-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathindex.html
More file actions
484 lines (400 loc) · 14.6 KB
/
Copy pathindex.html
File metadata and controls
484 lines (400 loc) · 14.6 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
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
<!DOCTYPE html>
<html>
<head>
<meta charset="utf-8">
<!-- Metadata -->
<meta name="description" content="PEPR: Privileged Event-based Predictive Regularization for Domain Generalization">
<meta name="robots" content="index, follow">
<meta name="keywords" content="PEPR, Event Cameras, RGB-Event, Domain Generalization, LUPI, Privileged Information, Predictive Regularization, Computer Vision">
<meta property="og:title" content="PEPR: Privileged Event-based Predictive Regularization for Domain Generalization">
<meta property="og:description" content="Training RGB models with event-camera privileged information for robust domain generalization.">
<meta property="og:url" content="https://miccunifi.github.io/PEPR/">
<meta property="og:image" content="static/images/stage_2.png">
<meta property="og:image:width" content="1200">
<meta property="og:image:height" content="630">
<meta name="twitter:title" content="PEPR: Privileged Event-based Predictive Regularization">
<meta name="twitter:description" content="Events as privileged information for robust RGB-only domain generalization.">
<meta name="twitter:image" content="static/images/stage_2.png">
<meta name="twitter:card" content="summary_large_image">
<meta name="viewport" content="width=device-width, initial-scale=1">
<title>PEPR: Privileged Event-based Predictive Regularization</title>
<!-- Fonts -->
<link href="https://fonts.googleapis.com/css?family=Google+Sans|Noto+Sans|Castoro" rel="stylesheet">
<!-- Bulma / icons -->
<link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/bulma@0.9.4/css/bulma.min.css">
<link rel="stylesheet" href="https://cdn.jsdelivr.net/gh/jpswalsh/academicons@1/css/academicons.min.css">
<script defer src="https://use.fontawesome.com/releases/v5.15.4/js/all.js"></script>
<style>
body {
font-family: 'Noto Sans', sans-serif;
}
.publication-title {
font-family: 'Google Sans', sans-serif;
font-weight: 700;
line-height: 1.15;
}
.publication-title .pepr {
color: #6a2ca0;
}
.publication-authors {
margin-top: 1rem;
}
.author-block a {
color: #3273dc;
}
.conference {
margin-top: 0.75rem;
font-weight: 600;
color: #6a2ca0;
}
.publication-links {
margin-top: 1.5rem;
}
.link-block {
margin: 0.25rem;
display: inline-block;
}
.hero-teaser {
background: #f7f7f7;
padding-top: 2rem;
padding-bottom: 2rem;
}
.teaser-image {
max-width: 85%;
border-radius: 12px;
box-shadow: 0 10px 30px rgba(0,0,0,0.12);
background: white;
}
.section-title {
font-family: 'Google Sans', sans-serif;
font-weight: 700;
}
.tldr-box {
background: #f4ecfa;
border-left: 6px solid #6a2ca0;
border-radius: 10px;
padding: 1.25rem 1.5rem;
margin-top: 1.5rem;
margin-bottom: 1.5rem;
}
.highlight {
color: #6a2ca0;
font-weight: 700;
}
.card-box {
border-radius: 14px;
border: 1px solid #e5e5e5;
padding: 1.25rem;
height: 100%;
box-shadow: 0 4px 18px rgba(0,0,0,0.04);
}
.card-box h3 {
color: #6a2ca0;
font-weight: 700;
margin-bottom: 0.5rem;
}
.result-image {
width: 100%;
border-radius: 10px;
border: 1px solid #e5e5e5;
box-shadow: 0 6px 22px rgba(0,0,0,0.07);
margin-top: 1rem;
margin-bottom: 2rem;
}
.dataset-list li {
margin-bottom: 0.6rem;
}
pre {
background-color: #1f1f1f;
color: #f5f5f5;
border-radius: 10px;
padding: 1.25rem;
overflow-x: auto;
}
.footer {
font-size: 0.95rem;
}
</style>
</head>
<body>
<!-- Hero -->
<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">
<span class="pepr">PEPR:</span> Privileged Event-based Predictive Regularization<br>
for Domain Generalization
</h1>
<div class="is-size-5 publication-authors">
<span class="author-block">
<a href="#" target="_blank">Gabriele Magrini</a><sup>1</sup>,
</span>
<span class="author-block">
<a href="#" target="_blank">Federico Becattini</a><sup>2</sup>,
</span>
<span class="author-block">
<a href="#" target="_blank">Niccolò Biondi</a><sup>3</sup>,
</span>
<span class="author-block">
<a href="#" target="_blank">Pietro Pala</a><sup>1</sup>
</span>
</div>
<div class="is-size-5 publication-authors">
<span class="author-block">
<sup>1</sup>University of Florence
<sup>2</sup>University of Siena
<sup>3</sup>University of Trento
</span>
</div>
<div class="conference">
Accepted at CVPR 2026 Findings
</div>
<div class="publication-links">
<span class="link-block">
<a href="https://arxiv.org/pdf/2602.04583" target="_blank"
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/2602.04583" 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>
<span class="link-block">
<a href="https://github.com/miccunifi/PEPR" 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>
<span class="link-block">
<a href="#BibTeX"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="fas fa-quote-right"></i>
</span>
<span>Cite</span>
</a>
</span>
</div>
</div>
</div>
</div>
</div>
</section>
<!-- Teaser -->
<section class="section hero-teaser">
<div class="container has-text-centered">
<img src="./static/images/stage_2.png" alt="PEPR Method Overview" class="teaser-image">
</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 section-title">Abstract</h2>
<div class="content has-text-justified">
<p>
Deep neural networks for visual perception are highly susceptible to domain shift,
limiting their deployment under conditions that differ from the training data.
We address this problem through a cross-modal Learning Using Privileged Information
framework, where event cameras are available only during training and the final model
remains RGB-only at inference.
</p>
<p>
RGB streams are semantically dense but domain-dependent, while event streams are sparse
yet more domain-invariant. Direct feature alignment between these modalities is therefore
suboptimal, as it can force RGB representations to mimic sparse event features and lose
semantic detail. To overcome this, we introduce
<span class="highlight">Privileged Event-based Predictive Regularization</span>:
RGB features are trained to predict event-derived latent representations in a shared
feature space, transferring event robustness without direct alignment or input reconstruction.
</p>
<p>
PEPR improves robustness to day-to-night and adverse domain shifts across object detection
and semantic segmentation, while preserving a standard RGB-only inference pipeline.
</p>
</div>
<div class="tldr-box has-text-left">
<strong>TL;DR:</strong>
PEPR uses events only during training. Instead of aligning RGB and event features,
it makes RGB features predict event latents, improving robustness while keeping RGB-only inference.
</div>
</div>
</div>
</div>
</section>
<!-- Key Ideas -->
<section class="section hero is-light">
<div class="container is-max-desktop">
<h2 class="title is-3 has-text-centered section-title">Key Ideas</h2>
<div class="columns is-centered">
<div class="column">
<div class="card-box">
<h3>Prediction, not alignment</h3>
<p>
PEPR avoids forcing dense RGB features to directly match sparse event features.
Instead, the RGB encoder predicts event-derived latent targets.
</p>
</div>
</div>
<div class="column">
<div class="card-box">
<h3>Events as privileged information</h3>
<p>
Events are used only during training as an additional supervisory signal.
They are discarded after training.
</p>
</div>
</div>
<div class="column">
<div class="card-box">
<h3>RGB-only deployment</h3>
<p>
At test time, PEPR uses the original RGB model without event input,
additional sensors, or extra inference modules.
</p>
</div>
</div>
</div>
</div>
</section>
<!-- Method -->
<section class="section">
<div class="container is-max-desktop">
<h2 class="title is-3 has-text-centered section-title">Method</h2>
<div class="content has-text-justified">
<p>
During training, PEPR uses an RGB encoder, a task-specific prediction head, a privileged
event encoder, and a predictor. The RGB stream is optimized with the standard supervised task
loss, while the predictor maps RGB features to event latent patches produced by the event encoder.
The event encoder and predictor are removed at inference, leaving a robust RGB-only model.
</p>
</div>
<div class="has-text-centered">
<img src="./static/images/stage_2.png" alt="PEPR training and inference pipeline" class="result-image">
</div>
</div>
</section>
<!-- Results -->
<section class="section hero is-light">
<div class="container">
<h2 class="title is-3 has-text-centered section-title">Results</h2>
<div class="container is-max-desktop">
<div class="content has-text-centered">
<p>
PEPR improves RGB-only detection and segmentation robustness under domain shift,
outperforming direct alignment-based regularization.
</p>
</div>
</div>
<div class="container is-max-desktop">
<h3 class="title is-4 has-text-centered">Semantic Segmentation</h3>
<img src="./static/images/segmentation.png" alt="Semantic segmentation results" class="result-image">
<h3 class="title is-4 has-text-centered">Object Detection — FRED</h3>
<img src="./static/images/fred.png" alt="FRED object detection results" class="result-image">
<h3 class="title is-4 has-text-centered">Object Detection — Hard-DSEC</h3>
<img src="./static/images/harddsec.png" alt="Hard-DSEC object detection results" class="result-image">
</div>
</div>
</section>
<!-- Datasets -->
<section class="section" id="datasets">
<div class="container is-max-desktop">
<h2 class="title is-3 has-text-centered section-title">Datasets</h2>
<div class="content">
<ul class="dataset-list">
<li>
<strong>FRED:</strong>
<a href="https://huggingface.co/datasets/GabrieleMagrini/FRED" target="_blank">
Hugging Face dataset
</a>
</li>
<li>
<strong>DSEC:</strong>
<a href="https://dsec.ifi.uzh.ch/" target="_blank">
official dataset website
</a>
</li>
<li>
<strong>Hard-DSEC-DET:</strong>
<a href="https://github.com/djessy1998/EA-DETR" target="_blank">
EA-DETR repository
</a>
</li>
<li>
<strong>Cityscapes:</strong>
<a href="https://www.cityscapes-dataset.com/" target="_blank">
official website
</a>
</li>
<li>
<strong>Cityscapes Adverse:</strong>
<a href="https://huggingface.co/datasets/naufalso/cityscape-adverse" target="_blank">
Hugging Face dataset
</a>
</li>
</ul>
<p>
To simulate the event version of Cityscapes, please refer to the official
<a href="https://github.com/uzh-rpg/rpg_vid2e" target="_blank">VID2E</a> repository.
</p>
</div>
</div>
</section>
<!-- Code -->
<section class="section hero is-light">
<div class="container is-max-desktop has-text-centered">
<h2 class="title is-3 section-title">Code</h2>
<p>
Code for semantic segmentation and object detection is coming soon.
</p>
<p>
<a href="https://github.com/miccunifi/PEPR" target="_blank"
class="button is-dark is-rounded">
<span class="icon">
<i class="fab fa-github"></i>
</span>
<span>GitHub Repository</span>
</a>
</p>
</div>
</section>
<!-- BibTeX -->
<section class="section" id="BibTeX">
<div class="container is-max-desktop content">
<h2 class="title section-title">BibTeX</h2>
<pre><code>@inproceedings{magrini2026pepr,
title={PEPR: Privileged Event-based Predictive Regularization for Domain Generalization},
author={Magrini, Gabriele and Becattini, Federico and Biondi, Niccol{\`o} and Pala, Pietro},
booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
year={2026}
}</code></pre>
</div>
</section>
<!-- Footer -->
<footer class="footer">
<div class="container">
<div class="content has-text-centered">
<p>
This website template follows the academic project page style inspired by
<a href="https://nerfies.github.io/" target="_blank">Nerfies</a>.
</p>
</div>
</div>
</footer>
</body>
</html>