-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathindex.html
More file actions
146 lines (126 loc) · 5.7 KB
/
index.html
File metadata and controls
146 lines (126 loc) · 5.7 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
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>ZoomEarth</title>
<link rel="stylesheet" href="style.css">
<head>
<link rel="icon" href="./images/icon.png" type="image/png">
</head>
</head>
<body>
<nav>
<div class="nav-container">
<a href="#home" class="logo">ZoomEarth</a>
<ul class="nav-links">
<li><a href="#abstract">Abstract</a></li>
<li><a href="#video">Video</a></li>
<li><a href="#experiment">Experiment</a></li>
<li><a href="#bibtex">BibTeX</a></li>
</ul>
<div class="menu-toggle" id="menu-toggle">
<span></span>
<span></span>
<span></span>
</div>
</div>
</nav>
<header id="home" class="hero">
<h1>ZoomEarth</h1>
<p class="subtitle">Active Perception for Ultra-High-Resolution Geospatial Vision-Language Tasks</p>
<p style="font-family:sans-serif; font-size:16px; line-height:1.4;">
Ruixun Liu<sup>1*</sup>, Bowen Fu<sup>1*</sup>, Jiayi Song<sup>1</sup>,
Kaiyu Li<sup>1</sup>, Wanchen Li<sup>1</sup>, Lanxuan Xue<sup>1</sup>,<br>
Hui Qiao<sup>2</sup>, Weizhan Zhang<sup>1</sup>, Deyu Meng<sup>1</sup>,
Xiangyong Cao<sup>1†</sup>
</p>
<p style="font-family:sans-serif; font-size:14px; color:#ffffff; margin-top:6px;">
<sup>1</sup>Xi'an Jiaotong University <sup>2</sup>China Telecom Shaanxi Branch<br>
<small>* Equal contribution. † Corresponding author.</small>
</p>
<p class="links">
<a href="https://arxiv.org/abs/2511.12267" class="btn">
<img src="./images/arxiv-logomark-small.svg" alt="icon" width="15">
<span style="margin: 5px;">Paper</span>
</a>
<a href="https://github.com/earth-insights/ZoomEarth" class="btn">
<img src="./images/github-mark.svg" alt="icon" width="20">
<span style="margin: 5px;">Code</span>
</a>
<a href="https://huggingface.co/HappyBug/ZoomEarth-3B" class="btn">
<img src="./images/hf-logo.svg" alt="icon" width="25">
<span>Model</span>
</a>
<a href="https://huggingface.co/datasets/HappyBug/LRS-GRO" class="btn">
<img src="./images/hf-logo.svg" alt="icon" width="25">
<span>Dataset</span>
</a>
</p>
</header>
<section class="teaser">
<img src="images/teaser.jpg" alt="Teaser Image">
</section>
<section id="video" class="sec">
<h2>Demo Video</h2>
<video src="https://github.com/user-attachments/assets/429a5ca9-6778-4e53-b4bf-dea32310c5e3" controls></video>
</section>
<section id="abstract">
<h2>Abstract</h2>
<p style="font-style: italic;">
Ultra-high-resolution (UHR) remote sensing (RS) images offer rich fine-grained information but also present challenges in effective processing. Existing dynamic resolution and token pruning methods are constrained by a passive perception paradigm, suffering from increased redundancy when obtaining finer visual inputs. In this work, we explore a new active perception paradigm that enables models to revisit information-rich regions. First, we present LRS-GRO, a large-scale benchmark dataset tailored for active perception in UHR RS processing, encompassing 17 question types across global, region, and object levels, annotated via a semi-automatic pipeline. Building on LRS-GRO, we propose ZoomEarth, an adaptive cropping–zooming framework with a novel Region-Guided reward that provides fine-grained guidance. Trained via supervised fine-tuning (SFT) and Group Relative Policy Optimization (GRPO), ZoomEarth achieves state-of-the-art performance on LRS-GRO and, in the zero-shot setting, on three public UHR remote sensing benchmarks. Furthermore, ZoomEarth can be seamlessly integrated with downstream models for tasks such as cloud removal, denoising, segmentation, and image editing through simple tool interfaces, demonstrating strong versatility and extensibility.
</p>
</section>
<section id="framework" class="sec">
<img src="images/flowchart-1.jpg" alt="Flowchart Image">
</section>
<section class="sec">
<img src="images/data_pipelien-1.jpg" alt="Data Pipelien Image">
</section>
<section class="sec">
<img src="images/downstream-1.jpg" alt="Downstream Image">
</section>
<section id="experiment" class="sec">
<h2>Experiment</h2>
<img src="images/exp-1.png" alt="Experiment 1">
<img src="images/exp-2.png" alt="Experiment 2">
</section>
<!-- <section id="video">
<h2>Video</h2>
<div class="video-container">
<iframe src="https://www.youtube.com/embed/XXXXXXX" frameborder="0" allowfullscreen></iframe>
</div>
</section> -->
<section id="bibtex">
<h2>BibTeX</h2>
<pre>
@article{liu2025zoomearth,
title={ZoomEarth: Active Perception for Ultra-High-Resolution Geospatial Vision-Language Tasks},
author={Liu, Ruixun and Fu, Bowen and Song, Jiayi and Li, Kaiyu and Li, Wanchen and Xue, Lanxuan and Qiao, Hui and Zhang, Weizhan and Meng, Deyu and Cao, Xiangyong},
journal={arXiv preprint arXiv:2511.12267},
year={2025}
}
</pre>
</section>
<footer>
<p>© 2025 ZoomEarth</p>
</footer>
<script>
const menuToggle = document.getElementById('menu-toggle');
const navLinks = document.querySelector('.nav-links');
menuToggle.addEventListener('click', () => {
navLinks.classList.toggle('active');
});
</script>
<script>
const nav = document.querySelector("nav");
window.addEventListener("scroll", () => {
if (window.scrollY > 520) {
nav.classList.add("scrolled");
} else {
nav.classList.remove("scrolled");
}
});
</script>
</body>
</html>