-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathindex.html
More file actions
437 lines (388 loc) · 19.4 KB
/
Copy pathindex.html
File metadata and controls
437 lines (388 loc) · 19.4 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
<!DOCTYPE html>
<html>
<head>
<meta charset="utf-8">
<meta name="description" content="ERGO: Efficient High-Resolution Visual Understanding for Vision-Language Models">
<meta name="keywords" content="ERGO, Vision-Language Models, Reinforcement Learning, High-Resolution, Efficient">
<meta name="viewport" content="width=device-width, initial-scale=1">
<title>ERGO</title>
<link href="https://fonts.googleapis.com/css2?family=Inter:wght@300;400;500;600;700;800&family=Source+Serif+4:ital,wght@0,400;0,600;0,700;1,400&family=JetBrains+Mono:wght@400;500&display=swap" rel="stylesheet">
<link rel="stylesheet" href="./static/css/bulma.min.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">
<link rel="stylesheet" href="./static/css/index.css">
<script defer src="./static/js/fontawesome.all.min.js"></script>
</head>
<body>
<!-- ======== Hero / Title ======== -->
<section class="hero">
<div class="hero-body">
<div class="container is-max-desktop">
<div class="columns is-centered">
<div class="column has-text-centered">
<h1 class="ergo-main-title">
ERGO: Efficient High-Resolution Visual Understanding for Vision-Language Models
</h1>
<!-- Authors -->
<div class="is-size-5 publication-authors">
<span class="author-block">Jewon Lee<sup>*</sup>,</span>
<span class="author-block">Wooksu Shin<sup>*</sup>,</span>
<span class="author-block">Seungmin Yang,</span>
<span class="author-block">Ki-Ung Song,</span>
<span class="author-block">DongUk Lim,</span>
<span class="author-block">Jaeyeon Kim</span>
<br>
<span class="author-block">Tae-Ho Kim<sup>†</sup>,</span>
<span class="author-block">Bo-Kyeong Kim<sup>†</sup></span>
</div>
<div class="is-size-5 publication-authors" style="margin-top: 6px;">
<span class="author-block">Nota Inc.</span>
</div>
<div class="publication-authors author-note">
<span class="author-block"><sup>*</sup>Equal contribution (listed alphabetically)</span>
<span class="author-block"> <sup>†</sup>Corresponding authors</span>
</div>
<!-- Venue Badge -->
<div class="is-size-5 publication-authors" style="margin-top: 10px;">
<a href="https://iclr.cc/virtual/2026/poster/10008298" target="_blank" class="publication-awards">ICLR 2026</a>
</div>
<!-- Links -->
<div class="column has-text-centered">
<div class="publication-links">
<!-- Paper -->
<span class="link-block">
<a href="https://arxiv.org/abs/2509.21991" class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="fas fa-file-pdf"></i>
</span>
<span>Paper</span>
</a>
</span>
<!-- GitHub -->
<span class="link-block">
<a href="https://github.com/nota-github/ERGO"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="fab fa-github"></i>
</span>
<span>Code</span>
</a>
</span>
<!-- HuggingFace -->
<span class="link-block">
<a href="https://huggingface.co/nota-ai/ERGO-7B"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<span style="font-size:18px">🤗</span>
</span>
<span>Model</span>
</a>
</span>
<!-- Demo (placeholder) -->
<span class="link-block">
<a href="https://ergo-demo.nota.ai" class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="fas fa-desktop"></i>
</span>
<span>Demo</span>
</a>
</span>
</div>
</div>
</div>
</div>
</div>
</div>
</section>
<!-- ======== Fig1: Comparison with Prior Work ======== -->
<section class="section" style="padding-bottom: 0;">
<div class="container is-max-desktop">
<div class="columns is-centered has-text-centered">
<div class="column" style="max-width: 960px;">
<div class="figure-box">
<img src="static/images/comparision_with_prior.png" alt="Comparison with prior methods" width="100%" />
<p class="figure-caption">
<strong>Comparison with prior work on high-resolution visual reasoning.</strong>
The yellow box marks the target object, which becomes indiscernible after input-image downsampling.
(a) DeepEyes (Zheng et al., 2025) succeeds when the object remains discernible, but at the cost of a large number of vision tokens.
(b) DeepEyes (Zheng et al., 2025) fails when the object is indiscernible at low resolution, where fewer vision tokens are available.
(c) Our <strong>ERGO</strong> performs reasoning-driven perception, correctly answering the question even on low-resolution images.
</p>
</div>
</div>
</div>
</div>
</section>
<!-- ======== Abstract ======== -->
<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">Abstract</h2>
<div class="content has-text-justified">
<p>
Existing Large Vision-Language Models (LVLMs) incur substantial computational overhead
when processing high-resolution images due to the massive number of vision tokens.
We introduce a two-stage “coarse-to-fine” reasoning pipeline:
a downsampled image is first analyzed to identify task-relevant regions,
then only those regions are cropped at full resolution for subsequent reasoning.
</p>
<p>
A key challenge is that prior methods rely on <em>perception-driven</em> reasoning—they
fail to locate relevant regions once fine-grained visual cues are lost through downsampling.
<strong>ERGO (Efficient Reasoning & Guided Observation)</strong> instead performs
<em>reasoning-driven perception</em>, leveraging multimodal context to determine where to focus
even when target objects become visually indiscernible.
We develop simple yet effective reward components in a reinforcement learning framework
that encourage both accurate region selection and efficient vision-token usage,
achieving higher accuracy with significantly fewer tokens across multiple benchmarks.
</p>
</div>
</div>
</div>
</div>
</section>
<!-- ======== Key Contributions ======== -->
<section class="section" style="padding-top: 0;">
<div class="container is-max-desktop">
<div class="columns is-centered">
<div class="column is-four-fifths">
<h2 class="title is-3 has-text-centered">Key Contributions</h2>
<ul class="contributions-list">
<li data-icon="▶">
<strong>Efficient coarse-to-fine pipeline.</strong>
A two-stage reasoning pipeline that first processes low-resolution inputs to identify task-relevant regions
and then re-encodes them at higher resolution, reducing computational cost while preserving essential information.
</li>
<li data-icon="🎯">
<strong>Reward for reasoning-driven perception.</strong>
With our proposed reward, the policy model learns that relying solely on accurate object localization
is not always optimal and that contextual knowledge can often be more beneficial.
We are the first to demonstrate this insight for high-resolution visual processing in LVLMs.
</li>
<li data-icon="📈">
<strong>State-of-the-art performance with fewer vision tokens.</strong>
ERGO surpasses competitive methods in accuracy on multiple high-resolution benchmarks,
while reducing vision token counts and delivering practical speedups.
</li>
</ul>
</div>
</div>
</div>
</section>
<!-- ======== Fig2: Training Pipeline ======== -->
<section class="section" style="padding-top: 0;">
<div class="container is-max-desktop">
<div class="columns is-centered has-text-centered">
<div class="column" style="max-width: 960px;">
<div class="figure-box">
<img src="static/images/training_pipeline.png" alt="Training Pipeline" width="100%" />
<p class="figure-caption">
<strong>Overview of RL-based training pipeline.</strong>
The red background highlights the components of the proposed TCE reward.
The green background highlights the conventional rewards adopted by most reasoning LVLMs.
</p>
</div>
</div>
</div>
</div>
</section>
<!-- ======== Method ======== -->
<section class="section">
<div class="container is-max-desktop">
<div class="columns is-centered">
<div class="column is-four-fifths">
<h2 class="title is-3 has-text-centered">Method</h2>
<div class="content has-text-justified" style="margin-bottom: 32px;">
<p>
ERGO’s training objective is explicitly aligned with vision-processing efficiency in a reinforcement learning (RL) framework.
Given an image and a text query, the pipeline operates in two stages:
(1) the policy model predicts bounding-box coordinates for the task-relevant region with a thinking trace,
and (2) the cropped region at original resolution is fed back for final answer generation.
</p>
</div>
<h2 class="title is-4 has-text-centered" style="margin-bottom: 20px;">Reward Design</h2>
<div class="method-card">
<h4>Region-Verification Reward</h4>
<p>
Task performance is evaluated using <em>only</em> the cropped region and the query, without access to the original image.
This encourages the policy model to identify informative, self-contained regions that preserve sufficient information for accurate reasoning.
</p>
</div>
<div class="method-card">
<h4>Box Adjustment Reward</h4>
<p>
A complementary reward that regularizes the size of the selected region.
It penalizes overly large crops based on the area ratio, preventing the trivial strategy of selecting the entire image
while allowing flexible region selection.
</p>
</div>
<div class="method-card">
<h4>Task-Driven Contextual Exploration (TCE) Reward</h4>
<p>
The main reward combining region-verification and box adjustment:
<strong>r<sub>TCE</sub> = α · r<sub>region</sub> + β · r<sub>box</sub></strong>.
This enables the policy model to learn robust and efficient region selection strategies for vision-grounded reasoning.
</p>
</div>
<div class="method-card">
<h4>Final Reward Formulation</h4>
<p>
The overall reward is a linear combination of three components:
<strong>R = r<sub>TCE</sub> + r<sub>acc</sub> + r<sub>format</sub></strong>,
where the accuracy reward bridges the training-test mismatch and the format reward enforces well-structured outputs.
</p>
</div>
</div>
</div>
</div>
</section>
<!-- ======== Results ======== -->
<section class="section">
<div class="container">
<h2 class="title is-2 has-text-centered" style="margin-bottom: 40px;">Results</h2>
<!-- Main Results -->
<div class="columns is-centered has-text-centered">
<div class="column is-four-fifths" style="max-width: 900px;">
<h3 class="title is-4">Main Results</h3>
<div class="figure-box">
<img src="static/images/table2.png" alt="Performance comparison under pixel constraints" width="100%" />
<p class="figure-caption">
<strong>Performance comparison under efficiency-considered scenarios with pixel constraints.</strong>
ERGO outperforms the original model and post-training methods across all benchmarks.
† denotes reproduction with their code using our data, while ‡ denotes inference with their original pipeline.
</p>
</div>
</div>
</div>
<!-- Efficiency Analysis -->
<div class="columns is-centered has-text-centered" style="margin-top: 32px;">
<div class="column is-four-fifths" style="max-width: 900px;">
<h3 class="title is-4">Efficiency Analysis</h3>
<div class="figure-box" style="margin-bottom: 24px;">
<img src="static/images/fig4.png" alt="Performance-efficiency trade-off" width="100%" />
<p class="figure-caption">
<strong>Performance-efficiency trade-off on the V* benchmark.</strong>
The total number of vision tokens is the sum of the tokens from the downsampled original image and those from the high-resolution cropped image.
</p>
</div>
<div class="columns">
<div class="column is-half">
<div class="figure-box">
<img src="static/images/table3.png" alt="Vision token counts comparison" width="100%" />
<p class="figure-caption">
<strong>Comparison of vision token counts in coarse-to-fine reasoning.</strong>
</p>
</div>
</div>
<div class="column is-half">
<div class="figure-box">
<img src="static/images/table4.png" alt="Latency comparison" width="100%" />
<p class="figure-caption">
<strong>Latency comparison with models that leverage multiple tool calls on V* using the vLLM engine.</strong>
Latency represents the average duration to produce a final answer for each image–query pair.
</p>
</div>
</div>
</div>
</div>
</div>
<!-- In-Depth Analysis -->
<div class="columns is-centered has-text-centered" style="margin-top: 32px;">
<div class="column is-four-fifths" style="max-width: 900px;">
<h3 class="title is-4">In-Depth Analysis</h3>
<div class="columns">
<div class="column is-half">
<div class="figure-box">
<img src="static/images/fig5.png" alt="Robustness under target-object masking" width="100%" />
<p class="figure-caption">
<strong>Evaluation of model robustness under target-object masking.</strong>
Models can only succeed by leveraging contextual information when the object is completely masked.
ERGO achieves the most robust performance in the masked condition.
</p>
</div>
</div>
<div class="column is-half">
<div class="figure-box">
<img src="static/images/table5.png" alt="Conventional benchmark results" width="100%" />
<p class="figure-caption">
<strong>Results on conventional vision–language benchmarks.</strong>
ERGO maintains or improves the capabilities of the base Qwen2.5-VL-7B model.
</p>
</div>
</div>
</div>
</div>
</div>
<!-- Ablation Studies -->
<div class="columns is-centered has-text-centered" style="margin-top: 32px;">
<div class="column is-four-fifths" style="max-width: 900px;">
<h3 class="title is-4">Ablation Studies</h3>
<div class="figure-box" style="margin-bottom: 24px;">
<img src="static/images/table6.png" alt="Ablation analysis" width="100%" />
<p class="figure-caption">
<strong>Ablation analysis.</strong>
Average performance is measured over six benchmarks.
(a) Reward design, (b) TCE reward weight, (c) Parameter size, (d) Reward model, (e) Box adjustment reward.
</p>
</div>
</div>
</div>
</div>
</section>
<!-- ======== Qualitative Results ======== -->
<section class="section">
<div class="container is-max-desktop">
<h2 class="title is-2 has-text-centered" style="margin-bottom: 40px;">Qualitative Results</h2>
<div class="columns is-centered has-text-centered">
<div class="column is-half">
<div class="figure-box">
<img src="static/images/1.png" alt="Qualitative result 1" width="100%" />
<p class="figure-caption">
ERGO utilizes coarse cues (“the region where the bottle is located”) to provide the answer.
</p>
</div>
</div>
<div class="column is-half">
<div class="figure-box">
<img src="static/images/3.png" alt="Qualitative result 3" width="100%" />
<p class="figure-caption">
ERGO can also exploit clear visual cues (the purple umbrella and the orange luggage) when the object is still discernible.
</p>
</div>
</div>
</div>
</div>
</section>
<!-- ======== Citation ======== -->
<section class="section" id="BibTeX">
<div class="container is-max-desktop content">
<h2 class="title is-3 has-text-centered">Citation</h2>
<pre><code>@misc{lee2025ergoefficienthighresolutionvisual,
title={ERGO: Efficient High-Resolution Visual Understanding for Vision-Language Models},
author={Jewon Lee and Wooksu Shin and Seungmin Yang and Ki-Ung Song and DongUk Lim and Jaeyeon Kim and Tae-Ho Kim and Bo-Kyeong Kim},
year={2025},
eprint={2509.21991},
archivePrefix={arXiv},
primaryClass={cs.CV},
url={https://arxiv.org/abs/2509.21991},
}</code></pre>
</div>
</section>
<!-- ======== Footer ======== -->
<footer class="footer">
<div class="columns is-centered">
<div class="column is-8">
<div class="content">
<p style="text-align: center;">
This website is adapted from <a href="https://nerfies.github.io/">Nerfies</a> and <a
href="https://mathvista.github.io/">MathVista</a>.
<br>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>
</footer>
</body>
</html>