em;">Stanford University</a>
</p>
<br>
<br>
<br>
<br>
<br>
<p style="text-align: center; margin: 0;">
<span class="author-note"><sup>*</sup>Equal contribution</span> 
</p>
</div>
<div class="byline-column" style="line-height: 1.2;">
<h3 style="font-size: 1.2em; margin-bottom: 0.2em;">Date</h3>
<p style="margin: 0; font-size: 0.9em;">
Dec. 18<sup style="font-size: 0.8em;">th</sup>, 2024
</p>
</div>
</div>
</div>
<p style="text-align: center; font-weight: bold; margin-top: 20px;">
Compare your spatial intelligence abilities with Gemini!
</p>
<div class="l-body">
<!-- Preview Images in a Flex Container - placed below the Q&A sections -->
<div class="preview-container">
<img id="Touristvideo1Preview" class="preview" src="static/img/cropped_thumbnails/42445026_thumbnail_cropped.png" alt="video-thumbnail2233" onclick="switchVideo('Tourist', 'video1Container', 'video1Preview')">
<img id="Touristvideo2Preview" class="preview" src="static/img/cropped_thumbnails/scene0663_00_thumbnail_cropped.png" alt="video-thumbnail3" onclick="switchVideo('Tourist','video2Container', 'video2Preview')">
<img id="Touristvideo3Preview" class="preview" src="static/img/cropped_thumbnails/45662987_thumbnail_cropped.png" alt="video-thumbnail18" onclick="switchVideo('Tourist', 'video3Container', 'video3Preview')">
<img id="Touristvideo4Preview" class="preview" src="static/img/cropped_thumbnails/42445026_thumbnail_cropped.png" alt="video-thumbnail2233" onclick="switchVideo('Tourist', 'video4Container', 'video4Preview')">
<img id="Touristvideo5Preview" class="preview" src="static/img/cropped_thumbnails/41069025_thumbnail_cropped.png" alt="video-thumbnail1236" onclick="switchVideo('Tourist', 'video5Container', 'video5Preview')">
<img id="Touristvideo6Preview" class="preview" src="static/img/cropped_thumbnails/47334096_thumbnail_cropped.png" alt="video-thumbnail1462" onclick="switchVideo('Tourist','video6Container', 'video6Preview')">
<img id="Touristvideo7Preview" class="preview" src="static/img/cropped_thumbnails/scene0050_01_thumbnail_cropped.png" alt="video-thumbnail879" onclick="switchVideo('Tourist', 'video7Container', 'video7Preview')">
<img id="Touristvideo8Preview" class="preview" src="static/img/cropped_thumbnails/09c1414f1b_thumbnail_cropped.png" alt="video-thumbnail2871" onclick="switchVideo('Tourist','video8Container', 'video8Preview')">
<img id="Touristvideo9Preview" class="preview" src="static/img/cropped_thumbnails/42897538_thumbnail_cropped.png" alt="video-thumbnail608" onclick="switchVideo('Tourist','video9Container', 'video9Preview')">
<img id="Touristvideo10Preview" class="preview" src="static/img/cropped_thumbnails/47331990_thumbnail_cropped.png" alt="video-thumbnail4983" onclick="switchVideo('Tourist','video10Container', 'video10Preview')">
</div>
<!-- Video 1; ID: 1129 -->
<div id="Touristvideo1Container" class="video-container">
<div class="video-label">Relative Direction</div>
<video class="video-music" controls preload="metadata" playsinline>
<source src="static/videos/42446049.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
<div class="speed-reminder">
<i class="fas fa-info-circle"></i>This video is playing at 2x speed. You can adjust the speed using the controls above.
</div>
<div class="video-qa">
<div class="qa-item">
<p><strong>Question:</strong> If I am standing by the refrigerator and facing the washer, is the stove to my left, right, or back?
An object is to my back if I would have to turn at least 135 degrees in order to face it.
</p>
<p><strong>Options:</strong></p>
<ul style="list-style-type: none; padding-left: 0; display: flex; justify-content: space-evenly; text-align: left; max-width: 60%; margin: 0 auto;">
<li><input type="radio" name="tables"> A. Back</li>
<li><input type="radio" name="tables"> B. Right</li>
<li><input type="radio" name="tables"> C. Left</li>
</ul>
<div id="answer-1" style="display:none; text-align: center; margin-top: 20px;">
<div style="display: flex; justify-content: center; align-items: center; gap: 20px; white-space: nowrap;">
<p>
<strong style="color: green;">Ground Truth:</strong> Right
</p>
<p>
<strong style="color: red;">Gemini-1.5 Pro:</strong> Left
</p>
</div>
</div>
<p class="click-hint" style="width:85%; cursor:pointer; margin-top: 20px;" onclick="toggleAnswer('answer-1', this)">
<img src="static/img/icons/teaser.gif" style="width:1.5rem">
<strong>Click to view Ground Truth and MLLM's answer!</strong>
</p>
</div>
</div>
</div>
<!-- Video 2; ID: 4453 -->
<div id="Touristvideo2Container" class="video-container" style="display:none;">
<div class="video-label">Object Count</div>
<video class="video-music" controls preload="metadata" playsinline>
<source src="static/videos/scene0663_00.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
<div class="speed-reminder">
<i class="fas fa-info-circle"></i>This video is playing at 2x speed. You can adjust the speed using the controls above.
</div>
<div class="video-qa">
<div class="qa-item">
<p><strong>Question:</strong> How many chair(s) are in this room?</p>
<div id="answer-2" style="display:none; text-align: center; margin-top: 20px;">
<div style="display: flex; justify-content: center; align-items: center; gap: 20px; white-space: nowrap;">
<p>
<strong style="color: green;">Ground Truth:</strong> 3
</p>
<p>
<strong style="color: red;">Gemini-1.5 Pro:</strong> 2
</p>
</div>
</div>
<p class="click-hint" style="width:85%; cursor:pointer; margin-top: 20px;" onclick="toggleAnswer('answer-2', this)">
<img src="static/img/icons/teaser.gif" style="width:1.5rem">
<strong>Click to view Ground Truth and MLLM's answer!</strong>
</p>
</div>
</div>
</div>
<!-- Video 3; ID: 91 -->
<div id="Touristvideo3Container" class="video-container" style="display:none;">
<div class="video-label">Object Count</div>
<video class="video-music" controls preload="metadata" playsinline>
<source src="static/videos/45662987.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
<div class="speed-reminder">
<i class="fas fa-info-circle"></i>This video is playing at 2x speed. You can adjust the speed using the controls above.
</div>
<div class="video-qa">
<div class="qa-item">
<p><strong>Question:</strong> How many sofa(s) are in this room?</p>
<div id="answer-3" style="display:none; text-align: center; margin-top: 20px;">
<div style="display: flex; justify-content: center; align-items: center; gap: 20px; white-space: nowrap;">
<p>
<strong style="color: green;">Ground Truth:</strong> 2
</p>
<p>
<strong style="color: red;">Gemini-1.5 Pro:</strong> 1
</p>
</div>
</div>
<p class="click-hint" style="width:85%; cursor:pointer; margin-top: 20px;" onclick="toggleAnswer('answer-3', this)">
<img src="static/img/icons/teaser.gif" style="width:1.5rem">
<strong>Click to view Ground Truth and MLLM's answer!</strong>
</p>
</div>
</div>
</div>
<!-- Video 4; ID: 212 -->
<div id="Touristvideo4Container" class="video-container" style="display:none;">
<div class="video-label">Object Size</div>
<video class="video-music" controls preload="metadata" playsinline>
<source src="static/videos/42445026.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
<div class="speed-reminder">
<i class="fas fa-info-circle"></i>This video is playing at 2x speed. You can adjust the speed using the controls above.
</div>
<div class="video-qa">
<div class="qa-item">
<p><strong>Question:</strong> What is the length of the longest dimension (length, width, or height) of the chair, measured in centimeters?</p>
<div id="answer-4" style="display:none; text-align: center;">
<div style="display: flex; justify-content: center; align-items: center; gap: 20px; white-space: nowrap;">
<p>
<strong style="color: green;">Ground Truth:</strong> 69
</p>
<p>
<strong style="color: red;">Gemini-1.5 Pro:</strong> 200
</p>
</div>
</div>
<div style="margin-top: 20px; text-align: center;">
<p class="click-hint" style="width:85%; cursor:pointer;" onclick="toggleAnswer('answer-4', this)">
<img src="static/img/icons/teaser.gif" style="width:1.5rem">
<strong>Click to view Ground Truth and MLLM's answer!</strong>
</p>
</div>
</div>
</div>
</div>
<!-- Video 5; ID: 1236 -->
<div id="Touristvideo5Container" class="video-container" style="display:none;">
<div class="video-label">Relative Direction</div>
<video class="video-music" controls preload="metadata" playsinline>
<source src="static/videos/41069025.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
<div class="speed-reminder">
<i class="fas fa-info-circle"></i>This video is playing at 2x speed. You can adjust the speed using the controls above.
</div>
<div class="video-qa">
<div class="qa-item">
<p><strong>Question:</strong> If I am standing by the stove and facing the sofa, is the TV to the left or the right of the sofa?</p>
<p><strong>Options:</strong></p>
<ul style="list-style-type: none; padding-left: 0; display: flex; justify-content: center; text-align: left; max-width: 60%; margin: 0 auto;">
<li style="margin-right: 30px;"><input type="radio" name="tables"> A. Left</li>
<li><input type="radio" name="tables"> B. Right</li>
</ul>
<div id="answer-5" style="display:none; text-align: center; margin-top: 20px;">
<div style="display: flex; justify-content: center; align-items: center; gap: 20px; white-space: nowrap;">
<p>
<strong style="color: green;">Ground Truth:</strong> Left
</p>
<p>
<strong style="color: red;">Gemini-1.5 Pro:</strong> Right
</p>
</div>
</div>
<p class="click-hint" style="width:85%; cursor:pointer; margin-top: 20px;" onclick="toggleAnswer('answer-5', this)">
<img src="static/img/icons/teaser.gif" style="width:1.5rem">
<strong>Click to view Ground Truth and MLLM's answer!</strong>
</p>
</div>
</div>
</div>
<!-- Video 6; ID: 1462 -->
<div id="Touristvideo6Container" class="video-container" style="display:none;">
<div class="video-label">Relative Distance</div>
<video class="video-music" controls preload="metadata" playsinline>
<source src="static/videos/47334096.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
<div class="speed-reminder">
<i class="fas fa-info-circle"></i>This video is playing at 2x speed. You can adjust the speed using the controls above.
</div>
<div class="video-qa">
<div class="qa-item">
<p><strong>Question:</strong> Measuring from the closest point of each object, which of these objects (table, stool, sofa, stove) is the closest to the TV?</p>
<p><strong>Options:</strong></p>
<ul style="list-style-type: none; padding-left: 0; display: flex; justify-content: space-between; text-align: left; max-width: 60%; margin: 0 auto;">
<li><input type="radio" name="tables"> A. Table</li>
<li><input type="radio" name="tables"> B. Stool</li>
<li><input type="radio" name="tables"> C. Sofa</li>
<li><input type="radio" name="tables"> D. Stove</li>
</ul>
<div id="answer-6" style="display:none; text-align: center; margin-top: 20px;">
<div style="display: flex; justify-content: center; align-items: center; gap: 20px; white-space: nowrap;">
<p>
<strong style="color: green;">Ground Truth:</strong> Table
</p>
<p>
<strong style="color: red;">Gemini-1.5 Pro:</strong> Sofa
</p>
</div>
</div>
<p class="click-hint" style="width:85%; cursor:pointer; margin-top: 20px;" onclick="toggleAnswer('answer-6', this)">
<img src="static/img/icons/teaser.gif" style="width:1.5rem">
<strong>Click to view Ground Truth and MLLM's answer!</strong>
</p>
</div>
</div>
</div>
<!-- Video 7; ID: 4870 -->
<div id="Touristvideo7Container" class="video-container" style="display:none;">
<div class="video-label">Absolute Distance</div>
<video class="video-music" controls preload="metadata" playsinline>
<source src="static/videos/scene0050_01.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
<div class="speed-reminder">
<i class="fas fa-info-circle"></i>This video is playing at 2x speed. You can adjust the speed using the controls above.
</div>
<div class="video-qa">
<div class="qa-item">
<p><strong>Question:</strong> Measuring from the closest point of each object, what is the distance between the table and the piano (in meters)?</p>
<div id="answer-7" style="display:none; text-align: center; margin-top: 20px;">
<div style="display: flex; justify-content: center; align-items: center; gap: 20px; white-space: nowrap;">
<p>
<strong style="color: green;">Ground Truth:</strong> 2.3
</p>
<p>
<strong style="color: red;">Gemini-1.5 Pro:</strong> 0.1
</p>
</div>
</div>
<p class="click-hint" style="width:85%; cursor:pointer; margin-top: 20px;" onclick="toggleAnswer('answer-7', this)">
<img src="static/img/icons/teaser.gif" style="width:1.5rem">
<strong>Click to view Ground Truth and MLLM's answer!</strong>
</p>
</div>
</div>
</div>
<!-- Video 8; ID: 2871 -->
<div id="Touristvideo8Container" class="video-container" style="display:none;">
<div class="video-label">Appearance Order</div>
<video class="video-music" controls preload="metadata" playsinline>
<source src="static/videos/09c1414f1b.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
<div class="speed-reminder">
<i class="fas fa-info-circle"></i>This video is playing at 2x speed. You can adjust the speed using the controls above.
</div>
<div class="video-qa">
<div class="qa-item">
<p><strong>Question:</strong> What will be the first-time appearance order of the following categories in the video: blanket, trash can, microwave, plant?</p>
<p><strong>Options:</strong></p>
<div style="display: flex; justify-content: center; flex-wrap: wrap; max-width: 800px; margin: 0 auto; gap: 20px;">
<div style="flex: 1; min-width: 300px; text-align: left;">
<input type="radio" id="optionA" name="appearanceOrder">
<label for="optionA">A. microwave, blanket, plant, trash can</label>
</div>
<div style="flex: 1; min-width: 300px; text-align: left;">
<input type="radio" id="optionB" name="appearanceOrder">
<label for="optionB">B. plant, blanket, microwave, trash can</label>
</div>
<div style="flex: 1; min-width: 300px; text-align: left;">
<input type="radio" id="optionC" name="appearanceOrder">
<label for="optionC">C. plant, blanket, trash can, microwave</label>
</div>
<div style="flex: 1; min-width: 300px; text-align: left;">
<input type="radio" id="optionD" name="appearanceOrder">
<label for="optionD">D. blanket, trash can, microwave, plant</label>
</div>
</div>
</div>
</div>
<div id="answer-8" style="display:none; text-align: center; margin-top: 20px;">
<div style="display: flex; justify-content: center; align-items: center; gap: 20px; white-space: nowrap;">
<p>
<strong style="color: green;">Ground Truth:</strong> C.
</p>
<p>
<strong style="color: red;">Gemini:</strong> B.
</p>
</div>
</div>
<p class="click-hint" style="width:85%; cursor:pointer; margin-top: 20px;" onclick="toggleAnswer('answer-8', this)">
<img src="static/img/icons/teaser.gif" style="width:1.5rem">
<strong>Click to view Ground Truth and MLLM's answer!</strong>
</p>
</div>
</div>
</div>
<!-- Video 9; ID: 567 -->
<div id="Touristvideo9Container" class="video-container" style="display:none;">
<div class="video-label">Room Size</div>
<video class="video-music" controls preload="metadata" playsinline>
<source src="static/videos/42897538.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
<div class="speed-reminder">
<i class="fas fa-info-circle"></i>This video is playing at 2x speed. You can adjust the speed using the controls above.
</div>
<div class="video-qa">
<div class="qa-item">
<p><strong>Question:</strong> What is the size of this room (in square meters)? If multiple rooms are shown, estimate the size of the combined space.</p>
<div id="answer-9" style="display:none; text-align: center; margin-top: 20px;">
<div style="display: flex; justify-content: center; align-items: center; gap: 20px; white-space: nowrap;">
<p><strong style="color: green;">Ground Truth:</strong> 29.0</p>
<p><strong style="color: red;">Gemini-1.5 Pro:</strong> 50</p>
</div>
</div>
<p class="click-hint" style="width:85%; cursor:pointer; margin-top: 20px;" onclick="toggleAnswer('answer-9', this)">
<img src="static/img/icons/teaser.gif" style="width:1.5rem">
<strong>Click to view Ground Truth and MLLM's answer!</strong>
</p>
</div>
</div>
</div>
<!-- Video 10; ID: 4983 -->
<div id="Touristvideo10Container" class="video-container" style="display:none;">
<div class="video-label">Route Plan</div>
<video class="video-music" controls preload="metadata" playsinline>
<source src="static/videos/47331990.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
<div class="speed-reminder" style="margin-bottom: -10px;">
<i class="fas fa-info-circle"></i>This video is playing at 2x speed. You can adjust the speed using the controls above.
</div>
<div class="video-qa" style="margin-top: 0px;">
<div class="qa-item">
<p><strong>Question:</strong> You are a robot beginning at the door and facing the bed. You want to navigate to the window. You will perform the following actions (Note: for each [please fill in], choose either 'turn back,' 'turn left,' or 'turn right.'): 1. Go forward until the edge of bed 2. [please fill in] 3. Go forward until the wall 4. [please fill in] 5. Go forward until the window. You have reached the final destination.</p>
<p><strong>Options:</strong></p>
<div style="display: flex; justify-content: center; flex-wrap: wrap; max-width: 800px; margin: 0 auto; gap: 20px;">
<div style="flex: 1; min-width: 300px; text-align: left;">
<input type="radio" id="optionA" name="appearanceOrder">
<label for="optionA">A. Turn Back, Turn Left</label>
</div>
<div style="flex: 1; min-width: 300px; text-align: left;">
<input type="radio" id="optionB" name="appearanceOrder">
<label for="optionB">B. Turn Left, Turn Right</label>
</div>
<div style="flex: 1; min-width: 300px; text-align: left;">
<input type="radio" id="optionC" name="appearanceOrder">
<label for="optionC">C. Turn Right, Turn Left</label>
</div>
<div style="flex: 1; min-width: 300px; text-align: left;">
<input type="radio" id="optionD" name="appearanceOrder">
<label for="optionD">D. Turn Left, Turn Left</label>
</div>
</div>
<div id="answer-10" style="visibility: hidden; opacity: 0; max-height: 0; transition: visibility 0s, opacity 0.5s ease-in-out, max-height 0.5s;">
<div style="display: flex; justify-content: center; align-items: center; gap: 20px; white-space: nowrap;">
<p><strong style="color: green;">Ground Truth:</strong> B</p>
<p><strong style="color: red;">Gemini-1.5 Pro:</strong> C</p>
</div>
</div>
<p class="click-hint" style="width:85%; cursor:pointer; margin-top: 0px;" onclick="toggleAnswer2('answer-10', this)">
<img src="static/img/icons/teaser.gif" style="width:1.5rem">
<strong>Click to view Ground Truth and MLLM's answer!</strong>
</p>
</div>
</div>
</div>
<script>
function toggleAnswer(answerId, element) {
const answerElement = document.getElementById(answerId);
const isVisible = answerElement.style.display === "block";
if (isVisible) {
answerElement.style.display = "none";
element.innerHTML = '<img src="static/img/icons/teaser.gif" style="width:1.5rem"> <strong>Click to view Ground Truth and MLLM\'s answer!</strong>';
} else {
answerElement.style.display = "block";
element.innerHTML = '<img src="static/img/icons/teaser.gif" style="width:1.5rem"> <strong>Hide answer</strong>';
if (window.innerWidth <= 768) {
// answerElement.scrollIntoView({ behavior: "smooth", block: "start" });
}
}
}
</script>
<script>
function toggleAnswer2(answerId, element) {
const answerElement = document.getElementById(answerId);
if (answerElement.style.visibility === "hidden" || answerElement.style.visibility === "") {
answerElement.style.transition = "none";
answerElement.style.visibility = "visible";
answerElement.style.opacity = "1";
answerElement.style.maxHeight = "500px";
element.innerHTML = '<img src="static/img/icons/teaser.gif" style="width:1.5rem"> <strong>Hide answer</strong>';
} else {
// Hide the answer instantly
answerElement.style.transition = "none";
answerElement.style.opacity = "0";
answerElement.style.maxHeight = "0";
answerElement.style.visibility = "hidden";
element.innerHTML = '<img src="static/img/icons/teaser.gif" style="width:1.5rem"> <strong>Click to view Ground Truth and MLLM\'s answer!</strong>';
}
}
</script>
</div>
<d-figure id="fig-teaser">
<figure>
<img data-zoomable="" draggable="false" src="static/img/teaser.png" alt="Visual-Spatial Intelligence Teaser">
<figcaption>
<strong>Figure 1:</strong> Can Multimodal LLMs “think spatially” when presented with a video recording of an environment? Can they build an accurate, implicit “cognitive map” that allows them to answer questions about a space? What are the strengths and limitations of using MLLMs to enhance spatial intelligence? We dig into these questions by setting up video data for MLLMs to watch, building a VQA benchmark to check their recall, and examining what the MLLMs actually remember and understand.
</figcaption>
</figure>
</d-figure>
<p class="text abstract">
We present a novel video-based visual-spatial intelligence benchmark (VSI-Bench) of over 5,000 question-answer pairs, and find that MLLMs exhibit competitive - though subhuman - visual-spatial intelligence.
Our evaluation reveals that MLLMs exhibit competitive visual-spatial intelligence, if still well short of human-level.
To understand the MLLMs' behavior, we probe models to express how they think in space both linguistically and visually and find that while spatial reasoning capabilities remain the primary bottleneck for MLLMs to reach higher benchmark performance, local world models and spatial awareness do emerge within these models.
<br>
</p>
<div class="icon-row">
<a href="#vsi-benchmark" class="icon-link">
<img src="./static/img/icons/bench.svg" alt="Visual Representation Logo" class="icon">
VSI-Bench<br>Construction
</a>
<a href="#vsi-evaluation" class="icon-link">
<img src="./static/img/icons/evaluation.svg" alt="Evaluation Logo" class="icon">
Evaluation<br>Protocol
</a>
<a href="#vsi-leaderboard" class="icon-link">
<img src="./static/img/leaderboard-star-svgrepo-com.svg" alt="Leaderboard Logo" class="icon">
VSI-Bench<br>Leaderboard
</a>
<a href="#visual-representation" class="icon-link">
<img src="./static/img/icons/speech.svg" alt="Speech Logo" class="icon">
Linguistic<br>Analysis
</a>
<a href="#vsi-improvements" class="icon-link">
<img src="./static/img/icons/vision.svg" alt="Vision Logo" class="icon">
Visual<br>Analysis
</a>
</div>
<p class="click-hint2" style="width: 85%;">
<img src="static/img/icons/click.gif" style="width: 1.5rem">
<strong>Click to jump to each section.</strong>
</p>
<hr>
<div id='vsi-benchmark' class="vsi-benchmark">
<div id="sec:vsi-overview" class="sub-section">
<h1 class="text">VSI-Bench</h1>
<p class="text" align="justify">
<p class="text" align="justify">
<strong>Benchmark Overview:</strong> We develop VSI-Bench, a benchmark to evaluate the visual-spatial intelligence of Multimodal LLMs (MLLMs) using over 5,000 question-answer pairs derived from 288 egocentric videos sourced from the validation sets of public indoor 3D scene reconstruction datasets ScanNet, ScanNet++, and ARKitScenes. VSI-Bench includes eight tasks under three task types: configurational, measurement estimation, and spatiotemporal. See Fig. 2 for an overview of the tasks in VSI-Bench and Fig. 3 for dataset statistics. Iteratively refined for quality, VSI-Bench provides a foundation to study the connection between MLLMs and 3D reconstruction.</p>
<d-figure id="fig-task-demo" >
<figure>
<img data-zoomable="" draggable="false" src="static/img/task-demo.png" alt="benchmark category">
<figcaption style="text-align: center; margin-top: 20px;">
<strong>Figure 2: Tasks demonstration of VSI-Bench.</strong>
Note: the questions above are simplified slightly for clarity and brevity.
</figcaption>
</figure>
</d-figure>
<d-figure id="fig-bench-stats">
<figure>
<!-- <div style="display: flex; align-items: center; gap: 0px; margin-left: -250px">
<div style="flex: 3; display: flex; justify-content: center; align-items: center; overflow: hidden; position: relative; margin: 0;">
<iframe
src="static/img/dataset_overview_donut_chart.html"
style="width: 100%; height: 450px; border: none; transform: scale(0.7); transform-origin: center; padding: 0; margin: 0;"
title="Dataset Overview Donut Chart">
</iframe>
</div>
-->
<!-- Display the image -->
<!-- <div style="flex: 1; height: 500px; margin-left: -300px;">
<img
data-zoomable=""
draggable="false"
src="static/img/video-length.png"
alt="Fit Visualization"
style="width: 100%; height: 100%; object-fit: contain; margin: 0px; padding: 0;">
</div>
</div> -->
<img data-zoomable="" draggable="false" src="static/img/benchmark-stats.png" alt="benchmark category" style="width: 90%; height: auto; display: block; margin: 0 auto;">
<figcaption style="text-align: center; margin-top: 20px;">
<strong>Figure 3: Benchmark Statistics.</strong>
<strong>Left</strong>: The distribution of tasks across three main categories. <strong>Right</strong>: The video length statistic.
</figcaption>
</figure>
</d-figure>
</div>
<div id="vsi-construct" class="sub-section">
<p class="text" align="justify"><strong>VSI-Bench Construction:</strong>
We develop a robust pipeline to construct VSI-Bench that enables high-quality question-answer (QA) pair generation at scale. Starting with data collection and unification, we standardize diverse 3D indoor scene datasets into a unified meta-information format, incorporating object categories, bounding boxes, and video specifications to support dataset-agnostic QA generation. QA pairs are generated using automated annotations from meta-information and task-specific question templates, with route planning tasks manually annotated. To ensure quality, we implement a human-in-the-loop review process, iteratively refining question templates, annotations, and QA generation rules by addressing ambiguities and errors flagged by evaluators. </p>
<d-figure id="fig-bench-pipeline" >
<figure>
<img data-zoomable="" draggable="false" src="static/img/benchmark_construction.png" alt="benchmark category">
<figcaption>
<p align="justify"><strong>Figure 4: Benchmark curation pipeline.</strong> The pipeline unifies datasets into a standardized format and semantic space for consistent processing. QA pairs are then generated through both human annotation and question templates. To ensure quality, human verification is implemented at all key stages for filtering low-quality videos, annotations, and ambiguous QA pairs.</p>
</figcaption>
</figure>
</d-figure>
</div>
<div id="vsi-evaluation" class="vsi-evaluation">
<h1 class="text">Evaluation on VSI-Bench</h1>
<!-- <p class="text">
<p class="text">
We benchmarked 15 video-supporting MLLMs from diverse model families. For proprietary models, we consider Gemini-1.5 and GPT-4o. For open-source models, we evaluate models from InternVL2, ViLA, 203 LongViLA, LongVA, LLaVA-OneVision, and LLaVA-NeXT-Video. All evaluations were conducted in zero-shot settings with default prompts and greedy decoding for reproducibility. Tasks were evaluated using either Multiple-Choice Answer (MCA) accuracy or our proposed Mean Relative Accuracy (MRA) for Numerical Answer (NA) tasks. Baselines include random selection and frequency-based answer selection to identify performance gains due to distribution biases. Additionally, human performance was assessed on a randomly sampled subset of 400 questions (VSI-Bench tiny), with metrics compared to Gemini-1.5 Pro.
</p> -->
<script type="text/javascript" id="MathJax-script" async
src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-mml-chtml.js">
</script>
<p class="text" align="justify"><strong>Evaluation Setup</strong>:
We benchmark 15 video-supporting MLLMs from diverse model families. For proprietary models, we consider Gemini-1.5 and GPT-4o. For open-source models, we evaluate models from InternVL2, ViLA, LongViLA, LongVA, LLaVA-OneVision, and LLaVA-NeXT-Video. All evaluations are conducted in zero-shot settings with default prompts and greedy decoding for reproducibility. Tasks are evaluated using either Multiple-Choice Answer (MCA) accuracy or our proposed Mean Relative Accuracy (MRA) for Numerical Answer (NA) tasks.
</p>
<div class="formula">