forked from Tencent/YOLO-Master
-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathindex.html
More file actions
362 lines (324 loc) · 15.3 KB
/
Copy pathindex.html
File metadata and controls
362 lines (324 loc) · 15.3 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>Introduction — YOLO-Master</title>
<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=Roboto:wght@300;400;500;700&family=Roboto+Mono:wght@400;500&display=swap" rel="stylesheet">
<link rel="stylesheet" href="https://cdnjs.cloudflare.com/ajax/libs/font-awesome/6.4.0/css/all.min.css">
<link rel="stylesheet" href="./assets/css/style.css">
</head>
<body>
<!-- ==================== App Bar ==================== -->
<header class="app-bar" role="banner">
<div class="app-bar-inner">
<a href="./index.html" class="app-bar-brand">
<span class="app-bar-brand-icon"><i class="fas fa-bolt"></i></span>
YOLO-Master
</a>
<nav class="app-bar-links" aria-label="Main navigation">
<a href="./index.html" class="active"><i class="fas fa-home"></i> Home</a>
<a href="./wiki.html"><i class="fas fa-book"></i> Wiki</a>
<a href="./docs.html"><i class="fas fa-file-lines"></i> Docs</a>
<a href="./model-zoo.html"><i class="fas fa-cubes"></i> Model Zoo</a>
<a href="./demo.html"><i class="fas fa-play"></i> Demo</a>
</nav>
<div class="app-bar-search">
<i class="fas fa-search"></i> Search
</div>
<div class="app-bar-meta">
<a href="https://github.com/Tencent/YOLO-Master" target="_blank" rel="noopener noreferrer">
<i class="fab fa-github"></i>
<img src="https://img.shields.io/github/stars/Tencent/YOLO-Master?style=flat-square&logo=github&logoColor=white&color=3f51b5&label=" alt="GitHub Stars" class="meta-badge">
</a>
<span><i class="fas fa-tag"></i> v26.02</span>
</div>
<button class="app-bar-toggle" aria-label="Toggle menu"><i class="fas fa-bars"></i></button>
</div>
</header>
<!-- ==================== Three-Column Layout ==================== -->
<div class="doc-layout">
<!-- Left Sidebar -->
<aside class="doc-sidebar">
<ul class="doc-sidebar-nav">
<li class="nav-section">Intro</li>
<li><a href="#introduction" class="active">Introduction</a></li>
<li><a href="#faq">FAQ</a></li>
<li class="nav-section">Navigate</li>
<li><a href="#quick-start">Quickstart</a></li>
<li><a href="./wiki.html">Wiki</a></li>
<li><a href="./docs.html">Docs</a></li>
<li><a href="./model-zoo.html">Model Zoo</a></li>
<li><a href="./demo.html">Demo</a></li>
</ul>
</aside>
<!-- Main Content -->
<main class="doc-content">
<!-- Hero Intro -->
<div class="hero-intro">
<h1>YOLO-Master</h1>
<p class="text-muted">A powerful MoE-accelerated framework for real-time object detection</p>
<!-- Mascot Logo -->
<img src="./assets/images/youtu-mascot-120.png" alt="YOLO-Master Mascot" class="hero-logo" width="120" height="120" loading="eager">
<!-- Org Logo -->
<div class="org-logo">
<img src="./assets/images/youtu-mascot-80.png" alt="Tencent Youtu Lab" class="org-logo-img" width="28" height="28" loading="lazy">
<span>Tencent Youtu Lab</span>
</div>
<!-- Badges -->
<div class="hero-badges">
<a href="https://arxiv.org/abs/2512.23273" target="_blank" rel="noopener noreferrer">
<img src="https://img.shields.io/badge/arXiv-2512.23273-b31b1b.svg?style=flat-square" alt="arXiv" loading="lazy">
</a>
<a href="https://github.com/Tencent/YOLO-Master" target="_blank" rel="noopener noreferrer">
<img src="https://img.shields.io/github/stars/isLinXu/YOLO-Master?style=social" alt="GitHub Stars" loading="lazy">
</a>
<a href="https://huggingface.co/spaces/gatilin/YOLO-Master-WebUI-Demo" target="_blank" rel="noopener noreferrer">
<img src="https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Demo-blue?style=flat-square" alt="HuggingFace" loading="lazy">
</a>
<img src="https://img.shields.io/badge/CVPR-2026-blue?style=flat-square&color=1e40af" alt="CVPR 2026" loading="lazy">
<a href="https://github.com/Tencent/YOLO-Master/blob/main/LICENSE" target="_blank" rel="noopener noreferrer">
<img src="https://img.shields.io/badge/License-AGPL%203.0-blue.svg?style=flat-square" alt="License" loading="lazy">
</a>
</div>
</div>
<!-- Introduction -->
<section id="introduction">
<h2>Introduction</h2>
<p>
Existing Real-Time Object Detection (RTOD) methods commonly adopt YOLO-like architectures
for their favorable trade-off between accuracy and speed. However, these models rely on
<strong>static dense computation</strong> that applies uniform processing to all inputs regardless
of scene complexity.
</p>
<p>
<strong>YOLO-Master</strong> introduces instance-conditional adaptive computation via
<strong>Efficient Sparse Mixture-of-Experts (ES-MoE)</strong> blocks. It dynamically allocates
computational resources according to scene complexity, outperforming state-of-the-art models
on large-scale benchmarks including COCO, VOC, VisDrone, KITTI, and SKU-110K.
</p>
<p>
Our framework is the first deep integration of Mixture-of-Experts (MoE) into YOLO architecture,
enabling dynamic intelligence for vision tasks. By routing each instance to a specialized subset
of experts, YOLO-Master achieves superior accuracy-efficiency trade-offs across diverse scenarios
ranging from autonomous driving to retail analytics.
</p>
</section>
<!-- Core Architecture -->
<section id="core-architecture">
<h2>Core Architecture</h2>
<p>
The YOLO-Master architecture consists of three main components: a lightweight backbone with
ES-MoE blocks, a feature pyramid neck, and a detection head with dynamic routing.
</p>
<img
src="https://github.com/user-attachments/assets/6caa1065-af77-4f77-8faf-7551c013dacd"
alt="YOLO-Master Architecture Diagram showing ES-MoE blocks with dynamic routing across backbone, neck, and head components"
loading="lazy"
>
<div class="abstract-box">
<p>
ES-MoE blocks enable compute-on-demand via dynamic routing. Each block contains multiple
expert networks and a lightweight router that determines which experts to activate for
each input instance. This sparse activation pattern reduces computational overhead while
maintaining expressive power.
</p>
</div>
<p>
The router network analyzes intermediate feature representations and produces a sparse
gating distribution over experts. Only the top-k experts are activated per instance,
enabling significant speedups on less complex scenes while preserving accuracy on
challenging inputs.
</p>
</section>
<!-- Key Features -->
<section id="key-features">
<h2>Key Features</h2>
<p class="text-muted">Six core innovations powering the next generation of real-time detection</p>
<div class="features-grid">
<article class="feature-card">
<div class="feature-icon purple"><i class="fas fa-brain"></i></div>
<h3>Mixture-of-Experts (MoE)</h3>
<p>ES-MoE blocks with dynamic routing for adaptive computation, allocating experts on a per-instance basis.</p>
</article>
<article class="feature-card">
<div class="feature-icon green"><i class="fas fa-sliders-h"></i></div>
<h3>LoRA Fine-Tuning</h3>
<p>Parameter-efficient fine-tuning with ~10% trainable parameters, enabling rapid domain adaptation.</p>
</article>
<article class="feature-card">
<div class="feature-icon orange"><i class="fas fa-bolt"></i></div>
<h3>Sparse SAHI</h3>
<p>3–5× speedup on 4K/8K images via content-adaptive slicing, skipping redundant regions.</p>
</article>
<article class="feature-card">
<div class="feature-icon blue"><i class="fas fa-project-diagram"></i></div>
<h3>Cluster-Weighted NMS</h3>
<p>Gaussian-weighted fusion for dense scenes, improving mAP in crowded environments.</p>
</article>
<article class="feature-card">
<div class="feature-icon pink"><i class="fas fa-globe"></i></div>
<h3>Mixture-of-Attention (MoA)</h3>
<p>CNN-native multi-scale attention fusion, bridging local inductive bias and global context.</p>
</article>
<article class="feature-card">
<div class="feature-icon cyan"><i class="fas fa-robot"></i></div>
<h3>Agent Skill System</h3>
<p>9 skill runners for AI agent orchestration, enabling autonomous vision pipelines.</p>
</article>
</div>
</section>
<!-- Quick Start -->
<section id="quick-start">
<h2>Quick Start</h2>
<p class="text-muted">Get up and running in under 5 minutes</p>
<h3>Installation</h3>
<div class="code-block">
<div class="code-header">
<span class="code-lang">Bash</span>
<button class="code-copy" aria-label="Copy code"><i class="far fa-copy"></i> Copy</button>
</div>
<pre><code>conda create -n yolo_master python=3.11 -y
conda activate yolo_master
git clone https://github.com/Tencent/YOLO-Master.git
cd YOLO-Master
pip install -r requirements.txt
pip install -e .</code></pre>
</div>
<h3>Python Inference</h3>
<div class="code-block">
<div class="code-header">
<span class="code-lang">Python</span>
<button class="code-copy" aria-label="Copy code"><i class="far fa-copy"></i> Copy</button>
</div>
<pre><code>from ultralytics import YOLO
model = YOLO("yolo_master_n.pt")
results = model("path/to/image.jpg")
results[0].show()</code></pre>
</div>
</section>
<!-- Citation -->
<section id="citation">
<h2>Citation</h2>
<p class="text-muted">If you find YOLO-Master useful in your research, please consider citing our paper</p>
<div class="code-block">
<div class="code-header">
<span class="code-lang">BibTeX</span>
<button class="code-copy" aria-label="Copy BibTeX"><i class="far fa-copy"></i> Copy</button>
</div>
<pre><code>@inproceedings{lin2026yolomaster,
title={{YOLO-Master}: MOE-Accelerated with Specialized Transformers for Enhanced Real-time Detection},
author={Lin, Xu and Peng, Jinlong and Gan, Zhenye and Zhu, Jiawen and Liu, Jun},
booktitle={CVPR},
year={2026}
}</code></pre>
</div>
</section>
<!-- FAQ -->
<section id="faq">
<h2>FAQ</h2>
<h3>What is ES-MoE?</h3>
<p>
ES-MoE (Efficient Sparse Mixture-of-Experts) is our novel building block that replaces
standard convolutional layers in YOLO. It routes each instance to a subset of expert
networks based on learned gating, enabling adaptive computation.
</p>
<h3>Which datasets are supported?</h3>
<p>
YOLO-Master supports COCO, VOC, VisDrone, KITTI, SKU-110K, and any custom dataset
compatible with the YOLO format. Pre-trained weights are available for all model sizes
(N/S/M/L/X).
</p>
<h3>Can I use MoE with my existing YOLO models?</h3>
<p>
Yes! YOLO-Master is designed as a drop-in replacement. You can use our ES-MoE blocks
in your existing YOLO pipelines with minimal code changes. See the <a href="./docs.html">Docs</a>
for integration guides.
</p>
</section>
</main>
<!-- Right TOC -->
<aside class="doc-toc">
<div class="doc-toc-title">On this page</div>
<ul>
<li><a href="#introduction">Introduction</a></li>
<li><a href="#core-architecture">Core Architecture</a></li>
<li><a href="#key-features">Key Features</a></li>
<li><a href="#quick-start">Quick Start</a></li>
<li><a href="#citation">Citation</a></li>
<li><a href="#faq">FAQ</a></li>
</ul>
</aside>
</div>
<!-- ==================== Footer ==================== -->
<footer class="footer" role="contentinfo">
<div class="footer-inner">
<div class="footer-grid">
<div>
<div class="footer-brand"><i class="fas fa-bolt"></i> YOLO-Master</div>
<p class="footer-desc">MoE-Accelerated Real-time Detection Framework</p>
</div>
<div>
<div class="footer-heading">Project</div>
<ul class="footer-links">
<li><a href="./index.html">Home</a></li>
<li><a href="./wiki.html">Wiki</a></li>
<li><a href="./docs.html">Docs</a></li>
<li><a href="./model-zoo.html">Model Zoo</a></li>
<li><a href="./demo.html">Demo</a></li>
</ul>
</div>
<div>
<div class="footer-heading">Resources</div>
<ul class="footer-links">
<li><a href="https://github.com/Tencent/YOLO-Master" target="_blank" rel="noopener noreferrer">GitHub</a></li>
<li><a href="https://arxiv.org/abs/2512.23273" target="_blank" rel="noopener noreferrer">Paper</a></li>
<li><a href="https://huggingface.co/spaces/gatilin/YOLO-Master-WebUI-Demo" target="_blank" rel="noopener noreferrer">HuggingFace</a></li>
</ul>
</div>
<div>
<div class="footer-heading">Legal</div>
<ul class="footer-links">
<li><a href="https://github.com/Tencent/YOLO-Master/blob/main/LICENSE" target="_blank" rel="noopener noreferrer">License</a></li>
</ul>
</div>
</div>
<div class="footer-bottom">
<span>© 2025 Tencent Youtu Lab. Released under AGPL-3.0.</span>
<span>Made with <i class="fas fa-heart" style="color: #ef4444;"></i></span>
</div>
</div>
</footer>
<!-- ==================== Scripts ==================== -->
<script>
// Copy code function
function copyCode(button) {
const codeBlock = button.closest('.code-block').querySelector('code');
const text = codeBlock.innerText;
navigator.clipboard.writeText(text).then(() => {
const original = button.innerHTML;
button.innerHTML = '<i class="fas fa-check"></i> Copied!';
setTimeout(() => {
button.innerHTML = original;
}, 2000);
});
}
// Attach copy handlers
document.querySelectorAll('.code-copy').forEach(btn => {
btn.addEventListener('click', () => copyCode(btn));
});
// Mobile nav toggle
const toggle = document.querySelector('.app-bar-toggle');
const navLinks = document.querySelector('.app-bar-links');
if (toggle && navLinks) {
toggle.addEventListener('click', () => {
const expanded = toggle.getAttribute('aria-expanded') === 'true';
toggle.setAttribute('aria-expanded', !expanded);
navLinks.style.display = expanded ? 'none' : 'flex';
});
}
</script>
</body>
</html>