vtool-r1.github.io/index.html at main · VTool-R1/vtool-r1.github.io · GitHub

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
<!DOCTYPE html>
<html>
<head>
  <meta charset="utf-8">
  <meta name="description"
        content="VTool-R1: VLMs Learn to Think with Images via Reinforcement Learning on Multimodal Tool Use"> <!-- TODO: add some description, visible outside -->
  <meta name="keywords" content="Reinforcement Learning, LLMs, Strategic Tool Use, UIUC, VTool-R1, VLMs, VLM, ICLR"> <!-- TODO: add some keywords for search engine -->
  <meta name="viewport" content="width=device-width, initial-scale=1">
  <title>VTool-R1</title>

  <link href="https://fonts.cdnfonts.com/css/br-sonoma" rel="stylesheet">

  <!-- Global site tag (gtag.js) - Google Analytics -->
  <script async src="https://www.googletagmanager.com/gtag/js?id=G-PYVRSFMDRL"></script>
  <script>
    window.dataLayer = window.dataLayer || [];

    function gtag() {
      dataLayer.push(arguments);
    }

    gtag('js', new Date());

    gtag('config', 'G-PYVRSFMDRL');
  </script>
  <script async src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-mml-chtml.js"></script>

  <link href="https://fonts.googleapis.com/css?family=Google+Sans|Noto+Sans|Castoro"
        rel="stylesheet">

  <link rel="stylesheet" href="./static/css/bulma.min.css">
  <link rel="stylesheet" href="./static/css/bulma-carousel.min.css">
  <link rel="stylesheet" href="./static/css/bulma-slider.min.css">
  <link rel="stylesheet" href="./static/css/fontawesome_6_7_2.all.css">
  <link rel="stylesheet"
        href="https://cdn.jsdelivr.net/gh/jpswalsh/academicons@1/css/academicons.min.css">
  <link rel="stylesheet" href="./static/css/index.css">

  <script src="https://ajax.googleapis.com/ajax/libs/jquery/3.5.1/jquery.min.js"></script>
  <script defer src="./static/js/fontawesome_6_7_2.all.js"></script>
  <script src="./static/js/bulma-carousel.min.js"></script>
  <script src="./static/js/bulma-slider.min.js"></script>
  <script src="./static/js/index.js"></script>
</head>
<body>

<nav class="navbar" role="navigation" aria-label="main navigation">
  <div class="navbar-brand">
    <a role="button" class="navbar-burger" aria-label="menu" aria-expanded="false">
      <span aria-hidden="true"></span>
      <span aria-hidden="true"></span>
      <span aria-hidden="true"></span>
    </a>
  </div>
  <!-- <div class="navbar-menu">
    <div class="navbar-start" style="flex-grow: 1; justify-content: center;">
      <a class="navbar-item" href="">
      <span class="icon">
          <i class="fas fa-home"></i>
      </span>
      </a> -->

      <!-- <div class="navbar-item has-dropdown is-hoverable"> -->
        <!-- <a class="navbar-link">
          More Research
        </a> -->
        <!-- <div class="navbar-dropdown"> -->
          <!-- <a class="navbar-item" href="https://hypernerf.github.io">
            HyperNeRF
          </a> -->
        <!-- </div> -->
      <!-- </div> -->
    <!-- </div> -->

  <!-- </div> -->
</nav>


<section class="hero">
  <div class="hero-body">
    <div class="container is-max-desktop">
      <div class="columns is-centered">
        <div class="column has-text-centered">
          <h1 class="title is-2 publication-title">
            ICLR 2026
          </h1>
          <h1 class="title is-2 publication-title">
            VTool-R1: VLMs Learn to Think with Images via Reinforcement Learning on Multimodal Tool Use
          </h1> <!-- TODO: fix the title -->
          <div class="is-size-5 publication-authors">
            <span class="author-block">Mingyuan Wu<sup>1*</sup>,</span>
            <span class="author-block">Jingcheng Yang<sup>1*</sup>,</span>
            <span class="author-block">Jize Jiang<sup>1</sup>,</span>
            <span class="author-block">Meitang Li<sup>2</sup>,</span>
            <span class="author-block">Kaizhuo Yan<sup>1</sup>,</span>
            <span class="author-block">Hanchao Yu<sup>3</sup>,</span>
            <span class="author-block">Minjia Zhang<sup>1</sup>,</span>
            <span class="author-block">Chengxiang Zhai<sup>1</sup>,</span>
            <span class="author-block">Klara Nahrstedt<sup>1</sup></span>
          </div>

          <br/>

          <div class="is-size-6 publication-authors">
            <span class="author-block"><sup>1</sup>University of Illinois Urbana Champaign</span>
            <br>
            <span class="author-block"><sup>2</sup>University of Michigan Ann Arbor</span>
            <br>
            <span class="author-block"><sup>3</sup>Independent Researcher</span>
            <br>
            <span class="author-block is-size-7"><sup>*</sup>Equal contribution</span>
          </div>

          <br>

          <div class="is-size-6 publication-authors">
            <span class="author-block">{mw34, klara}@cs.illinois.edu</span>
          </div>

          <div class="column has-text-centered">
            <div class="publication-links">
              <span class="link-block">
                <a href="https://arxiv.org/pdf/2505.19255"
                   class="external-link button is-normal is-rounded is-dark">
                  <span class="icon">
                    <img src="./static/images/ar.svg" alt="img" style="width: 100%; height: 100%">
                  </span>
                  <span>ArXiv</span>
                </a>
              </span>
              <!-- Code Link. -->
              <span class="link-block">
                <a href="https://github.com/VTOOL-R1/vtool-r1"
                   class="external-link button is-normal is-rounded is-dark"> <!-- TODO: fix repo link -->
                  <span class="icon">
                      <i class="fab fa-github"></i>
                  </span>
                  <span>Code</span>
                  </a>
              </span>
              <!-- Dataset Link. -->
              <!--<span class="link-block">
                <a href=""
                   class="external-link button is-normal is-rounded is-dark">
                  <span class="icon">
                      <i class="fa-regular fa-database"></i>
                  </span>
                  <span>Dataset</span>
                  </a>
              </span>-->
              <!-- Model Link. -->
              <span class="link-block">
                <a href="https://huggingface.co/VTOOL"
                   class="external-link button is-normal is-rounded is-dark"> <!-- TODO: fix model link -->
                  <span class="icon">
                    <!-- <i class="fa-solid fa-face-smiling-hands"></i> -->
                    <i class="fa-solid fa-face-smiling-hands" style="color: #FFD43B;"></i>
                  </span>
                  <span>Model</span>
                  </a>
              </span>
            </div>

          </div>
        </div>
      </div>
    </div>
  </div>
</section>

<!-- Teaser -->
<!-- <section class="hero teaser">
  <div class="container is-max-desktop">
    <div class="hero-body">
      <h2 class="subtitle has-text-centered">
        <img src="static/images/puffin.png"/>
        <span class="dnerf">Puffin-Zero</span> is an open reasoning language model that demonstrates the potential of
        large-scale RL training from pretrained checkpoints on solving competition-level math problems.
      </h2>
      <h2 class="subtitle has-text-centered">
        <img src="static/images/score.png"/>
        <span class="dnerf">Puffin-Zero</span> achieves 45.4 points on AIME 2024, a comparable performance to DeepSeek-R1-Zero-
        Qwen-32B.
      </h2>
    </div>
  </div>
</section> -->

<section class="section" style="margin-top: -3rem; margin-bottom: -2rem;">

  <div class="container is-max-desktop">
    <div class="columns is-centered">
      <div class="column is-four-fifths">
        <div class="content">
          <p class="title is-5 has-text-left" style="font-weight: bold;">Updates</p>
          <ul>
            <li>[2026/01/26] VTool-R1 is accepted to ICLR 2026! See you in Brazil 🇧🇷. <span style="color: red;">[New!]</span></li>
            <li>[2025/06/18] Updated ArXiv with better results!</span></li>
            <li>[2025/05/31] Code and model weights available.</li>
            <li>[2025/05/25] ArXiv preprint available. <!--<span style="color: red;">[New!]</span>--></li>
          </ul>
        </div>
      </div>
    </div>

  </div>
</section>

<section class="section">


  <div class="container is-max-desktop">

    <!-- Abstract -->
    <div class="columns is-centered has-text-centered">
      <div class="column is-four-fifths">
        <h2 class="title is-3">Introduction</h2>
        <div class="content has-text-justified">
          <p>
We introduce VTool-R1, one of the first frameworks that trains VLMs to generate multimodal chains of thought by interleaving text and intermediate visual reasoning steps. VTool-R1 integrates Python-based visual editing tools into the RFT process, enabling VLMs to learn when and how to generate visual reasoning steps that benefit final reasoning. Trained with outcome-based rewards tied to task accuracy, our approach elicits strategic visual tool use for reasoning without relying on process-based supervision. Experiments on structured visual question answering over charts and tables show that VTool-R1 enhances reasoning performance by teaching VLMs to "think with images" and generate multimodal chain of thoughts with tools.
          </p>
        </div>
      </div>
    </div>
    <div class="columns is-centered has-text-centered">
      <div class="column is-four-fifths">
        <h2 class="subtitle has-text-centered">
          <img src="./static/images/vtool_example.png"/>
        </h2>
        <b>Figure 2:</b> Qualitative Example from VTool-R1 (3B): The Model Successfully Integrates Intermediate Visual Steps.
      </div>
    </div>

    <div class="columns is-centered has-text-centered">
      <div class="column is-four-fifths">
        <h2 class="subtitle has-text-centered">
          <img src="./static/images/accuracy_table.png" style="width: 300px;"/>
        </h2>
            VTool-R1 7B achieved a <b>71.7%</b> accuracy on the ReFOCUS-TableVQA dataset. Which is <b>10% higher</b> than the base accuracy of 64.7% using Qwen2.5-VL.
      </div>
    </div>
  </div>
</section>


<section class="section">
  <div class="container is-max-desktop">
    <!-- Fully Open-Source -->
    <div class="columns is-centered has-text-centered">
      <div class="column is-four-fifths">
        <div class="content has-text-justified">
          <h3 class="title is-4">Code</h3>
          <p>
            We released the full training and evaluation code. We used <a href="https://github.com/hiyouga/EasyR1">EasyR1</a>, a fork of <a href="https://github.com/volcengine/verl">veRL</a> with support of vision language models for training.
          </p>
          <h3 class="title is-4">Datasets</h3>
          <p>
            For training and validation, we used datasets and tools from <a href="https://zeyofu.github.io/ReFocus/">ReFocus</a>. Please follow the instructions in our repository.
          </p>
          <h3 class="title is-4">BibTeX</h3>
<p>If you find our project helpful, please cite:</p>
<pre style="background-color: #f5f5f5; padding: 0.8rem 1rem 0.4rem 1rem; border-radius: 8px; overflow-x: auto; font-size: 0.9rem;">
@misc{wu2025vtoolr1vlmslearnthink,
      title={VTool-R1: VLMs Learn to Think with Images via Reinforcement Learning on Multimodal Tool Use},
      author={Mingyuan Wu and Jingcheng Yang and Jize Jiang and Meitang Li and Kaizhuo Yan and Hanchao Yu and Minjia Zhang and Chengxiang Zhai and Klara Nahrstedt},
      year={2025},
      eprint={2505.19255},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/2505.19255},
}
</pre>

<p>If you find the dataset helpful, please consider citing Refocus paper:</p>
<pre style="background-color: #f5f5f5; padding: 0.8rem 1rem 0.4rem 1rem; border-radius: 8px; overflow-x: auto; font-size: 0.9rem;">
@misc{fu2025refocusvisualeditingchain,
      title={ReFocus: Visual Editing as a Chain of Thought for Structured Image Understanding},
      author={Xingyu Fu and Minqian Liu and Zhengyuan Yang and John Corring and Yijuan Lu and Jianwei Yang and Dan Roth and Dinei Florencio and Cha Zhang},
      year={2025},
      eprint={2501.05452},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2501.05452},
}
</pre>


          <h3 class="title is-4">Acknowledgements</h3>
          <p>This research used the Delta advanced computing and data resource which is supported by the National Science Foundation (award OAC 2005572) and the State of Illinois. Delta is a joint effort of the University of Illinois Urbana-Champaign and its National Center for Supercomputing Applications.
<br><br>
We would also like to acknowledge Bowen Jin (author of Search-R1) and Xingyu Fu (author of Refocus) for their valuable suggestions and contributions to our project.
<br><br>
This work was supported by the National Science Foundation grants NSF CNS 21-06592, NSF OAC 18-35834 KN, NSF CNS 19-00875 and NSF CCF 22-17144. Any results and opinions are our own and do not represent views of National Science Foundation.
          </p>
        </div>
      </div>
    </div>
  </div>
</section>


<!-- <section class="section">
  <div class="container is-max-desktop">
    <div class="columns is-centered has-text-centered">
      <div class="column is-four-fifths">
        <h2 class="title is-3">Algorithm Insights</h2>
        <div class="content has-text-justified">
          <p>
            We propose the <b>D</b>ecoupled Clip and <b>D</b>ynamic s<b>A</b>mpling <b>P</b>olicy <b>O</b>ptimization (DAPO) algorithm, which includes several key techniques as below. Detailed analysis and insights can be found in our technical report.
          </p>
          <ul>
            <li>
              <b>Clip-Higher</b>, which promotes the diversity of the system and avoids entropy collapse. We observe an entropy-collapsing phenomenon in our initial experiments. We propose increasing the upper clip range of the importance sampling ratio in policy gradient loss to mitigate this problem.
            </li>
          </ul>
          <ul>
            <li>
              <b>Dynamic Sampling</b>, which improves training efficiency and stability. We propose a strategy that performs dynamic sampling and filters out prompt groups with the accuracy equal to 1 and 0, keeping a consistent number of prompts with effective gradients across batches.
            </li>
          </ul>
          <ul>
            <li>
              <b>Token-level Policy Gradient Loss</b>, which is critical in long-CoT RL scenarios.
            </li>
          </ul>
          <ul>
            <li>
              <b>Overlong Reward Shaping</b>, which reduces reward noise and stabilizes training.
            </li>
          </ul>
        </div>
      </div>
    </div>
  </div>
</section> -->

<!-- <section class="section">
  <div class="container is-max-desktop">
    <div class="columns is-centered has-text-centered">
      <div class="column is-four-fifths">
        <h2 class="title is-3">Contributions</h2>
        <div class="content has-text-justified">
          <p>

          </p>
        </div>
      </div>
    </div>
  </div>
</section> -->

<!-- <section class="section">
  <div class="container is-max-desktop">
    <div class="columns is-centered">
      <div class="column is-full-width">
        <h2 class="title is-3">Related Links</h2>
        <div class="content has-text-justified">

        </div>
      </div>
    </div>
  </div>
</section> -->


<!-- <section class="section" id="BibTeX">
  <div class="container is-max-desktop content">
    <h2 class="title">BibTeX</h2>
    <pre><code>@article{2025puffinzero,
                author    = {Author 1, Author 2, Author 3},
                title     = {Puffin-Zero},
                journal   = {Arxiv},
                year      = {2025},
              }</code></pre>
  </div>
</section> -->


<footer class="footer">
  <div class="container">
    <div class="content has-text-centered">
      <!-- <a class="icon-link"
         href="./static/pdf/dapo_paper.pdf">
        <i class="fa-solid fa-file-pdf" style="color: #ec4646;"></i>
      </a> -->
      <!-- <a class="icon-link" href="" class="external-link" disabled>
        <i class="fab fa-github"></i>
      </a> -->
    </div>
    <div class="columns is-centered">
      <div class="column is-8">
        <div class="content">
          <p>
            This website is adapted from <a href="https://nerfies.github.io/">Nerfies</a> and licensed under a <a rel="license"
                                                href="http://creativecommons.org/licenses/by-sa/4.0/">Creative
            Commons Attribution-ShareAlike 4.0 International License</a>.
          </p>
        </div>
      </div>
    </div>
  </div>
</footer>

</body>
</html>