-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathindex.html
More file actions
422 lines (386 loc) · 19.6 KB
/
index.html
File metadata and controls
422 lines (386 loc) · 19.6 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
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="utf-8">
<meta name="description" content="
FlowCLAS is a novel normalizing flow-based anomaly segmentation method that improves
normalizing flows by introducing a contrastive learning framework with pseudo-outliers.">
<meta name="keywords"
content="road anomaly segmentation, space anomaly segmentation, autonomous driving, computer vision, robotics">
<meta name="viewport" content="width=device-width, initial-scale=1">
<title>FlowCLAS: Enhancing Normalizing Flow-Based Anomaly Segmentation Via Contrastive Learning</title>
<!-- Google Tag Manager -->
<script>(function (w, d, s, l, i) {
w[l] = w[l] || []; w[l].push({
'gtm.start':
new Date().getTime(), event: 'gtm.js'
}); var f = d.getElementsByTagName(s)[0],
j = d.createElement(s), dl = l != 'dataLayer' ? '&l=' + l : ''; j.async = true; j.src =
'https://www.googletagmanager.com/gtm.js?id=' + i + dl; f.parentNode.insertBefore(j, f);
})(window, document, 'script', 'dataLayer', 'GTM-WNQ8P9M7');</script>
<!-- End Google Tag Manager -->
<link href="https://fonts.googleapis.com/css?family=Google+Sans|Noto+Sans|Castoro" rel="stylesheet">
<link rel="stylesheet" href="./static/css/bulma.min.css?v=20260305">
<link rel="stylesheet" href="https://cdn.jsdelivr.net/gh/jpswalsh/academicons@1/css/academicons.min.css">
<link rel="stylesheet" href="./static/css/index.css?v=20260305">
<link rel="icon" href="./static/images/TRAIL_BLACK_ICON.png">
<script src="https://ajax.googleapis.com/ajax/libs/jquery/3.5.1/jquery.min.js"></script>
<script defer src="./static/js/fontawesome.all.min.js"></script>
<script src="./static/js/index.js?v=20260305"></script>
</head>
<body>
<!-- Google Tag Manager (noscript) -->
<noscript><iframe src="https://www.googletagmanager.com/ns.html?id=GTM-WNQ8P9M7" height="0" width="0"
style="display:none;visibility:hidden"></iframe></noscript>
<!-- End Google Tag Manager (noscript) -->
<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">
FlowCLAS: Enhancing Normalizing Flow-Based Anomaly Segmentation Via Contrastive Learning
</h1>
<div class="is-size-3 publication-venue">WACV 2026</div>
<div class="is-size-5 publication-authors">
<span class="author-block">
<a href="https://johncwlee.github.io/">Chang Won Lee</a><sup>1</sup>,</span>
<span class="author-block">
Selina Leveugle<sup>1</sup>,</span>
<span class="author-block">
Paul Grouchy<sup>2</sup>,</span>
<span class="author-block">
Chris Langley<sup>2</sup>,</span>
<span class="author-block">
Svetlana Stolpner<sup>2</sup>,</span>
</div>
<div class="is-size-5 publication-authors">
<span class="author-block">
<a href="https://starslab.ca/people/prof-jonathan-kelly/">Jonathan Kelly</a><sup>1</sup>,</span>
<span class="author-block">
<a href="https://www.trailab.utias.utoronto.ca/steven-waslander">Steven L. Waslander</a><sup>1</sup>
</span>
</div>
<div class="is-size-5 publication-authors">
<span class="author-block"><sup>1</sup>University of Toronto</span>
<span class="author-block" style="margin-left: 1.5em;"><sup>2</sup>MDA Space</span>
</div>
<div class="column has-text-centered">
<div class="publication-links">
<!-- PDF Link. -->
<span class="link-block">
<a href="" 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>
<span class="link-block">
<a href="https://arxiv.org/abs/2411.19888" 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>
<!-- Video Link. -->
<span class="link-block">
<a href="https://youtu.be/CUTDzFNhzaY" class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="fab fa-youtube"></i>
</span>
<span>Video</span>
</a>
</span>
<!-- Code Link. -->
<span class="link-block">
<!-- TODO: Add code (https://github.com/TRAILab/FlowCLAS)-->
<a href=""
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="fab fa-github"></i>
</span>
<span>Code coming soon</span>
</a>
</span>
</div>
<p class="availability-note">
Code is not public yet. The GitHub repository will be updated here once the release is ready.
</p>
</div>
</div>
</div>
</div>
</div>
</section>
<!--section class="section">
<div class="container is-max-desktop">
<div class="columns is-centered">
<div class="column is-four-fifths">
<div class="content abstract-card">
<h2 class="title is-3 has-text-centered">Abstract</h2>
<p>
Anomaly segmentation is an essential capability for safety-critical robotics applications that must be aware of unexpected events. Normalizing flows (NFs), a class of generative models, are a promising approach for this task due to their ability to model the inlier data distribution efficiently. However, their performance falters in dynamic scenes, where complex, multi-modal data distributions cause them to struggle with identifying out-of-distribution samples, leaving a performance gap to leading discriminative methods.
To address this limitation, we introduce FlowCLAS, a hybrid framework that enhances the traditional maximum likelihood objective of NFs with a discriminative, contrastive loss. Leveraging Outlier Exposure, this objective explicitly enforces a separation between normal and anomalous features in the latent space, retaining the probabilistic foundation of NFs while embedding the discriminative power they lack.
The strength of this approach is demonstrated by FlowCLAS establishing new state-of-the-art (SOTA) performance across multiple challenging anomaly segmentation benchmarks for robotics, including Fishyscapes Lost \& Found, Road Anomaly, SegmentMeIfYouCan-ObstacleTrack, and ALLO. Our experiments also show that this contrastive approach is more effective than other outlier-based training strategies for NFs, successfully bridging the performance gap to leading discriminative methods.
</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">🌟 Key Features</h2>
<div class="content has-text-justified">
<ul>
<li>
<strong>Hybrid objective for normalizing flows.</strong>
We introduce FlowCLAS, a framework that combines maximum-likelihood training with an outlier-aware contrastive objective, encouraging a more separable latent space for robust anomaly segmentation in dynamic scenes.
</li>
<li>
<strong>Contrastive learning is the key ingredient.</strong>
Ablations show that contrastive learning outperforms other outlier-based training strategies for normalizing flows and is critical for learning object-level semantic features rather than low-level patterns.
</li>
<li>
<strong>Strong performance across multiple domains.</strong>
FlowCLAS delivers strong results on both road anomaly segmentation benchmarks for autonomous driving (Fishyscapes
Lost & Found, Road Anomaly, SegmentMeIfYouCan-ObstacleTrack) and the ALLO benchmark for space anomaly segmentation, showing that the approach transfers well across very different robotics settings.
</li>
</ul>
</div>
</div>
</div>
</div>
</section>
<section class="hero teaser">
<div class="container is-max-desktop">
<div class="hero-body">
<div class="has-text-centered">
<h2 class="title is-3" style="margin-bottom: 1em;">🧠 Method Overview</h2>
</div>
<div class="has-text-centered" style="margin-top: 1em;">
<img src="./static/images/flowclas_overview.png" style="width: 80%;" alt="FlowCLAS diagram">
</div>
<h2 class="subtitle" style="text-align: justify;">
<strong>FlowCLAS Overview.</strong> The framework operates in two stages. During training, a frozen vision
encoder <i>f</i><sub>φ</sub> extracts features from a mixed-content image <i>x</i><sup>mix</sup> and an
auxiliary outlier image <i>x</i><sup>out</sup>. A normalizing flow network <i>f</i><sub>θ</sub> then
maps these features to a latent space, producing <i>z</i><sup>{mix,out}</sup>. The model is optimized via a
hybrid objective: (1) a maximum likelihood loss (<i>L</i><sub>ml</sub>) pushes the latent samples <i>z</i>
corresponding to normal regions toward a base Multivariate Gaussian distribution, and (2) a contrastive loss
(<i>L</i><sub>con</sub>) enforces a separation between the latent representations of normal and anomalous
features in the projection space. During inference, the normalizing flow computes a likelihood-based anomaly
score map for a given test image, localizing regions that deviate from the learned distribution of normal
data.
</h2>
</div>
</div>
</section>
<section class="section results-section">
<div class="container is-max-desktop">
<div class="has-text-centered section-heading">
<h2 class="title is-3">📊 Results at a Glance</h2>
<p class="section-lede">
FlowCLAS establishes new state-of-the-art on multiple road and space anomaly segmentation benchmarks while
remaining competitive on the most challenging SMIYC split.
</p>
</div>
<div class="results-grid">
<article class="result-card">
<p class="result-card-eyebrow">Road anomaly segmentation</p>
<h3 class="title is-5">Fishyscapes Lost & Found</h3>
<div class="result-metrics">
<div class="metric-chip">
<span class="metric-label">AUPRC</span>
<span class="metric-value">88.8</span>
</div>
<div class="metric-chip">
<span class="metric-label">FPR95</span>
<span class="metric-value">0.7</span>
</div>
</div>
<p class="result-card-text">
Best overall result on FS-L&F, improving AUPRC by 3.2 points and reducing FPR95 by 0.6 relative to the
strongest prior method reported in the paper.
</p>
</article>
<article class="result-card">
<p class="result-card-eyebrow">Road anomaly segmentation</p>
<h3 class="title is-5">Road Anomaly</h3>
<div class="result-metrics">
<div class="metric-chip">
<span class="metric-label">AUPRC</span>
<span class="metric-value">93.0</span>
</div>
<div class="metric-chip">
<span class="metric-label">FPR95</span>
<span class="metric-value">3.3</span>
</div>
</div>
<p class="result-card-text">
Sets a new best AUPRC and improves FPR95 by 2.0 points over the previous strongest baseline on Road
Anomaly.
</p>
</article>
<article class="result-card">
<p class="result-card-eyebrow">SegmentMeIfYouCan benchmark</p>
<h3 class="title is-5">SegmentMeIfYouCan ObstacleTrack</h3>
<div class="result-metrics">
<div class="metric-chip">
<span class="metric-label">AUPRC</span>
<span class="metric-value">94.2</span>
</div>
<div class="metric-chip">
<span class="metric-label">FPR95</span>
<span class="metric-value">0.1</span>
</div>
</div>
<p class="result-card-text">
Achieves the best ObstacleTrack result. On the harder AnomalyTrack split, FlowCLAS reaches 94.3 AUPRC and
remains competitive with the top entry.
</p>
</article>
<article class="result-card">
<p class="result-card-eyebrow">Space anomaly segmentation</p>
<h3 class="title is-5">ALLO</h3>
<div class="result-metrics">
<div class="metric-chip">
<span class="metric-label">AUPRC</span>
<span class="metric-value">88.4</span>
</div>
<div class="metric-chip">
<span class="metric-label">FPR95</span>
<span class="metric-value">6.6</span>
</div>
</div>
<p class="result-card-text">
Delivers a 7.6-point AUPRC gain over the previous best result on ALLO while keeping false positives
competitive in this more difficult low-light orbital setting.
</p>
</article>
</div>
</div>
</section>
<section class="section why-section">
<div class="container is-max-desktop">
<div class="has-text-centered section-heading">
<h2 class="title is-3">Why It Works</h2>
<p class="section-lede">
The paper's ablations show that contrastive learning is the decisive ingredient rather than a minor training
detail.
</p>
</div>
<div class="insight-grid">
<article class="insight-card">
<h3 class="title is-5">Contrastive learning beats other OE strategies</h3>
<p>
Outlier exposure helps, but the strongest results come from enforcing latent-space separation with a
contrastive loss instead of relying only on segmentation-style supervision or outlier likelihood
minimization.
</p>
</article>
<article class="insight-card">
<h3 class="title is-5">The objective transfers to existing NF models</h3>
<p>
On ALLO, adding the FlowCLAS contrastive objective improves FastFlow from 29.1 to 40.8 AUPRC and UFlow
from 20.7 to 48.7 AUPRC, showing the framework is a reusable upgrade rather than a one-off architecture.
</p>
</article>
<article class="insight-card">
<h3 class="title is-5">Strong features help, but they are not enough alone</h3>
<p>
Better pre-trained encoders improve the ceiling, yet the paper shows that rich features still need the
contrastive objective to avoid low-level shortcuts and achieve reliable object-level anomaly separation.
</p>
</article>
</div>
</div>
</section>
<section class="section visualization-section">
<div class="container is-max-desktop">
<div class="columns is-centered">
<div class="column">
<div class="has-text-centered visualization-header">
<h2 class="title is-3">🖼️ Visualizations</h2>
</div>
<div class="visualization-group">
<h3 class="title is-5 visualization-group-title">Space anomaly segmentation</h3>
<figure class="visualization-figure">
<img
class="visualization-image"
src="./static/images/allo_1.png"
alt="ALLO test set, sample 1">
<figcaption class="visualization-caption">
<strong>ALLO</strong> qualitative result on a held-out orbital scene. The visualization highlights
FlowCLAS's ability to localize an anomalous object while preserving the surrounding ISS structure.
</figcaption>
</figure>
<figure class="visualization-figure">
<img
class="visualization-image"
src="./static/images/allo_2.png"
alt="ALLO test set, sample 2">
<figcaption class="visualization-caption">
<strong>ALLO</strong> qualitative result under a second viewpoint and lighting condition, illustrating
the method's transfer to dynamic space robotics imagery.
</figcaption>
</figure>
</div>
<div class="visualization-group">
<h3 class="title is-5 visualization-group-title">Road anomaly segmentation</h3>
<figure class="visualization-figure">
<img
class="visualization-image"
src="./static/images/fishyscapes.png"
alt="Road anomaly segmentation from the Fishyscapes Lost and Found dataset">
<figcaption class="visualization-caption">
<strong>Fishyscapes Lost & Found</strong> validation example showing dense anomaly localization in a
cluttered road scene with small hazardous objects.
</figcaption>
</figure>
<figure class="visualization-figure">
<img
class="visualization-image"
src="./static/images/roadanomaly.png"
alt="Road anomaly segmentation from the Road Anomaly dataset">
<figcaption class="visualization-caption">
<strong>Road Anomaly</strong> validation example illustrating object-level segmentation in a diverse
urban scene with strong appearance variation.
</figcaption>
</figure>
</div>
</div>
</div>
</div>
</section>
<section class="section" id="Citation">
<div class="container is-max-desktop content">
<h2 class="title">Citation</h2>
<pre><code>@inproceedings{leeFlowCLAS2026,
author = {Lee, Chang Won and Leveugle, Selina and Grouchy, Paul and Langley, Chris and Stolpner, Svetlana and Kelly, Jonathan and Waslander, Steven L.},
title = {FlowCLAS: Enhancing Normalizing Flow-Based Anomaly Segmentation Via Contrastive Learning},
booktitle = {Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)},
year = {2026},
}</code></pre>
</div>
</section>
<section class="section" id="Acknowledgements">
<div class="container is-max-desktop content">
<h2 class="title">Acknowledgements</h2>
<p>
This website is 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>
<p>
Thank you to the authors of <a href="https://github.com/nerfies/nerfies.github.io">Nerfies</a> for the
website template.
</p>
</div>
</section>
</body>
</html>