-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathindex.html
More file actions
751 lines (713 loc) · 78.8 KB
/
Copy pathindex.html
File metadata and controls
751 lines (713 loc) · 78.8 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
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8" />
<title>Chart2Code</title>
<meta name="description" content="From Charts to Code: A Hierarchical Benchmark for Multimodal Models" />
<meta http-equiv="X-UA-Compatible" content="IE=edge,chrome=1" />
<meta name="viewport" content="width=device-width, initial-scale=1, maximum-scale=1, user-scalable=no" />
<link rel="icon" href="https://github.com/CSU-JPG/Chart2Code.github.io/assets/logo_2.png" type="image/png">
<link rel="shortcut icon" href="favicon_mm.ico" type="image/x-icon" />
<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=Lobster&display=swap" rel="stylesheet">
<link href="https://fonts.googleapis.com/css?family=Google+Sans|Noto+Sans|Castoro" rel="stylesheet">
<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=Nunito:ital@0;1&family=Pacifico&family=Poppins:wght@500&display=swap" rel="stylesheet">
<link href="https://fonts.googleapis.com/css2?family=Playfair+Display:ital,wght@0,400..900;1,400..900&family=Raleway:wght@400;500&family=Lora:ital@0;1&display=swap" rel="stylesheet">
<link rel="stylesheet" href="https://cdnjs.cloudflare.com/ajax/libs/font-awesome/6.5.2/css/all.min.css" integrity="sha512-SnH5WK+bZxgPHs44uWIX+LLJAJ9/2PkPKZ5QiAj6Ta86w+fsb2TkcmfRyVX3pBnMFcV7oQPJkl9QevSCWr3W6A==" crossorigin="anonymous" referrerpolicy="no-referrer" />
<link rel="stylesheet" href="css/normalize.css" />
<link rel="stylesheet" href="css/fonts.css" />
<link rel="stylesheet" href="css/styles.css" />
<link rel="stylesheet" href="css/leaderboard.css" />
<script async src="https://www.googletagmanager.com/gtag/js?id=G-H9XFCMDPNS"></script>
<script>
window.dataLayer = window.dataLayer || [];
function gtag() { dataLayer.push(arguments); }
gtag("js", new Date());
gtag("config", "G-H9XFCMDPNS");
</script>
</head>
<body>
<div class="page-container">
<section class="title-section">
<div class="content-wrapper title-wrapper">
<div class="title-header">
<h1 class="chart2code-title">
<span class="word-chart">Chart</span><span class="word-2">2</span><span class="word-code">Code</span>
</h1>
</div>
<h2 class="subtitle">From Charts to Code: A Hierarchical Benchmark for Multimodal Models</h2>
<div class="authors">
<span class="author-block"><a href="#">Jiahao Tang</a><sup>1</sup>,</span>
<span class="author-block"><a href="https://zhaohengyuan1.github.io/">Henry Hengyuan Zhao</a><sup>2</sup>,</span>
<span class="author-block"><a href="#">Lijian Wu</a><sup>1</sup>,</span>
<span class="author-block"><a href="#">Yifei Tao</a><sup>3</sup>,</span>
<span class="author-block"><a href="https://scholar.google.com/citations?user=RLVSYY0AAAAJ&hl=en">Dongxing Mao</a><sup>1</sup>,</span><br>
<span class="author-block"><a href="#">Yang Wan</a><sup>1</sup>,</span>
<span class="author-block"><a href="https://scholar.google.com/citations?user=l18d7kcAAAAJ&hl=en">Jingru Tan</a><sup>1</sup>,</span>
<span class="author-block"><a href="https://minzeng1990.github.io/">Min Zeng</a><sup>1</sup>,</span>
<span class="author-block"><a href="https://scholar.google.com.hk/citations?user=w47WJE4AAAAJ&hl=en">Min Li</a><sup>1</sup>,</span>
<span class="author-block"><a href="https://fingerrec.github.io/">Alex Jinpeng Wang</a><sup>1</sup></span>
</div>
<div class="affiliations">
<span class="author-block"><a href="https://github.com/CSU-JPG"><sup>1</sup>CSU-JPG, </a>Central South University,</span>
<span class="author-block"><sup>2</sup>National University of Singapore,</span>
<span class="author-block"><sup>3</sup>Nanyang Technological University,</span>
</div>
<div class="button-group">
<a href="https://arxiv.org/abs/2510.17932"><button class="outline multimodal"><i class="fas fa-file-pdf"></i> Paper </button></a>
<a href="https://github.com/CSU-JPG/Chart2Code"><button class="outline multimodal"><i class="fa-brands fa-github"></i> Code </button></a>
<a href="https://huggingface.co/datasets/CSU-JPG/Chart2Code"><button class="outline multimodal"><i class="fas fa-images"></i> Data </button></a>
</div>
</div>
</section>
<section class="main-container">
<div class="content-wrapper">
<div class="content-box">
<h2 class="text-title">What's new with Chart2Code benchmark</h2>
<p class="text-content">TL;DR: We introduce Chart2Code, a new benchmark designed to evaluate chart generation capabilities of LMMs under progressively challenging conditions. There are five tasks in the Chart2Code benchmark.</p>
<div class="leaderboard-container">
<div class="tabs-container">
<div class="tab-links">
<button class="tab-link active" data-tab="dr-sample">Level 1: Direct Reproduction</button>
<button class="tab-link" data-tab="crd-sample">Level 1: Custom Raw Darta</button>
<button class="tab-link" data-tab="cfd-sample">Level 1: Custom Figure Data</button>
<button class="tab-link" data-tab="level2-sample">Level 2</button>
<button class="tab-link" data-tab="level3-sample">Level 3</button>
</div>
<div class="tab-content">
<div id="dr-sample" class="tab-pane active">
<img src="./assets/illustration_DR.png" alt="Direct Reproduction Example" class="tab-image">
</div>
<div id="crd-sample" class="tab-pane">
<img src="./assets/illustration_CRD.png" alt="Customized Raw Data Example" class="tab-image">
</div>
<div id="cfd-sample" class="tab-pane">
<img src="./assets/illustration_CFD.png" alt="Customized Figure Data Example" class="tab-image">
</div>
<div id="level2-sample" class="tab-pane">
<img src="./assets/illustration_level2.png" alt="Level 2 Example" class="tab-image">
</div>
<div id="level3-sample" class="tab-pane">
<img src="./assets/illustration_level3.png" alt="Level 3 Example" class="tab-image">
</div>
</div>
</div>
</div>
<h2 class="text-title">Overview</h2>
<p class="text-content">Chart2Code is a new benchmark for evaluating the chart understanding and code generation capabilities of large multimodal models. Chart2Code is explicitly designed from a user-driven perspective, capturing diverse real-world scenarios and progressively increasing task difficulty.</p>
<img src="./assets/figure1.png" class="responsive-image" alt="Chart2Code Overview"/>
<p class="text-content">To our knowledge, Chart2Code is the <b>first hierarchical benchmark</b> that reflects practical chart2code usage while systematically scaling task complexity. It consists of three levels as illustrated in above figure:
<br>Level1(<b>Chart Reproduction</b>) reproduces charts from a reference figure and user query;
<br>Level2(<b>Chart Editing</b>) involves complex modifications such as changing chart types or adding elements;
<br>Level3(<b>Long-Table to Chart Generation</b>) requires models to transform long, information-dense tables into faithful charts following user instructions.
</p>
<h2 class="text-title">Data Statistic</h2>
<img src="./assets/figure2.png" class="responsive-image" alt="Data Statistics"/>
<p class="text-content">In total, Chart2Code contains 2,023 tasks across 22 chart types, paired with multi-level evaluation metrics that assess both code correctness and the visual fidelity of rendered charts.</p>
<h2 class="text-title">Benchmark Comparison</h2>
<img src="./assets/figure3.png" class="responsive-image" alt="Benchmark Comparison Table"/>
<p class="text-content text-center">Table 1: Chart2Code is a unique benchmark featuring a more comprehensive set of tasks that better reflect real-world scenarios.</p>
<p class="text-content"><b>Comparison of existing chart-to-code benchmarks.</b>: Chart2Code is a new benchmark designed to rigorously evaluate chart generation capabilities of LMMs under progressively challenging conditions. This hierarchical design reflects real-world usage while progressively increasing difficulty, and its distinctions from prior benchmarks are highlighted in the above table.</p>
<h2 class="text-title">Human-Model Performance Comparison</h2>
<div class="leaderboard-container">
<div class="tabs-container">
<div class="tab-links">
<button class="tab-link active" data-tab="dr-sample">Level 1: Direct Reproduction</button>
<button class="tab-link" data-tab="crd-sample">Level 1: Custom Raw Darta</button>
<button class="tab-link" data-tab="cfd-sample">Level 1: Custom Figure Data</button>
<button class="tab-link" data-tab="level2-sample">Level 2 Example</button>
<button class="tab-link" data-tab="level3-sample">Level 3 Example</button>
</div>
<div class="tab-content">
<div id="dr-sample" class="tab-pane active">
<img src="assets/human_model_comparison_DR.png" alt="Direct Reproduction Example" class="tab-image">
</div>
<div id="crd-sample" class="tab-pane">
<img src="assets/human_model_comparison_CRD.png" alt="Customized Raw Data Example" class="tab-image">
</div>
<div id="cfd-sample" class="tab-pane">
<img src="assets/human_model_comparison_CFD.png" alt="Customized Figure Data Example" class="tab-image">
</div>
<div id="level2-sample" class="tab-pane">
<img src="assets/human_model_comparison_level2.png" alt="Level 2 Example" class="tab-image">
</div>
<div id="level3-sample" class="tab-pane">
<img src="assets/human_model_comparison_level3.png" alt="Level 3 Example" class="tab-image">
</div>
</div>
</div>
</div>
<div class="leaderboard-container">
<div class="tabs-container">
<div class="tab-links">
<button class="tab-link active" data-tab="level1">Level 1: Chart Reproduction</button>
<button class="tab-link" data-tab="level2">Level 2: Chart Editing</button>
<button class="tab-link" data-tab="level3">Level 3: Long-Table to Chart</button>
</div>
<div class="tab-content">
<div id="level1" class="tab-pane active">
<table class="sortable-table">
<thead>
<tr>
<th rowspan="2"><strong>Model</strong></th>
<th colspan="4"><strong>Direct Reproduction(DR)</strong></th>
<th colspan="4"><strong>Customize Raw Data(CRD)</strong></th>
<th colspan="4"><strong>Customize Figure Data(CFD)</strong></th>
</tr>
<tr>
<th><strong>Exec.Rate</strong></th>
<th><strong>Base</strong></th>
<th><strong>LLM</strong></th>
<th><strong>LMM</strong></th>
<th><strong>Exec.Rate</strong></th>
<th><strong>Base</strong></th>
<th><strong>LLM</strong></th>
<th><strong>LMM</strong></th>
<th><strong>Exec.Rate</strong></th>
<th><strong>Base</strong></th>
<th><strong>LLM</strong></th>
<th><strong>LMM</strong></th>
</tr>
</thead>
<tbody>
<tr class="group-header"><td colspan="13"><em><strong>Proprietary</strong></em></td></tr>
<tr><td>Gemini-3-Pro</td><td><strong>97.50</strong></td><td>78.65</td><td>75.86</td><td><strong>45.42</strong></td><td><strong>100.0</strong></td><td>69.23</td><td><strong>69.76</strong></td><td><strong>40.72</strong></td><td><strong>99.07</strong></td><td>70.78</td><td>71.12</td><td>32.85</td></tr>
<tr><td>Claude-Sonnet-4</td><td>96.52</td><td>65.60</td><td>70.69</td><td>32.36</td><td><strong>100.0</strong></td><td>61.46</td><td>55.03</td><td>40.63</td><td>93.52</td><td>65.27</td><td>65.99</td><td>26.44</td></tr>
<tr><td>GPT-5.2</td><td>97.08</td><td><strong>79.91</strong></td><td><strong>77.88</strong></td><td>43.73</td><td>97.22</td><td>66.31</td><td>65.51</td><td>39.26</td><td><strong>99.07</strong></td><td><strong>73.02</strong></td><td><strong>71.42</strong></td><td><strong>35.40</strong></td></tr>
<tr><td>Seed-1.5-VL</td><td>87.34</td><td>63.85</td><td>53.84</td><td>26.40</td><td>97.22</td><td>65.76</td><td>58.73</td><td>34.74</td><td>79.63</td><td>65.58</td><td>64.19</td><td>19.53</td></tr>
<tr><td>Seed-1.6-VL</td><td>82.61</td><td>62.46</td><td>50.85</td><td>26.16</td><td>94.44</td><td>60.69</td><td>57.87</td><td>28.44</td><td>79.63</td><td>66.22</td><td>63.12</td><td>25.51</td></tr>
<tr class="group-header"><td colspan="13"><em><strong>Open-Source LMMs (non-thinking)</strong></em></td></tr>
<tr><td>LLaVA-OV-Qwen2-7B-SI</td><td>82.48</td><td>31.33</td><td>16.81</td><td>2.58</td><td>19.44</td><td>49.73</td><td>35.93</td><td>15.14</td><td>0.00</td><td>-</td><td>-</td><td>-</td></tr>
<tr><td>LLaVA-OV-Qwen2-7B-OV</td><td>72.60</td><td>34.33</td><td>54.85</td><td>3.42</td><td>8.33</td><td>32.68</td><td>15.67</td><td>6.00</td><td>0.00</td><td>-</td><td>-</td><td>-</td></tr>
<tr><td>DeepSeek-VL-7B</td><td>36.44</td><td>37.49</td><td>22.02</td><td>3.37</td><td>55.56</td><td>50.73</td><td>29.28</td><td>9.25</td><td>4.63</td><td>30.37</td><td>25.20</td><td>5.60</td></tr>
<tr><td>kimi-VL-A3B</td><td>70.79</td><td>52.76</td><td>40.36</td><td>14.15</td><td>66.67</td><td>50.29</td><td>46.85</td><td>33.04</td><td>64.81</td><td>53.62</td><td>45.36</td><td>16.41</td></tr>
<tr><td>Qwen2-VL-7B</td><td>62.45</td><td>40.99</td><td>28.74</td><td>7.02</td><td>77.78</td><td>52.21</td><td>44.50</td><td>19.61</td><td>31.48</td><td>51.02</td><td>41.03</td><td>8.65</td></tr>
<tr><td>Qwen2-VL-72B</td><td>74.97</td><td>50.74</td><td>38.27</td><td>12.75</td><td>86.11</td><td>56.02</td><td>46.89</td><td>28.23</td><td>64.81</td><td>59.66</td><td>56.41</td><td>15.81</td></tr>
<tr><td>InternVL-2.5-8B</td><td>67.73</td><td>42.76</td><td>27.49</td><td>7.80</td><td>69.44</td><td>54.16</td><td>42.80</td><td>25.88</td><td>37.04</td><td>51.58</td><td>38.81</td><td>12.85</td></tr>
<tr><td>InternVL-2.5-38B</td><td>85.67</td><td>51.97</td><td>39.83</td><td>15.18</td><td>86.11</td><td>55.74</td><td>47.98</td><td>24.71</td><td>77.78</td><td>59.86</td><td>55.51</td><td>22.88</td></tr>
<tr><td>InternVL-3-8B</td><td>29.49</td><td>46.58</td><td>55.68</td><td>9.20</td><td>36.11</td><td>53.47</td><td>31.38</td><td>6.23</td><td>12.96</td><td>50.73</td><td>35.61</td><td>3.14</td></tr>
<tr><td>InternVL-3-38B</td><td>85.26</td><td>53.57</td><td>42.85</td><td>16.68</td><td>86.11</td><td>58.17</td><td>51.10</td><td>34.00</td><td>36.11</td><td>60.17</td><td>60.42</td><td>18.56</td></tr>
<tr><td>GLM-4V-9B</td><td>54.24</td><td>32.50</td><td>23.12</td><td>4.80</td><td>75.00</td><td>50.35</td><td>42.91</td><td>19.70</td><td>41.67</td><td>31.64</td><td>12.84</td><td>1.00</td></tr>
<tr><td>GLM-4.6v-Flash</td><td>54.52</td><td>66.67</td><td>55.38</td><td>31.70</td><td>61.11</td><td>63.44</td><td>52.75</td><td>32.45</td><td>55.56</td><td>65.64</td><td>60.23</td><td>22.57</td></tr>
<tr><td>Intern-VL-3.5-8B</td><td>72.18</td><td>52.01</td><td>41.29</td><td>15.41</td><td>69.44</td><td>37.87</td><td>14.58</td><td>1.68</td><td>49.07</td><td>53.18</td><td>48.85</td><td>10.87</td></tr>
<tr><td>Intern-VL-3.5-38B</td><td>84.01</td><td>55.78</td><td>42.54</td><td>19.78</td><td>86.11</td><td>60.66</td><td>53.16</td><td>38.83</td><td>24.07</td><td>61.15</td><td>60.69</td><td>21.84</td></tr>
<tr><td>MiMo-VL-7B-RL</td><td>43.12</td><td>64.41</td><td>49.82</td><td>21.43</td><td>66.67</td><td>59.73</td><td>52.21</td><td>25.38</td><td>42.59</td><td>64.39</td><td>57.68</td><td>21.44</td></tr>
<tr><td>MiMo-VL-7B-SFT</td><td>51.74</td><td>63.99</td><td>49.07</td><td>22.61</td><td>75.00</td><td>53.34</td><td>37.80</td><td>22.74</td><td>32.41</td><td>64.68</td><td>54.74</td><td>22.83</td></tr>
<tr><td>Qwen2.5-VL-7B</td><td>69.26</td><td>48.20</td><td>35.46</td><td>10.54</td><td>83.33</td><td>55.42</td><td>46.72</td><td>20.14</td><td>44.44</td><td>50.30</td><td>41.78</td><td>7.81</td></tr>
<tr><td>Qwen2.5-VL-72B</td><td>63.84</td><td>59.05</td><td>47.95</td><td>20.97</td><td>94.44</td><td>58.12</td><td>50.97</td><td>27.53</td><td>47.22</td><td>63.95</td><td>61.37</td><td>24.48</td></tr>
<tr><td>Molmo-7B-D</td><td>4.31</td><td>25.26</td><td>11.48</td><td>2.42</td><td>2.78</td><td><strong>70.30</strong></td><td>40.00</td><td>4.00</td><td>1.85</td><td>45.58</td><td>24.00</td><td>0.00</td></tr>
<tr><td>Qwen3-VL-30B-A3B</td><td>72.18</td><td>63.09</td><td>50.84</td><td>15.03</td><td>69.44</td><td>55.84</td><td>52.26</td><td>23.28</td><td>74.07</td><td>63.02</td><td>60.23</td><td>23.56</td></tr>
<tr><td>Qwen3-VL-32B</td><td>76.91</td><td>63.10</td><td>54.87</td><td>18.44</td><td>77.78</td><td>60.49</td><td>52.64</td><td>36.79</td><td>80.56</td><td>66.64</td><td>65.27</td><td>25.84</td></tr>
<tr class="group-header"><td colspan="13"><em><strong>Open-Source LMMs (thinking)</strong></em></td></tr>
<tr><td>MiMo-VL-7B-RL</td><td>21.14</td><td>70.45</td><td>60.00</td><td>30.04</td><td>75.00</td><td>64.58</td><td>57.96</td><td>24.96</td><td>37.96</td><td>68.05</td><td>64.38</td><td>29.83</td></tr>
<tr><td>MiMo-VL-7B-SFT</td><td>41.03</td><td>65.49</td><td>54.33</td><td>16.95</td><td>77.78</td><td>61.33</td><td>53.89</td><td>29.93</td><td>37.96</td><td>64.04</td><td>56.04</td><td>23.29</td></tr>
<tr><td>Qwen3-VL-30B-A3B</td><td>76.08</td><td>62.37</td><td>55.48</td><td>23.83</td><td>61.11</td><td>64.78</td><td>57.63</td><td>35.00</td><td>80.56</td><td>66.57</td><td>62.86</td><td>26.34</td></tr>
</tbody>
</table>
</div>
<div id="level2" class="tab-pane">
<table class="sortable-table">
<thead>
<tr>
<th rowspan="2"><strong>Model</strong></th>
<th rowspan="2"><strong>Exec.<br>Rate</strong></th>
<th colspan="10"><strong>Code-Level</strong></th>
<th rowspan="2"><strong>Chart-Level<br>LMM-Score</strong></th>
</tr>
<tr>
<th><strong>Color</strong></th>
<th><strong>Grid</strong></th>
<th><strong>Layout</strong></th>
<th><strong>Legend</strong></th>
<th><strong>Visual</strong></th>
<th><strong>Data</strong></th>
<th><strong>Text</strong></th>
<th><strong>Type</strong></th>
<th><strong>Base</strong></th>
<th><strong>LLM</strong></th>
</tr>
</thead>
<tbody>
<tr class="group-header"><td colspan="13"><em><strong>Proprietary</strong></em></td></tr>
<tr><td>Gemini-3-Pro</td><td><strong>97.23</strong></td><td><strong>52.32</strong></td><td>75.31</td><td><strong>86.45</strong></td><td>63.33</td><td>81.58</td><td>62.75</td><td>77.16</td><td>93.86</td><td>70.78</td><td>72.21</td><td><strong>33.41</strong></td></tr>
<tr><td>Claude-Sonnet-4</td><td>90.20</td><td>47.17</td><td>65.29</td><td>55.32</td><td>56.51</td><td>81.50</td><td>54.88</td><td>80.52</td><td>93.29</td><td>63.65</td><td>66.45</td><td>25.40</td></tr>
<tr><td>GPT-5.2</td><td>96.04</td><td>58.44</td><td>80.83</td><td>61.51</td><td>58.16</td><td><strong>84.65</strong></td><td><strong>64.66</strong></td><td><strong>83.77</strong></td><td><strong>94.52</strong></td><td><strong>70.93</strong></td><td><strong>75.66</strong></td><td>33.03</td></tr>
<tr><td>Seed-1.5-VL</td><td>60.20</td><td>44.39</td><td>69.37</td><td>52.06</td><td>52.80</td><td>79.85</td><td>52.02</td><td>77.21</td><td>92.66</td><td>61.67</td><td>65.45</td><td>18.30</td></tr>
<tr><td>Seed-1.6-VL</td><td>70.00</td><td>43.61</td><td>69.37</td><td>52.25</td><td>51.32</td><td>79.70</td><td>53.11</td><td>79.46</td><td>92.32</td><td>61.77</td><td>65.40</td><td>17.43</td></tr>
<tr class="group-header"><td colspan="13"><em><strong>Open-Source LMMs (non-thinking)</strong></em></td></tr>
<tr><td>LLaVA-OV-Qwen2-7B-SI</td><td>2.57</td><td>15.35</td><td>59.38</td><td>53.85</td><td>30.79</td><td>56.34</td><td>22.59</td><td>52.70</td><td>55.33</td><td>38.43</td><td>37.40</td><td>6.40</td></tr>
<tr><td>LLaVA-OV-Qwen2-7B-OV</td><td>1.68</td><td>15.37</td><td>63.48</td><td>38.24</td><td><strong>64.56</strong></td><td>50.15</td><td>18.13</td><td>48.81</td><td>58.43</td><td>39.07</td><td>35.44</td><td>7.35</td></tr>
<tr><td>DeepSeek-VL-7B</td><td>30.46</td><td>15.32</td><td>42.58</td><td>34.44</td><td>26.24</td><td>52.95</td><td>22.36</td><td>51.56</td><td>68.43</td><td>35.16</td><td>26.87</td><td>4.20</td></tr>
<tr><td>kimi-VL-A3B</td><td>50.69</td><td>26.22</td><td>64.50</td><td>40.68</td><td>39.93</td><td>66.08</td><td>34.05</td><td>64.97</td><td>83.03</td><td>47.96</td><td>44.43</td><td>9.01</td></tr>
<tr><td>Qwen2-VL-7B</td><td>24.55</td><td>18.72</td><td>57.60</td><td>41.87</td><td>32.73</td><td>58.65</td><td>25.41</td><td>55.07</td><td>76.38</td><td>41.06</td><td>34.82</td><td>5.27</td></tr>
<tr><td>Qwen2-VL-72B</td><td>57.23</td><td>27.82</td><td>64.07</td><td>42.86</td><td>40.59</td><td>69.53</td><td>36.83</td><td>64.35</td><td>84.76</td><td>49.55</td><td>39.88</td><td>10.02</td></tr>
<tr><td>InternVL-2.5-8B</td><td>24.06</td><td>23.86</td><td>62.25</td><td>41.21</td><td>36.70</td><td>65.03</td><td>32.72</td><td>62.46</td><td>81.14</td><td>46.20</td><td>41.14</td><td>7.48</td></tr>
<tr><td>InternVL-2.5-38B</td><td>30.30</td><td>30.40</td><td>68.25</td><td>50.34</td><td>48.97</td><td>72.92</td><td>37.38</td><td>70.78</td><td>87.96</td><td>53.48</td><td>52.90</td><td>10.67</td></tr>
<tr><td>InternVL-3-8B</td><td>4.55</td><td>23.22</td><td>62.68</td><td>57.03</td><td>31.34</td><td>66.79</td><td>29.34</td><td>58.63</td><td>84.49</td><td>46.41</td><td>38.02</td><td>5.48</td></tr>
<tr><td>InternVL-3-38B</td><td>69.80</td><td>35.87</td><td>68.36</td><td>48.66</td><td>49.30</td><td>75.86</td><td>45.33</td><td>72.28</td><td>90.53</td><td>56.74</td><td>56.98</td><td>13.32</td></tr>
<tr><td>GLM-4V-9B</td><td>10.50</td><td>19.71</td><td>61.79</td><td>48.11</td><td>34.54</td><td>59.98</td><td>26.53</td><td>57.80</td><td>65.80</td><td>42.05</td><td>34.78</td><td>3.87</td></tr>
<tr><td>GLM-4.6V-Flash</td><td>18.22</td><td>48.06</td><td>69.23</td><td>50.13</td><td>51.20</td><td>81.56</td><td>54.78</td><td>77.90</td><td>91.86</td><td>62.76</td><td>63.25</td><td>17.91</td></tr>
<tr><td>InternVL-3.5-8B</td><td>37.52</td><td>29.55</td><td>68.21</td><td>43.87</td><td>41.56</td><td>69.85</td><td>39.58</td><td>67.41</td><td>83.82</td><td>51.30</td><td>49.52</td><td>9.71</td></tr>
<tr><td>InternVL-3.5-38B</td><td>4.26</td><td>44.11</td><td><strong>92.54</strong></td><td>67.22</td><td>60.24</td><td>72.35</td><td>40.38</td><td>76.53</td><td>88.56</td><td>62.64</td><td>57.48</td><td>8.28</td></tr>
<tr><td>MiMo-VL-7B-RL</td><td>17.23</td><td>42.72</td><td>70.76</td><td>54.30</td><td>46.81</td><td>74.50</td><td>49.23</td><td>74.57</td><td>89.75</td><td>59.42</td><td>56.95</td><td>14.54</td></tr>
<tr><td>MiMo-VL-7B-SFT</td><td>30.40</td><td>37.36</td><td>67.37</td><td>48.40</td><td>47.24</td><td>75.98</td><td>46.53</td><td>70.80</td><td>88.04</td><td>56.56</td><td>53.97</td><td>15.23</td></tr>
<tr><td>Qwen2.5-VL-7B</td><td>31.98</td><td>26.87</td><td>67.22</td><td>45.41</td><td>38.75</td><td>65.34</td><td>37.21</td><td>63.68</td><td>83.65</td><td>49.22</td><td>44.35</td><td>9.44</td></tr>
<tr><td>Qwen2.5-VL-72B</td><td>74.55</td><td>41.32</td><td>67.44</td><td>52.16</td><td>50.04</td><td>77.74</td><td>48.65</td><td>74.72</td><td>91.12</td><td>59.31</td><td>49.02</td><td>16.69</td></tr>
<tr><td>Molmo-7B-D</td><td>0.59</td><td>16.53</td><td>44.44</td><td>41.67</td><td>50.00</td><td>49.70</td><td>29.14</td><td>41.57</td><td>54.44</td><td>37.32</td><td>29.92</td><td>15.50</td></tr>
<tr><td>Qwen3-VL-30B-A3B</td><td>38.91</td><td>40.07</td><td>70.02</td><td>51.69</td><td>50.06</td><td>76.81</td><td>48.01</td><td>75.52</td><td>92.39</td><td>59.26</td><td>45.77</td><td>17.46</td></tr>
<tr><td>Qwen3-32B</td><td>63.76</td><td>42.90</td><td>71.98</td><td>50.76</td><td>52.24</td><td>77.96</td><td>52.35</td><td>77.43</td><td>92.09</td><td>61.30</td><td>64.15</td><td>18.27</td></tr>
<tr class="group-header"><td colspan="13"><em><strong>Open-Source LMMs (thinking)</strong></em></td></tr>
<tr><td>MiMo-VL-7B-RL</td><td>26.14</td><td>42.47</td><td>69.05</td><td>48.47</td><td>45.96</td><td>80.76</td><td>50.41</td><td>74.80</td><td>90.56</td><td>59.53</td><td>61.04</td><td>17.81</td></tr>
<tr><td>MiMo-VL-7B-SFT</td><td>30.50</td><td>40.56</td><td>68.22</td><td>48.10</td><td>46.65</td><td>78.42</td><td>48.68</td><td>73.48</td><td>89.86</td><td>58.32</td><td>59.33</td><td>16.38</td></tr>
<tr><td>Qwen3-VL-30B-A3B</td><td>13.76</td><td>44.00</td><td>65.02</td><td>48.92</td><td>49.19</td><td>78.19</td><td>50.18</td><td>76.18</td><td>89.60</td><td>59.55</td><td>63.04</td><td>14.37</td></tr>
</tbody>
</table>
</div>
<div id="level3" class="tab-pane">
<table class="sortable-table">
<thead>
<tr>
<th rowspan="2"><strong>Model</strong></th>
<th rowspan="2"><strong>Exec.<br>Rate</strong></th>
<th colspan="10"><strong>Code-Level</strong></th>
<th rowspan="2"><strong>Chart-Level<br>LMM-Score</strong></th>
</tr>
<tr>
<th><strong>Color</strong></th>
<th><strong>Grid</strong></th>
<th><strong>Layout</strong></th>
<th><strong>Legend</strong></th>
<th><strong>Visual</strong></th>
<th><strong>Data</strong></th>
<th><strong>Text</strong></th>
<th><strong>Type</strong></th>
<th><strong>Base</strong></th>
<th><strong>LLM-Score</strong></th>
</tr>
</thead>
<tbody>
<tr class="group-header"><td colspan="13"><em><strong>Proprietary</strong></em></td></tr>
<tr><td>Gemini-3-Pro</td><td>30.03</td><td><strong>56.29</strong></td><td><strong>83.71</strong></td><td>92.55</td><td>68.39</td><td><strong>84.00</strong></td><td><strong>62.26</strong></td><td><strong>80.45</strong></td><td><strong>96.60</strong></td><td><strong>74.28</strong></td><td><strong>77.90</strong></td><td><strong>35.97</strong></td></tr>
<tr><td>Claude-Sonnet-4</td><td><strong>46.65</strong></td><td>35.43</td><td>68.97</td><td>84.08</td><td>61.77</td><td>74.44</td><td>46.73</td><td>67.77</td><td>89.95</td><td>61.13</td><td>62.04</td><td>16.60</td></tr>
<tr><td>GPT-5.2</td><td>20.13</td><td>34.46</td><td>82.14</td><td>80.87</td><td>52.64</td><td>78.34</td><td>43.56</td><td>76.09</td><td>93.81</td><td>61.99</td><td>68.40</td><td>16.29</td></tr>
<tr><td>Seed-1.5-VL</td><td>10.86</td><td>44.55</td><td>76.89</td><td>93.94</td><td>69.30</td><td>73.52</td><td>51.75</td><td>73.57</td><td>95.96</td><td>67.58</td><td>75.59</td><td>22.85</td></tr>
<tr><td>Seed-1.6-VL</td><td>26.52</td><td>44.26</td><td>82.28</td><td><strong>95.48</strong></td><td><strong>74.73</strong></td><td>75.42</td><td>52.62</td><td>77.82</td><td>86.47</td><td>68.6</td><td>76.35</td><td>25.77</td></tr>
</tbody>
</table>
</div>
</div>
</div>
</div>
<h2 class="text-title">Results Analysis</h2>
<img src="assets/figure7.png" class="responsive-image" alt="Correlation of model performance">
<p class="text-content text-center">Correlation of the model performance (i.e., LMM-score) on different manually annotated difficulties (i.e., Easy, Medium, Hard) on Level 1, 2, 3, respectively.</p>
<img src="assets/figure8.png" class="responsive-image" alt="Model generalization analysis">
<p class="text-content">Left: Both proprietary and open-source models generalize well on Level 1 and Level 2 tasks when calculating the LLM-score for predicted code assessment. Right: Proprietary models tend to obtain higher LMM-scores on the Level 1 task rather than the Level 2, while open-source models perform poorly on both tasks (scores are lower than 0.5).</p>
<img src="assets/figure9.png" class="responsive-image" alt="Performance analysis on different task cases">
<p class="text-content text-center">Analysis of model performance on different task cases with LLM-score and LMM-score.</p>
<h2 class="text-title">Example of Data and Error Cases</h2>
<div class="leaderboard-container">
<div class="tabs-container">
<div class="tab-links">
<button class="tab-link active" data-tab="dr-sample">Level 1: Direct Reproduction</button>
<button class="tab-link" data-tab="crd-sample">Level 1: Custom Raw Darta</button>
<button class="tab-link" data-tab="cfd-sample">Level 1: Custom Figure</button>
<button class="tab-link" data-tab="level2-sample">Level 2 Example</button>
<button class="tab-link" data-tab="level3-sample">Level 3 Example</button>
<button class="tab-link" data-tab="error-cases">Error Cases</button>
</div>
<div class="tab-content">
<div id="dr-sample" class="tab-pane active">
<h3 class="text-title">An Example of Level 1: Direct Reproduction</h3>
<img src="assets/DR_sample.png" alt="Direct Reproduction Example" class="tab-image">
</div>
<div id="crd-sample" class="tab-pane">
<h3 class="text-title">An Example of Level 1: Customized Text-Format Table Data</h3>
<img src="assets/CRD_sample.png" alt="Customized Raw Data Example" class="tab-image">
</div>
<div id="cfd-sample" class="tab-pane">
<h3 class="text-title">An Example of Level 1: Figure-Format Table Data</h3>
<img src="assets/CFD_sample.png" alt="Customized Figure Data Example" class="tab-image">
</div>
<div id="level2-sample" class="tab-pane">
<h3 class="text-title">An Example of Level 2</h3>
<img src="assets/level2_sample.png" alt="Level 2 Example" class="tab-image">
</div>
<div id="level3-sample" class="tab-pane">
<h3 class="text-title">An Example of Level 3</h3>
<img src="assets/level3_sample.png" alt="Level 3 Example" class="tab-image">
</div>
<div id="error-cases" class="tab-pane">
<h3 class="text-title">Error Cases Visualization</h3>
<img src="assets/error_analyse.png" alt="Error Analysis 1" class="tab-image" style="margin-bottom: 20px;">
<img src="assets/error_analyse_2.png" alt="Error Analysis 2" class="tab-image">
</div>
</div>
</div>
</div>
<!-- Evaluation Cases Section -->
<h2 class="text-title">Evaluation Cases</h2>
<div class="leaderboard-container">
<div class="tabs-container">
<!-- 选项卡链接 (保证了3个按钮互不重复) -->
<div class="tab-links">
<button class="tab-link active" data-tab="eval-case-1">Case 1: Radar Chart (Radar 36)</button>
<button class="tab-link" data-tab="eval-case-2">Case 2: Bar Chart (Bar 62)</button>
<button class="tab-link" data-tab="eval-case-3">Case 3: Combination Chart (Combination 45)</button>
</div>
<div class="tab-content" style="background-color: #ffffff; border-radius: 0 0 8px 8px; padding: 20px;">
<!-- ==================== Case 1: Radar Chart ==================== -->
<div id="eval-case-1" class="tab-pane active">
<div style="display: flex; flex-wrap: wrap; gap: 20px; justify-content: center; margin-bottom: 25px;">
<div style="flex: 1 1 45%; min-width: 280px; padding: 15px; background: #fafafa; border: 1px solid #e2e8f0; border-radius: 8px; display: flex; flex-direction: column; align-items: center; box-shadow: 0 2px 4px rgba(0,0,0,0.02); box-sizing: border-box;">
<div style="width: 100%; display: flex; justify-content: center; background: #fff; border: 1px solid #eee; border-radius: 4px; padding: 10px;">
<img src="./assets/radar_36.png" alt="GT Figure" style="max-width: 100%; max-height: 350px; object-fit: contain;">
</div>
<p style="margin: 12px 0 0 0; font-size: 0.95em; color: #334155;"><strong>GT_Figure (Left)</strong></p>
</div>
<div style="flex: 1 1 45%; min-width: 280px; padding: 15px; background: #fafafa; border: 1px solid #e2e8f0; border-radius: 8px; display: flex; flex-direction: column; align-items: center; box-shadow: 0 2px 4px rgba(0,0,0,0.02); box-sizing: border-box;">
<div style="width: 100%; display: flex; justify-content: center; background: #fff; border: 1px solid #eee; border-radius: 4px; padding: 10px;">
<img src="./assets/claude_radar_36.png" alt="Generation Figure" style="max-width: 100%; max-height: 350px; object-fit: contain;">
</div>
<p style="margin: 12px 0 0 0; font-size: 0.95em; color: #334155;"><strong>Generation_Figure (Right)</strong></p>
</div>
</div>
<div style="background-color: #f8fafc; border-left: 4px solid #3b82f6; padding: 12px 16px; margin-bottom: 20px; border-radius: 0 4px 4px 0;">
<p style="margin: 0; font-size: 0.95em; line-height: 1.5; color: #334155;">
<strong>Detailed multi-dimensional evaluation statistics.</strong> By pairing related metrics, we observe comprehensive performance across API, Base, and standard Evaluation reports.
</p>
</div>
<div style="overflow-x: auto; width: 100%; border: 1px solid #e2e8f0; border-radius: 8px;">
<table style="width: 100%; min-width: 700px; border-collapse: collapse; text-align: left; font-size: 0.95em; background-color: #fff;">
<thead>
<tr style="background-color: #f1f5f9; border-bottom: 2px solid #cbd5e1;">
<th style="padding: 12px 16px; width: 20%; color: #1e293b;"><strong>Metric 1 (Score)</strong></th>
<th style="padding: 12px 16px; width: 20%; color: #1e293b;"><strong>Metric 2 (Score)</strong></th>
<th style="padding: 12px 16px; width: 60%; color: #1e293b;"><strong>Combined Reason / Key Insight</strong></th>
</tr>
</thead>
<tbody>
<tr>
<td colspan="3" style="padding: 10px 16px; background-color: #e2e8f0; color: #0f172a; border-bottom: 1px solid #cbd5e1;">
<em><strong>LLM Report (Overall Score: 38.50)</strong></em>
</td>
</tr>
<tr style="border-bottom: 1px solid #f1f5f9;">
<td style="padding: 12px 16px; color: #334155;">Chart Type (0.0)</td>
<td style="padding: 12px 16px; color: #334155;">Data Sim (0.0)</td>
<td style="padding: 12px 16px; color: #475569;">GT is normalized radial radar (radius [0.1,0.9]); GEN plots raw absolute values (0-100). Fundamentally different data mapping.</td>
</tr>
<tr style="border-bottom: 1px solid #f1f5f9;">
<td style="padding: 12px 16px; color: #334155;">Visual Params (30.0)</td>
<td style="padding: 12px 16px; color: #334155;">Color (80.0)</td>
<td style="padding: 12px 16px; color: #475569;">Similar basic colors used, but GEN uses thicker lines, different markers ('o'), and varying radius scaling.</td>
</tr>
<tr style="border-bottom: 1px solid #f1f5f9;">
<td style="padding: 12px 16px; color: #334155;">Layout (50.0)</td>
<td style="padding: 12px 16px; color: #334155;">Grid (40.0)</td>
<td style="padding: 12px 16px; color: #475569;">Figure size, theta offset, and limits differ. GT uses subtle dashed grey grid without y-ticks; GEN shows visible default radial ticks.</td>
</tr>
<tr style="border-bottom: 1px solid #cbd5e1;">
<td style="padding: 12px 16px; color: #334155;">Legend (60.0)</td>
<td style="padding: 12px 16px; color: #334155;">Text Content (45.0)</td>
<td style="padding: 12px 16px; color: #475569;">Legend locations differ (bottom vs. upper-right). Numeric annotation formats, offsets, and tick labels differ significantly.</td>
</tr>
<tr>
<td colspan="3" style="padding: 10px 16px; background-color: #e2e8f0; color: #0f172a; border-bottom: 1px solid #cbd5e1; border-top: 1px solid #cbd5e1;">
<em><strong>Base Report (Overall Weighted Score: 45.84)</strong></em>
</td>
</tr>
<tr style="border-bottom: 1px solid #f1f5f9;">
<td style="padding: 12px 16px; color: #334155;">Type F1 (1.00)</td>
<td style="padding: 12px 16px; color: #334155;">Layout F1 (1.00)</td>
<td style="padding: 12px 16px; color: #475569;">Perfect match in basic layout container and chart type detection.</td>
</tr>
<tr style="border-bottom: 1px solid #f1f5f9;">
<td style="padding: 12px 16px; color: #334155;">Legend F1 (0.78)</td>
<td style="padding: 12px 16px; color: #334155;">Text F1 (0.78)</td>
<td style="padding: 12px 16px; color: #475569;">High similarity in legend content and overall text OCR results.</td>
</tr>
<tr style="border-bottom: 1px solid #f1f5f9;">
<td style="padding: 12px 16px; color: #334155;">Visual Param (0.74)</td>
<td style="padding: 12px 16px; color: #334155;">Data Param (0.14)</td>
<td style="padding: 12px 16px; color: #475569;">Visual parameters closely match, but substantial errors exist in data extraction and mapping.</td>
</tr>
<tr style="border-bottom: 1px solid #cbd5e1;">
<td style="padding: 12px 16px; color: #334155;">Color F1 (0.00)</td>
<td style="padding: 12px 16px; color: #334155;">Grid F1 (0.00)</td>
<td style="padding: 12px 16px; color: #475569;">Complete mismatch in color precision and grid configuration.</td>
</tr>
<tr>
<td colspan="3" style="padding: 10px 16px; background-color: #e2e8f0; color: #0f172a; border-bottom: 1px solid #cbd5e1; border-top: 1px solid #cbd5e1;">
<em><strong>LMM Report (Final Similarity Score: 18.0)</strong></em>
</td>
</tr>
<tr style="border-bottom: 1px solid #f1f5f9;">
<td style="padding: 12px 16px; color: #334155;">Chart Type (100)</td>
<td style="padding: 12px 16px; color: #334155;">Color Style (80)</td>
<td style="padding: 12px 16px; color: #475569;">Both use radar charts. Colors are semantically consistent (red/green/blue/purple) with only minor hue/opacity differences.</td>
</tr>
<tr style="border-bottom: 1px solid #cbd5e1;">
<td style="padding: 12px 16px; color: #334155;">Data (25)</td>
<td style="padding: 12px 16px; color: #334155;">Layout (30)</td>
<td style="padding: 12px 16px; color: #475569;"><strong>Major Errors:</strong> Axis alignment/mapping differs drastically; numeric labels do not match GT. Legend moved, title changed, and axis orientation rotated.</td>
</tr>
<tr style="background-color: #f8fafc;">
<td colspan="3" style="padding: 16px; line-height: 1.6; color: #334155; border-top: 1px solid #cbd5e1;">
<small><strong>Summary:</strong> Charts share the same radar chart type and overall high/low ordering by series, but there are multiple clear visual mismatches: major data-axis alignment/annotation differences and substantial layout/annotation shifts (legend, ticks, orientation), plus minor color/style and numeric-label placement differences.</small>
</td>
</tr>
</tbody>
</table>
</div>
</div>
<!-- ==================== Case 2: Bar Chart ==================== -->
<div id="eval-case-2" class="tab-pane">
<div style="display: flex; flex-wrap: wrap; gap: 20px; justify-content: center; margin-bottom: 25px;">
<div style="flex: 1 1 45%; min-width: 280px; padding: 15px; background: #fafafa; border: 1px solid #e2e8f0; border-radius: 8px; display: flex; flex-direction: column; align-items: center; box-shadow: 0 2px 4px rgba(0,0,0,0.02); box-sizing: border-box;">
<div style="width: 100%; display: flex; justify-content: center; background: #fff; border: 1px solid #eee; border-radius: 4px; padding: 10px;">
<img src="./assets/bar_62.png" alt="GT Figure" style="max-width: 100%; max-height: 350px; object-fit: contain;">
</div>
<p style="margin: 12px 0 0 0; font-size: 0.95em; color: #334155;"><strong>GT_Figure (Left)</strong></p>
</div>
<div style="flex: 1 1 45%; min-width: 280px; padding: 15px; background: #fafafa; border: 1px solid #e2e8f0; border-radius: 8px; display: flex; flex-direction: column; align-items: center; box-shadow: 0 2px 4px rgba(0,0,0,0.02); box-sizing: border-box;">
<div style="width: 100%; display: flex; justify-content: center; background: #fff; border: 1px solid #eee; border-radius: 4px; padding: 10px;">
<img src="./assets/claude_bar_62.png" alt="Generation Figure" style="max-width: 100%; max-height: 350px; object-fit: contain;">
</div>
<p style="margin: 12px 0 0 0; font-size: 0.95em; color: #334155;"><strong>Generation_Figure (Right)</strong></p>
</div>
</div>
<div style="background-color: #f8fafc; border-left: 4px solid #3b82f6; padding: 12px 16px; margin-bottom: 20px; border-radius: 0 4px 4px 0;">
<p style="margin: 0; font-size: 0.95em; line-height: 1.5; color: #334155;">
<strong>Detailed multi-dimensional evaluation statistics.</strong> By pairing related metrics, we observe comprehensive performance across API, Base, and standard Evaluation reports.
</p>
</div>
<div style="overflow-x: auto; width: 100%; border: 1px solid #e2e8f0; border-radius: 8px;">
<table style="width: 100%; min-width: 700px; border-collapse: collapse; text-align: left; font-size: 0.95em; background-color: #fff;">
<thead>
<tr style="background-color: #f1f5f9; border-bottom: 2px solid #cbd5e1;">
<th style="padding: 12px 16px; width: 20%; color: #1e293b;"><strong>Metric 1 (Score)</strong></th>
<th style="padding: 12px 16px; width: 20%; color: #1e293b;"><strong>Metric 2 (Score)</strong></th>
<th style="padding: 12px 16px; width: 60%; color: #1e293b;"><strong>Combined Reason / Key Insight</strong></th>
</tr>
</thead>
<tbody>
<tr>
<td colspan="3" style="padding: 10px 16px; background-color: #e2e8f0; color: #0f172a; border-bottom: 1px solid #cbd5e1;">
<em><strong>LLM Report (Overall Score: 85.00)</strong></em>
</td>
</tr>
<tr style="border-bottom: 1px solid #f1f5f9;">
<td style="padding: 12px 16px; color: #334155;">Chart Type (100.0)</td>
<td style="padding: 12px 16px; color: #334155;">Data Sim (65.0)</td>
<td style="padding: 12px 16px; color: #475569;">Both render side-by-side grouped vertical bar charts. Most values match, but several numeric differences exist (e.g., Microsoft 212 vs 218), producing perceptible height differences.</td>
</tr>
<tr style="border-bottom: 1px solid #f1f5f9;">
<td style="padding: 12px 16px; color: #334155;">Visual Params (75.0)</td>
<td style="padding: 12px 16px; color: #334155;">Color (100.0)</td>
<td style="padding: 12px 16px; color: #475569;">Same color arrays used. Exact bar and figure sizes differ (13x8 vs 14x6), which slightly changes spacing and relative bar thickness.</td>
</tr>
<tr style="border-bottom: 1px solid #f1f5f9;">
<td style="padding: 12px 16px; color: #334155;">Layout (80.0)</td>
<td style="padding: 12px 16px; color: #334155;">Grid (85.0)</td>
<td style="padding: 12px 16px; color: #475569;">Both have 1x2 subplots with horizontal y-axis grids. GT's grid is dashed and more prominent; placement methods for legends cause slight margin differences.</td>
</tr>
<tr style="border-bottom: 1px solid #cbd5e1;">
<td style="padding: 12px 16px; color: #334155;">Legend (85.0)</td>
<td style="padding: 12px 16px; color: #334155;">Text Content (95.0)</td>
<td style="padding: 12px 16px; color: #475569;">Legends are centered above plots with matching entries. Titles and labels match exactly; GT explicitly sets larger font sizes.</td>
</tr>
<tr>
<td colspan="3" style="padding: 10px 16px; background-color: #e2e8f0; color: #0f172a; border-bottom: 1px solid #cbd5e1; border-top: 1px solid #cbd5e1;">
<em><strong>Base Report (Overall Weighted Score: 77.67)</strong></em>
</td>
</tr>
<tr style="border-bottom: 1px solid #f1f5f9;">
<td style="padding: 12px 16px; color: #334155;">Type F1 (1.00)</td>
<td style="padding: 12px 16px; color: #334155;">Layout F1 (1.00)</td>
<td style="padding: 12px 16px; color: #475569;">Perfect structural match in basic layout container and chart type detection.</td>
</tr>
<tr style="border-bottom: 1px solid #f1f5f9;">
<td style="padding: 12px 16px; color: #334155;">Data Param (0.99)</td>
<td style="padding: 12px 16px; color: #334155;">Color F1 (0.96)</td>
<td style="padding: 12px 16px; color: #475569;">Exceptional performance in extracting data values and precise color matching.</td>
</tr>
<tr style="border-bottom: 1px solid #f1f5f9;">
<td style="padding: 12px 16px; color: #334155;">Text F1 (0.88)</td>
<td style="padding: 12px 16px; color: #334155;">Visual Param (0.85)</td>
<td style="padding: 12px 16px; color: #475569;">Strong alignment in text OCR results and visual parameter rendering.</td>
</tr>
<tr style="border-bottom: 1px solid #cbd5e1;">
<td style="padding: 12px 16px; color: #334155;">Grid F1 (0.12)</td>
<td style="padding: 12px 16px; color: #334155;">Legend F1 (0.00)</td>
<td style="padding: 12px 16px; color: #475569;">Model struggles significantly with extracting legend configurations and grid properties.</td>
</tr>
<tr>
<td colspan="3" style="padding: 10px 16px; background-color: #e2e8f0; color: #0f172a; border-bottom: 1px solid #cbd5e1; border-top: 1px solid #cbd5e1;">
<em><strong>LMM Report (Final Similarity Score: 85.0)</strong></em>
</td>
</tr>
<tr style="border-bottom: 1px solid #f1f5f9;">
<td style="padding: 12px 16px; color: #334155;">Chart Type (100)</td>
<td style="padding: 12px 16px; color: #334155;">Color Style (80)</td>
<td style="padding: 12px 16px; color: #475569;">Both are grouped vertical bar charts. Color hues are consistent; minor style differences exist in stroke thickness and bar edges.</td>
</tr>
<tr style="border-bottom: 1px solid #cbd5e1;">
<td style="padding: 12px 16px; color: #334155;">Data (95)</td>
<td style="padding: 12px 16px; color: #334155;">Layout (80)</td>
<td style="padding: 12px 16px; color: #475569;">Heights, ordering, and trends match with negligible pixel differences. Minor shifts in font sizes, legend marker spacing, and subplot margins.</td>
</tr>
<tr style="background-color: #f8fafc;">
<td colspan="3" style="padding: 16px; line-height: 1.6; color: #334155; border-top: 1px solid #cbd5e1;">
<small><strong>Summary:</strong> Charts are visually very similar with matching chart types, data trends, axes, and layout; only minor differences in styling (font weight/size, slight legend marker spacing) and extremely small numeric/text rendering variations are present.</small>
</td>
</tr>
</tbody>
</table>
</div>
</div>
<!-- ==================== Case 3: Combination Chart ==================== -->
<div id="eval-case-3" class="tab-pane">
<div style="display: flex; flex-wrap: wrap; gap: 20px; justify-content: center; margin-bottom: 25px;">
<div style="flex: 1 1 45%; min-width: 280px; padding: 15px; background: #fafafa; border: 1px solid #e2e8f0; border-radius: 8px; display: flex; flex-direction: column; align-items: center; box-shadow: 0 2px 4px rgba(0,0,0,0.02); box-sizing: border-box;">
<div style="width: 100%; display: flex; justify-content: center; background: #fff; border: 1px solid #eee; border-radius: 4px; padding: 10px;">
<img src="./assets/combination_45.png" alt="GT Figure" style="max-width: 100%; max-height: 350px; object-fit: contain;">
</div>
<p style="margin: 12px 0 0 0; font-size: 0.95em; color: #334155;"><strong>GT_Figure (Left)</strong></p>
</div>
<div style="flex: 1 1 45%; min-width: 280px; padding: 15px; background: #fafafa; border: 1px solid #e2e8f0; border-radius: 8px; display: flex; flex-direction: column; align-items: center; box-shadow: 0 2px 4px rgba(0,0,0,0.02); box-sizing: border-box;">
<div style="width: 100%; display: flex; justify-content: center; background: #fff; border: 1px solid #eee; border-radius: 4px; padding: 10px;">
<img src="./assets/claude_combination_45.png" alt="Generation Figure" style="max-width: 100%; max-height: 350px; object-fit: contain;">
</div>
<p style="margin: 12px 0 0 0; font-size: 0.95em; color: #334155;"><strong>Generation_Figure (Right)</strong></p>
</div>
</div>
<div style="background-color: #f8fafc; border-left: 4px solid #3b82f6; padding: 12px 16px; margin-bottom: 20px; border-radius: 0 4px 4px 0;">
<p style="margin: 0; font-size: 0.95em; line-height: 1.5; color: #334155;">
<strong>Detailed multi-dimensional evaluation statistics (Combination).</strong> By pairing related metrics, we observe comprehensive performance across API, Base, and standard Evaluation reports.
</p>
</div>
<div style="overflow-x: auto; width: 100%; border: 1px solid #e2e8f0; border-radius: 8px;">
<table style="width: 100%; min-width: 700px; border-collapse: collapse; text-align: left; font-size: 0.95em; background-color: #fff;">
<thead>
<tr style="background-color: #f1f5f9; border-bottom: 2px solid #cbd5e1;">
<th style="padding: 12px 16px; width: 20%; color: #1e293b;"><strong>Metric 1 (Score)</strong></th>
<th style="padding: 12px 16px; width: 20%; color: #1e293b;"><strong>Metric 2 (Score)</strong></th>
<th style="padding: 12px 16px; width: 60%; color: #1e293b;"><strong>Combined Reason / Key Insight</strong></th>
</tr>
</thead>
<tbody>
<!-- LLM Report -->
<tr>
<td colspan="3" style="padding: 10px 16px; background-color: #e2e8f0; color: #0f172a; border-bottom: 1px solid #cbd5e1;">
<em><strong>LLM Report (Overall Score: 39.50)</strong></em>
</td>
</tr>
<tr style="border-bottom: 1px solid #f1f5f9;">
<td style="padding: 12px 16px; color: #334155;">Chart Type (20.0)</td>
<td style="padding: 12px 16px; color: #334155;">Data Sim (25.0)</td>
<td style="padding: 12px 16px; color: #475569;">GT uses deterministic analytic plots and specific categorical values; GEN samples random data (np.random) and uses histogram-based density plots, completely altering the geometry and underlying distributions.</td>
</tr>
<tr style="border-bottom: 1px solid #f1f5f9;">
<td style="padding: 12px 16px; color: #334155;">Visual Params (40.0)</td>
<td style="padding: 12px 16px; color: #334155;">Color (30.0)</td>
<td style="padding: 12px 16px; color: #475569;">GT applies specific bar widths, lw=2 overlays, and a light color palette (e.g., lightgreen); GEN uses default histogram styles and a mismatched green/blue/orange palette.</td>
</tr>
<tr style="border-bottom: 1px solid #f1f5f9;">
<td style="padding: 12px 16px; color: #334155;">Layout (50.0)</td>
<td style="padding: 12px 16px; color: #334155;">Grid (70.0)</td>
<td style="padding: 12px 16px; color: #475569;">GT arranges a 2x4 grid (17x10) with specific polar dashed grids; GEN uses a 3x4 gridspec (16x12) with default Cartesian histogram grids, causing noticeable aspect differences.</td>
</tr>
<tr style="border-bottom: 1px solid #cbd5e1;">
<td style="padding: 12px 16px; color: #334155;">Legend (60.0)</td>
<td style="padding: 12px 16px; color: #334155;">Text Content (45.0)</td>
<td style="padding: 12px 16px; color: #475569;">GT features precise legend locations and annotation texts (exact percentages); GEN uses default placements and simulates a word cloud with different text.</td>
</tr>
<!-- Base Report -->
<tr>
<td colspan="3" style="padding: 10px 16px; background-color: #e2e8f0; color: #0f172a; border-bottom: 1px solid #cbd5e1; border-top: 1px solid #cbd5e1;">
<em><strong>Base Report (Overall Weighted Score: 46.02)</strong></em>
</td>
</tr>
<tr style="border-bottom: 1px solid #f1f5f9;">
<td style="padding: 12px 16px; color: #334155;">Text F1 (83.92)</td>
<td style="padding: 12px 16px; color: #334155;">Type F1 (75.00)</td>
<td style="padding: 12px 16px; color: #475569;">Strong performance in text OCR and reasonable detection of the multiple basic chart types present.</td>
</tr>
<tr style="border-bottom: 1px solid #f1f5f9;">
<td style="padding: 12px 16px; color: #334155;">Legend F1 (71.50)</td>
<td style="padding: 12px 16px; color: #334155;">Data Param (62.10)</td>
<td style="padding: 12px 16px; color: #475569;">Moderate alignment in legend extraction and data parameter identification, though overall recall remains insufficient.</td>
</tr>
<tr style="border-bottom: 1px solid #f1f5f9;">
<td style="padding: 12px 16px; color: #334155;">Visual Param (54.71)</td>
<td style="padding: 12px 16px; color: #334155;">Color F1 (25.41)</td>
<td style="padding: 12px 16px; color: #475569;">Model struggles significantly with visual rendering parameters and shows severe color mismatches.</td>
</tr>
<tr style="border-bottom: 1px solid #cbd5e1;">
<td style="padding: 12px 16px; color: #334155;">Layout F1 (0.00)</td>
<td style="padding: 12px 16px; color: #334155;">Grid F1 (0.00)</td>
<td style="padding: 12px 16px; color: #475569;">Complete failure in capturing the complex multi-subplot arrangement and grid configurations.</td>
</tr>
<!-- LMM Report -->
<tr>
<td colspan="3" style="padding: 10px 16px; background-color: #e2e8f0; color: #0f172a; border-bottom: 1px solid #cbd5e1; border-top: 1px solid #cbd5e1;">
<em><strong>LMM Report (Final Similarity Score: 0.0)</strong></em>
</td>
</tr>
<tr style="border-bottom: 1px solid #f1f5f9;">
<td style="padding: 12px 16px; color: #334155;">Chart Type (100)</td>
<td style="padding: 12px 16px; color: #334155;">Color Style (0)</td>
<td style="padding: 12px 16px; color: #475569;">Successfully identifies the complex chart mix. However, <strong>Major Errors</strong> exist: missing KDE lines, hatching, bar outlines, and severe donut/wordcloud color mismatches.</td>
</tr>
<tr style="border-bottom: 1px solid #cbd5e1;">
<td style="padding: 12px 16px; color: #334155;">Data (0)</td>
<td style="padding: 12px 16px; color: #334155;">Layout (0)</td>
<td style="padding: 12px 16px; color: #475569;"><strong>Major Errors:</strong> Fundamental data deviations (donut rings collapsed from 90%/40% to a single 50% ring). Subplot composition, aspect ratios, and KDE curve presence are entirely altered.</td>
</tr>
<!-- Summary Row -->
<tr style="background-color: #f8fafc;">
<td colspan="3" style="padding: 16px; line-height: 1.6; color: #334155; border-top: 1px solid #cbd5e1;">
<small><strong>Summary:</strong> Automatic failure (0 score) triggered by multiple major errors. The model fails fundamentally in data presentation (e.g., random data sampling instead of deterministic plotting, collapsed donut rings) and layout composition (missing KDE curves, altered grid arrangement).</small>
</td>
</tr>
</tbody>
</table>
</div>
</div>
</div>
</div>
</div>
<h2 class="text-title">Citation</h2>
<pre id="citation"><code>
@misc{tang2025chartscodehierarchicalbenchmark,
title={From Charts to Code: A Hierarchical Benchmark for Multimodal Models},
author={Jiahao Tang and Henry Hengyuan Zhao and Lijian Wu and Yifei Tao and Dongxing Mao and Yang Wan and Jingru Tan and Min Zeng and Min Li and Alex Jinpeng Wang},
year={2025},
eprint={2510.17932},
archivePrefix={arXiv},
primaryClass={cs.SE},
url={https://arxiv.org/abs/2510.17932},
}
</code></pre>
<p class="acknowledgements">
<b>Acknowledge:</b> Thanks to Carlos & John for this webpage template. Also thanks to the SWE-bench team and their benchmark <a href="https://www.swebench.com/multimodal.html">https://www.swebench.com/multimodal.html</a>.
</p>
<p class="acknowledgements">
<b>Template Usage:</b> If you would like to use this website template for your own leaderboard, please <span style="color:brown">send Carlos & John an email requesting permission.</span> If granted, please make sure to acknowledge the SWE-bench team and link to this leaderboard on the home page of the website.
</p>
<div class="logo-group">
<a href="https://www.csu.edu.cn/"><img src="./assets/csu.gif" /></a>
<a href="https://www.nus.edu.sg/"><img src="https://www.nus.edu.sg/images/default-source/base/logo.png" alt="NUS Logo" /></a>
<a href="https://www.ntu.edu.sg//"><img src="./assets/NTU.png" alt="NUS Logo" /></a>
</div>
</div>
</div>
</section>
</div>
<script src="css/leaderboard.js"></script>
</body>
</html>