-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathmain.html
242 lines (215 loc) · 13.2 KB
/
main.html
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
<!DOCTYPE html>
<html>
<head>
<meta charset="utf-8">
<!-- Meta tags for social media banners, these should be filled in appropriatly as they are your "business card" -->
<!-- Replace the content tag with appropriate information -->
<meta name="description" content="Foundation Models, Surgical Data Science, AI4Health">
<meta property="og:title" content="PeskaVLP"/>
<meta property="og:description" content="Procedure-Aware Surgical Video-language Pretraining with Hierarchical Knowledge Augmentation"/>
<meta property="og:url" content="URL OF THE WEBSITE"/>
<!-- Path to banner image, should be in the path listed below. Optimal dimenssions are 1200X630-->
<meta property="og:image" content=""/>
<meta property="og:image:width" content="1200"/>
<meta property="og:image:height" content="630"/>
<meta name="twitter:title" content="PeskaVLP">
<meta name="twitter:description" content="Procedure-Aware Surgical Video-language Pretraining with Hierarchical Knowledge Augmentation">
<!-- Path to banner image, should be in the path listed below. Optimal dimenssions are 1200X600-->
<meta name="twitter:image" content="static-peskavlp/images/your_twitter_banner_image.png">
<meta name="twitter:card" content="summary_large_image">
<!-- Keywords for your paper to be indexed by-->
<meta name="keywords" content="Foundation Models, Surgical Data Science, AI4Health">
<meta name="viewport" content="width=device-width, initial-scale=1">
<title>Procedure-Aware Surgical Video-language Pretraining with Hierarchical Knowledge Augmentation</title>
<link rel="icon" type="image/x-icon" href="static-peskavlp/images/favicon.png">
<link href="https://fonts.googleapis.com/css?family=Google+Sans|Noto+Sans|Castoro" rel="stylesheet">
<link rel="stylesheet" href="static-peskavlp/css/bulma.min.css">
<link rel="stylesheet" href="static-peskavlp/css/bulma-carousel.min.css">
<link rel="stylesheet" href="static-peskavlp/css/bulma-slider.min.css">
<link rel="stylesheet" href="static-peskavlp/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-peskavlp/css/index.css">
<script src="https://ajax.googleapis.com/ajax/libs/jquery/3.5.1/jquery.min.js"></script>
<script src="https://documentcloud.adobe.com/view-sdk/main.js"></script>
<script defer src="static-peskavlp/js/fontawesome.all.min.js"></script>
<script src="static-peskavlp/js/bulma-carousel.min.js"></script>
<script src="static-peskavlp/js/bulma-slider.min.js"></script>
<script src="static-peskavlp/js/index.js"></script>
</head>
<body>
<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-1 publication-title">Procedure-Aware Surgical Video-language Pretraining with Hierarchical Knowledge Augmentation</h1>
<div class="is-size-5 publication-authors">
<!-- Paper authors -->
<span class="author-block"><a href="https://flaick.github.io/" target="_blank">Kun Yuan</a>,</span>
<span class="author-block"><a href="https://www.linkedin.com/in/vinkle-srivastav/" target="_blank">Vinkle Srivastav</a>,</span>
<span class="author-block"><a href="https://camma.unistra.fr/npadoy/" target="_blank">Nicolas Padoy</a>,</span>
<span class="author-block"><a href="https://www.professoren.tum.de/en/navab-nassir" target="_blank">Nassir Navab</span>
</div>
<div class="is-size-5 publication-authors">
<span class="author-block"><sup>1</sup>University of Strasbourg, <sup>2</sup>Technical University of Munich<br>NeurIPS Spotlight 2024</span>
</div>
<div class="column has-text-centered">
<div class="publication-links">
<!-- ArXiv abstract Link -->
<span class="link-block">
<a href="https://arxiv.org/abs/2410.00263" target="_blank"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="ai ai-arxiv"></i>
</span>
<span>arXiv</span>
</a>
</span>
<!-- Github link -->
<span class="link-block">
<a href="https://github.com/CAMMA-public/SurgVLP" target="_blank"
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>
</div>
</div>
</div>
</div>
</div>
</div>
</section>
<!-- Teaser video-->
<section class="hero teaser">
<div class="container is-max-desktop">
<div class="hero-body">
<img src="static-peskavlp/images/demo_peskavlp.gif" alt="PeskaVLP Demo GIF" style="max-height: 100%; max-width: 100%;">
<h2 class="subtitle has-text-centered">
A simple GIF demonstrates the discrimination and generative tasks that PeskaVLP can do related to laparoscopic surgical images. Please refer to our paper for more details.
</h2>
</div>
</div>
</section>
<!-- End teaser video -->
<!-- Paper abstract -->
<section class="section hero is-light">
<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>
Surgical video-language pretraining (VLP) faces unique challenges due to the knowledge domain gap and the scarcity of multi-modal data. This study aims to bridge the gap by addressing issues regarding textual information loss in surgical lecture videos and the spatial-temporal challenges of surgical VLP. We propose a hierarchical knowledge augmentation approach and a novel Procedure-Encoded Surgical Knowledge-Augmented Video-Language Pretraining (PeskaVLP) framework to tackle these issues. The knowledge augmentation uses large language models (LLM) for refining and enriching surgical concepts, thus providing comprehensive language supervision and reducing the risk of overfitting. PeskaVLP combines language supervision with visual self-supervision, constructing hard negative samples and employing a Dynamic Time Warping (DTW) based loss function to effectively comprehend the cross-modal procedural alignment. Extensive experiments on multiple public surgical scene understanding and cross-modal retrieval datasets show that our proposed method significantly improves zero-shot transferring performance and offers a generalist visual representation for further advancements in surgical scene understanding.
</p>
<figure>
<img src="static-peskavlp/images/overview.png" alt="Overview" class="center-image blend-img-background">
</figure>
</div>
</div>
</div>
</div>
</section>
<!-- End paper abstract -->
<!-- SVL-Pretrain -->
<script type="module" src="https://gradio.s3-us-west-2.amazonaws.com/4.14.0/gradio.js"></script>
<section class="section hero">
<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">Data: SVL-Pretrain</h2>
<div class="content has-text-justified">
<p>
We adopt the previous surgical video-language pretraining dataset SVL-Pretrain, first proposed by <a href="https://arxiv.org/abs/2307.15220" target="_blank">SurgVLP</a>. SVL-Pretrain dataset includes thousands of video-text pairs from surgical lecture videos' audio, metadata.
</p>
<figure>
<img src="static-peskavlp/images/svl_pretrain.png" alt="SVL_Pretrain" class="center-image blend-img-background">
</figure>
<p>
We perform text augmentations to summarize/explain/correct the abstract/keystep/narration texts based on the large language model's knowledge.
</p>
<figure>
<img src="static-peskavlp/images/rewrite.png" alt="Rewrite" class="center-image blend-img-background">
</figure>
</div>
</div>
</div>
</div>
</section>
<!-- End SVL-Pretrain -->
<!-- PeskaVLP -->
<script type="module" src="https://gradio.s3-us-west-2.amazonaws.com/4.14.0/gradio.js"></script>
<section class="section hero">
<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">Pretraining: PeskaVLP</h2>
<div class="content has-text-justified">
<p>
We perform the hierarchical video-language as <a href="https://arxiv.org/abs/2405.10075" target="_blank">HecVL</a> did. We adopt a hierarchical video-language pretraining approach, alternating between video-narration, video-keystep, and video-abstract pairs. Video-narration pairs provide detailed contextual understanding, video-keystep pairs focus on critical procedural actions, and video-abstract pairs offer high-level overviews. This strategy ensures the model captures both fine-grained details and broader contexts, enhancing its performance across various video-language tasks.
</p>
<figure>
<img src="static-peskavlp/images/hecvl.png" alt="hierarchical_pretrain" class="center-image blend-img-background">
</figure>
<p>
At the clip level, we apply augmentations to enhance data diversity and robustness. At the phase and video levels, we focus on procedure understanding, enabling the model to capture and interpret the sequential and hierarchical structure of the entire procedure.
</p>
<figure>
<img src="static-peskavlp/images/pipeline.png" alt="pipeline" class="center-image blend-img-background">
</figure>
</div>
</div>
</div>
</div>
</section>
<!-- End PeskaVLP -->
<!-- Zeroshot -->
<script type="module" src="https://gradio.s3-us-west-2.amazonaws.com/4.14.0/gradio.js"></script>
<section class="section hero is-light">
<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">Zero Shot Evaluation</h2>
<div class="content has-text-justified">
<p>
Zero-shot phase recognition results. We report Accuracy / F1-Score. PeskaVLP outperforms the other methods across different tasks. We report the state-of-the-art methods that are fine-tuned on the downstream dataset in a fully-supervised manner. However, models fine-tuned on specific downstream datasets show limited generalizability across procedures and institutions.
</p>
<figure>
<img src="static-peskavlp/images/zeroshot.png" alt="Zeroshot" class="center-image blend-img-background">
</figure>
</div>
</div>
</div>
</div>
</section>
<!-- End Zeroshot -->
<!--BibTex citation -->
<section class="section" id="BibTeX">
<div class="container is-max-desktop content">
<h2 class="title">BibTeX</h2>
<pre><code>@article{yuan2024procedure,
title={Procedure-Aware Surgical Video-language Pretraining with Hierarchical Knowledge Augmentation},
author={Yuan, Kun and Srivastav, Vinkle and Navab, Nassir and Padoy, Nicolas},
journal={arXiv preprint arXiv:2410.00263},
year={2024}
}</code></pre>
</div>
</section>
<!--End BibTex citation -->
<footer class="footer">
<div class="container">
<div class="columns is-centered">
<div class="column is-8">
<div class="content">
<p>
Last update: 2024/01. Template credited to <a href="https://github.com/eliahuhorwitz/Academic-project-page-template" target="_blank">Academic Project Page Template</a>.
</p>
</div>
</div>
</div>
</div>
</footer>
</body>
</html>