-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathindex.html
More file actions
587 lines (536 loc) · 27.2 KB
/
index.html
File metadata and controls
587 lines (536 loc) · 27.2 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
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1">
<!-- Primary Meta Tags -->
<!-- TODO: Replace with your paper title and author names -->
<meta name="title" content="SCNP Same Class Neighbor Penalization - Towards High-Quality Image Segmentation: Improving Topology Accuracy by Penalizing Neighbor Pixels - Juan Miguel Valverde, Dim P. Papadopoulos, Rasmus Larsen, Anders Dahl">
<!-- TODO: Write a compelling 150-160 character description of your research -->
<meta name="description" content="SCNP Same Class Neighbor Penalization - Towards High-Quality Image Segmentation: Improving Topology Accuracy by Penalizing Neighbor Pixels">
<!-- TODO: Add 5-10 relevant keywords for your research area -->
<meta name="keywords" content="scnp, topology, image segmentation, machine learning, computer vision, deep learning, AI">
<!-- TODO: List all authors -->
<meta name="author" content="Juan Miguel Valverde, Dim P. Papadopoulos, Rasmus Larsen, Anders Dahl">
<meta name="robots" content="index, follow">
<meta name="language" content="English">
<!-- Open Graph / Facebook -->
<meta property="og:type" content="article">
<!-- TODO: Replace with your institution or lab name -->
<meta property="og:site_name" content="Technical University of Denmark">
<!-- TODO: Same as paper title above -->
<meta property="og:title" content="SCNP Same Class Neighbor Penalization - Towards High-Quality Image Segmentation: Improving Topology Accuracy by Penalizing Neighbor Pixels">
<!-- TODO: Same as description above -->
<meta property="og:description" content="SCNP Same Class Neighbor Penalization - Towards High-Quality Image Segmentation: Improving Topology Accuracy by Penalizing Neighbor Pixels">
<!-- TODO: Replace with your actual website URL -->
<meta property="og:url" content="https://jmlipman.github.io/SCNP-SameClassNeighborPenalization/">
<!-- TODO: Create a 1200x630px preview image and update path -->
<meta property="og:image" content="https://jmlipman.github.io/SCNP-SameClassNeighborPenalization/static/images/social_preview.png">
<meta property="og:image:width" content="1200">
<meta property="og:image:height" content="630">
<meta property="og:image:alt" content="Towards High-Quality Image Segmentation: Improving Topology Accuracy by Penalizing Neighbor Pixels - Research Preview">
<meta property="article:published_time" content="2026-06-01T00:00:00.000Z">
<meta property="article:author" content="Juan Miguel Valverde">
<meta property="article:section" content="Research">
<meta property="article:tag" content="Topology">
<meta property="article:tag" content="Image Segmentation">
<!-- Academic/Research Specific -->
<meta name="citation_title" content="Towards High-Quality Image Segmentation: Improving Topology Accuracy by Penalizing Neighbor Pixels">
<meta name="citation_author" content="Valverde, Juan Miguel">
<meta name="citation_author" content="Papadopoulos, Dim P.">
<meta name="citation_author" content="Larsen, Rasmus">
<meta name="citation_author" content="Dahl, Anders">
<meta name="citation_publication_date" content="2026">
<meta name="citation_conference_title" content="CVPR">
<meta name="citation_pdf_url" content="https://jmlipman.github.io/SCNP-SameClassNeighborPenalization/static/pdfs/paper.pdf">
<!-- Additional SEO -->
<meta name="theme-color" content="#2563eb">
<meta name="msapplication-TileColor" content="#2563eb">
<meta name="apple-mobile-web-app-capable" content="yes">
<meta name="apple-mobile-web-app-status-bar-style" content="default">
<!-- Preconnect for performance -->
<link rel="preconnect" href="https://fonts.googleapis.com">
<link rel="preconnect" href="https://fonts.gstatic.com" crossorigin>
<link rel="preconnect" href="https://ajax.googleapis.com">
<link rel="preconnect" href="https://documentcloud.adobe.com">
<link rel="preconnect" href="https://cdn.jsdelivr.net">
<!-- TODO: Replace with your paper title and authors -->
<title>Towards High-Quality Image Segmentation: Improving Topology Accuracy by Penalizing Neighbor Pixels - Juan Miguel Valverde, Dim P. Papadopoulos, Rasmus Larsen, Anders Dahl</title>
<!-- Favicon and App Icons -->
<!--
<link rel="icon" type="image/x-icon" href="static/images/favicon.ico">
<link rel="apple-touch-icon" href="static/images/favicon.ico">
-->
<!-- Critical CSS - Load synchronously -->
<link rel="stylesheet" href="static/css/bulma.min.css">
<link rel="stylesheet" href="static/css/index.css">
<!-- Non-critical CSS - Load asynchronously -->
<link rel="preload" href="static/css/bulma-carousel.min.css" as="style" onload="this.onload=null;this.rel='stylesheet'">
<link rel="preload" href="static/css/bulma-slider.min.css" as="style" onload="this.onload=null;this.rel='stylesheet'">
<link rel="preload" href="static/css/fontawesome.all.min.css" as="style" onload="this.onload=null;this.rel='stylesheet'">
<link rel="preload" href="https://cdn.jsdelivr.net/gh/jpswalsh/academicons@1/css/academicons.min.css" as="style" onload="this.onload=null;this.rel='stylesheet'">
<!-- Fallback for browsers that don't support preload -->
<noscript>
<link rel="stylesheet" href="static/css/bulma-carousel.min.css">
<link rel="stylesheet" href="static/css/bulma-slider.min.css">
<link rel="stylesheet" href="static/css/fontawesome.all.min.css">
<link rel="stylesheet" href="https://cdn.jsdelivr.net/gh/jpswalsh/academicons@1/css/academicons.min.css">
</noscript>
<!-- Fonts - Optimized loading -->
<link href="https://fonts.googleapis.com/css2?family=Inter:wght@400;500;600;700;800&display=swap" rel="stylesheet">
<!-- Defer non-critical JavaScript -->
<script defer src="https://ajax.googleapis.com/ajax/libs/jquery/3.5.1/jquery.min.js"></script>
<script defer src="https://documentcloud.adobe.com/view-sdk/main.js"></script>
<script defer src="static/js/fontawesome.all.min.js"></script>
<script defer src="static/js/bulma-carousel.min.js"></script>
<script defer src="static/js/bulma-slider.min.js"></script>
<script defer src="static/js/index.js"></script>
<!-- Structured Data for Academic Papers -->
<script type="application/ld+json">
{
"@context": "https://schema.org",
"@type": "ScholarlyArticle",
"headline": "Towards High-Quality Image Segmentation: Improving Topology Accuracy by Penalizing Neighbor Pixels",
"description": "Towards High-Quality Image Segmentation: Improving Topology Accuracy by Penalizing Neighbor Pixels",
"author": [
{
"@type": "Person",
"name": "Juan Miguel Valverde",
"affiliation": {
"@type": "Organization",
"name": "Technical University of Denmark"
}
},
{
"@type": "Person",
"name": "Dim P. Papadopoulos",
"affiliation": {
"@type": "Organization",
"name": "Technical University of Denmark"
}
},
{
"@type": "Person",
"name": "Rasmus Larsen",
"affiliation": {
"@type": "Organization",
"name": "Technical University of Denmark"
}
},
{
"@type": "Person",
"name": "Anders Dahl",
"affiliation": {
"@type": "Organization",
"name": "Technical University of Denmark"
}
},
],
"datePublished": "2026-06-01",
"publisher": {
"@type": "Organization",
"name": "CVPR"
},
"url": "https://jmlipman.github.io/SCNP-SameClassNeighborPenalization",
"image": "https://YOUR_DOMAIN.com/static/images/social_preview.png",
"keywords": ["scnp", "deep learning", "topology", "image segmentation", "machine learning", "computer vision"],
"abstract": "FULL_ABSTRACT_TEXT_HERE",
"citation": "BIBTEX_CITATION_HERE",
"isAccessibleForFree": true,
"license": "https://creativecommons.org/licenses/by/4.0/",
"mainEntity": {
"@type": "WebPage",
"@id": "https://jmlipman.github.io/SCNP-SameClassNeighborPenalization"
},
"about": [
{
"@type": "Thing",
"name": "RESEARCH_AREA_1"
},
{
"@type": "Thing",
"name": "RESEARCH_AREA_2"
}
]
}
</script>
<!-- Website/Organization Structured Data -->
<script type="application/ld+json">
{
"@context": "https://schema.org",
"@type": "Organization",
"name": "Technical University of Denmark",
"url": "https://YOUR_INSTITUTION_WEBSITE.com",
"logo": "https://YOUR_DOMAIN.com/static/images/favicon.ico",
"sameAs": [
"https://twitter.com/YOUR_TWITTER_HANDLE",
"https://github.com/YOUR_GITHUB_USERNAME"
]
}
</script>
</head>
<body>
<!-- Scroll to Top Button -->
<button class="scroll-to-top" onclick="scrollToTop()" title="Scroll to top" aria-label="Scroll to top">
<i class="fas fa-chevron-up"></i>
</button>
<!-- More Works Dropdown -->
<div class="more-works-container">
<button class="more-works-btn" onclick="toggleMoreWorks()" title="View More Works from Our Lab">
<i class="fas fa-flask"></i>
More Works
<i class="fas fa-chevron-down dropdown-arrow"></i>
</button>
<div class="more-works-dropdown" id="moreWorksDropdown">
<!--<div class="dropdown-header">
<h4>More Works from Our Lab</h4>
<button class="close-btn" onclick="toggleMoreWorks()">
<i class="fas fa-times"></i>
</button>
</div>-->
<div class="works-list">
<!-- TODO: Replace with your lab's related works -->
<a href="https://jmlipman.github.io/TopoMortar" class="work-item" target="_blank">
<div class="work-info">
<!-- TODO: Replace with actual paper title -->
<h5>TopoMortar: A dataset to evaluate image segmentation methods focused on topology accuracy</h5>
<!-- TODO: Replace with brief description -->
<!--<p>A dataset to.</p>-->
<!-- TODO: Replace with venue and year -->
<span class="work-venue">BMVC 2025 (<b>Oral</b>)</span>
</div>
<i class="fas fa-external-link-alt"></i>
</a>
<!-- TODO: Add more related works or remove extra items -->
<!--
<a href="https://arxiv.org/abs/PAPER_ID_2" class="work-item" target="_blank">
<div class="work-info">
<h5>Paper Title 2</h5>
<p>Brief description of the work and its main contribution.</p>
<span class="work-venue">Conference/Journal 2023</span>
</div>
<i class="fas fa-external-link-alt"></i>
</a>
<a href="https://arxiv.org/abs/PAPER_ID_3" class="work-item" target="_blank">
<div class="work-info">
<h5>Paper Title 3</h5>
<p>Brief description of the work and its main contribution.</p>
<span class="work-venue">Conference/Journal 2023</span>
</div>
<i class="fas fa-external-link-alt"></i>
</a>
-->
</div>
</div>
</div>
<main id="main-content">
<section class="hero">
<div class="hero-body">
<div class="container is-max-desktop">
<div class="columns is-centered">
<div class="column has-text-centered">
<!-- TODO: Replace with your paper title -->
<h1 class="title is-1 publication-title">Towards High-Quality Image Segmentation: Improving Topology Accuracy by Penalizing Neighbor Pixels</h1>
<div class="is-size-5 publication-authors">
<div class="centered">
<div class="author-row">
<div class="col-author text-center">
<a href="https://jmlipman.github.io" target="_blank"><img src="static/images/miguel.jpg" alt="Juan Miguel Valverde" class="" style="border-radius: 100%">
<p>Juan Miguel Valverde</p></a>
</div>
<div class="col-author text-center">
<a href="https://dimipapa.github.io/" target="_blank"><img src="static/images/dim.png" alt="Dim P. Papadopoulos" class="">
<p>Dim P. Papadopoulos</p></a>
</div>
<div class="col-author text-center">
<a href="https://orbit.dtu.dk/en/persons/rasmus-larsen/" target="_blank"><img src="static/images/rasmus.png" alt="Rasmus Larsen" class="">
<p>Rasmus Larsen</p></a>
</div>
<div class="col-author text-center">
<a href="http://people.compute.dtu.dk/abda/" target="_blank"><img src="static/images/anders.png" alt="Anders Bjorholm Dahl" class="">
<p>Anders Bjorholm Dahl</p></a>
</div>
</div>
<div class="publication-authors">
<span class="affiliation-block"><img src="static/images/dtu_logo.png" style="height:25px; padding-right:5px">Department of Applied Mathematics and Computer Science, Technical University of Denmark</span>
</div>
</div> <!-- from the "centered" -->
</div> <!-- from the "is-size-5 publication-authors that contains the authors' names -->
<div class="column has-text-centered">
<div class="publication-links">
<!-- TODO: Update with your arXiv paper ID -->
<span class="link-block">
<a href="https://arxiv.org/abs/2603.18671" 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>
<!-- TODO: Add your supplementary material PDF or remove this section -->
<span class="link-block">
<a href="https://arxiv.org/abs/2603.18671" target="_blank"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="fas fa-file-pdf"></i>
</span>
<span>Supplementary</span>
</a>
</span>
<!-- TODO: Replace with your GitHub repository URL -->
<span class="link-block">
<a href="https://github.com/jmlipman/SCNP-SameClassNeighborPenalization" 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>
<!-- TODO: Update with your arXiv paper ID -->
<span class="link-block">
<a href="https://arxiv.org/abs/2603.18671" 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>
</div>
</div>
<div class="column has-text-centered is-size-5">
<b>Accepted to <font color="red">CVPR 2026</font></b>
</div>
</div>
</div>
</div>
</div>
</section>
<!-- Teaser video-->
<section class="hero teaser" style="margin-top:-40px">
<div class="container is-max-desktop">
<div class="hero-body">
<h2 class="subtitle has-text-centered" style="font-weight: bold">
Summary: SCNP discourages trained models from generating topological errors, such as clusters of false positives, thereby improving topology accuracy. This is achieved by penalizing the poorest-classified neighbor of each logit during training.
</h2>
<img src="static/images/teaser.png" width="500px" style="display:flex; margin: auto"/>
</div>
</div>
</section>
<!-- End teaser video -->
<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">Image segmentation quality and Topology accuracy</h2>
<div class="content has-text-justified">
<p>
In image segmentation, performance is typically measured with pixel-level metrics, like Dice coefficient and accuracy.
However, these metrics alone tell very little about image segmentation quality.
Topology accuracy (i.e., the accuracy in the number of connected components and holes between the prediction and the ground truth) is a good indicator of image segmentation quality, particularly when the downstream tasks rely on precise counts of objects or structures and their connectivity.
For example, the connectivity of the roads in satellite images or in blood vessels is more important than their accuracy at the border.
</p>
<table width="100%"><tr><td width="100%"><img src="static/images/motivation_topologyaccuracy.png"></td></tr></table>
</div>
</div>
</div>
</div>
</section>
<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">SCNP — Same Class Neighbor Penalization</h2>
<div class="content has-text-justified">
<p>
We propose SCNP, a method to improve topology accuracy that is <b>fast</b>, <b>efficient</b>, works with <b>any kind of structures</b>, and it's <b>easy to use and incorporate</b> into existing training pipelines.
SCNP is applied to the logits, allowing you to use your favorite architecture, optimization strategy, data augmentation, and loss function.
In practice, this means that you only need to add a few (three) lines to your code to use SCNP.
</p>
<table width="100%"><tr><td width="100%"><img src="static/images/scnp_pipeline.png"></td></tr></table>
</div>
</div>
</div>
</div>
</section>
<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">How does SCNP work?</h2>
<div class="content has-text-justified">
<!--<p>
TODO
</p>-->
<!--<table width="100%"><tr><td width="100%"><img src="static/images/motivation_topologyaccuracy.png"></td></tr></table>-->
<video poster="" id="tree" autoplay controls muted loop height="500px" preload="metadata">
<source src="static/videos/scnp_teaser.mp4" type="video/mp4">
</video>
</div>
</div>
</div>
</div>
</section>
<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">Effect on the loss and the gradients</h2>
<div class="content has-text-justified">
<p>
SCNP has <b>three main effects</b> on the loss and the gradients.
</p>
<video poster="" id="tree" autoplay controls muted loop height="500px" preload="metadata">
<source src="static/videos/scnp_effects.mp4" type="video/mp4">
</video>
</div>
</div>
</div>
</div>
</section>
<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">Implementation</h2>
<div class="content has-text-justified">
<p>
SCNP can be implemented in literally three lines of code. These need to be executed <u>only during training</u>:
</p>
<div style="background-color: #f5f5f5; padding: 15px; border-radius: 5px; font-family: 'Consolas', 'Monaco', 'Courier New', monospace; overflow-x: auto;">
<pre style="margin: 0; line-height: 1.5;">
<code>
<span style="color: #0000ff;">logits</span> = model(X)
<span style="color: #008000;">### vvv SCNP vvv ### </span>
<span style="color: #008000;"># MinPooling in the foreground</span>
<span style="color: #0000ff;">t1</span> = -torch.nn.functional.max_pool2d(-(<span style="color: #0000ff;">logits</span>*<span style="color: #0000ff;">y_onehot</span>+<span style="color: #800080;">9999</span>*(<span style="color: #800080;">1</span>-<span style="color: #0000ff;">y_onehot</span>)), kernel_size=<span style="color: #800080;">3</span>, stride=<span style="color: #800080;">1</span>, padding=<span style="color: #800080;">1</span>)
<span style="color: #008000;"># MaxPooling in the background</span>
<span style="color: #0000ff;">t2</span> = torch.nn.functional.max_pool2d((<span style="color: #0000ff;">logits</span>*(<span style="color: #800080;">1</span>-<span style="color: #0000ff;">y_onehot</span>)-<span style="color: #800080;">9999</span>*<span style="color: #0000ff;">y_onehot</span>), kernel_size=<span style="color: #800080;">3</span>, stride=<span style="color: #800080;">1</span>, padding=<span style="color: #800080;">1</span>)
<span style="color: #008000;"># Combining 't1' and 't2'</span>
<span style="color: #0000ff;">z_tilde</span> = <span style="color: #0000ff;">t1</span>*<span style="color: #0000ff;">y_onehot</span> + <span style="color: #0000ff;">t2</span>*(<span style="color: #800080;">1</span>-<span style="color: #0000ff;">y_onehot</span>)
<span style="color: #008000;">### ^^^ SCNP ^^^ ### </span>
<span style="color: #0000ff;">loss</span> = YourFavouriteLoss(SoftmaxOrSigmoid(<span style="color: #0000ff;">z_tilde</span>), Y)
</code>
</pre>
</div>
<!--<table width="100%"><tr><td width="100%"><img src="static/images/motivation_topologyaccuracy.png"></td></tr></table>-->
</div>
</div>
</div>
</div>
</section>
<!-- Image carousel -->
<section class="hero is-small">
<div class="hero-body">
<div class="container">
<h2 class="title is-3">Experiments and results</h2>
<div class="content has-text-justified">
<p>
We conducted 1) a benchmark across 13 datasets comparing 5-7 loss functions vs. Cross Entropy Dice loss with our SCNP; 2) a sensitivity analysis on SCNP's only hyper-parameter (neighborhood size); and 3) an ablation study comparing nine loss functions with and without SCNP.
Each experiment was run with five different random seeds.
In the benchmark, we employed medical and non-medical datasets, datasets with tubular and non-tubular structures, semantic and instance, binary and multi-class segmentation tasks, and three state-of-the-art DL frameworks (nnUNetv2, Detectron2, InstanSeg).
</p>
</div>
<div id="results-carousel" class="carousel results-carousel">
<div class="item">
<!-- 1) Benchmark (results), 2) Qualitative results, 3) Sensitivity analysis, 4) Ablation
-->
<!-- TODO: Replace with your research result images -->
<img src="static/images/results_dataset.png" alt="Results on the 13 datasets" loading="lazy"/>
<!-- TODO: Replace with description of this result -->
<h2 class="subtitle has-text-centered">
SCNP improved topology accuracy without deteriorating Dice coefficient.
</h2>
</div>
<div class="item">
<!-- Your image here -->
<img src="static/images/results_qualitative.png" alt="Qualitative results" loading="lazy"/>
<h2 class="subtitle has-text-centered">
Qualitative results.
</h2>
</div>
<div class="item">
<!-- Your image here -->
<img src="static/images/sensitivity_analysis.png" alt="Sensitivity analysis" loading="lazy"/>
<h2 class="subtitle has-text-centered">
Our sensitivity analysis indicates that the optimal neighborhood size of SCNP is correlated with the structure size.
</h2>
</div>
<div class="item">
<!-- Your image here -->
<img src="static/images/ablation_study.png" alt="Ablation study" loading="lazy"/>
<h2 class="subtitle has-text-centered">
Our ablation study demonstrates that SCNP improves topology accuracy with different (topology and non-topology) loss functions.
</h2>
</div>
</div>
</div>
</div>
</section>
<!-- End image carousel -->
<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">Learn more</h2>
<div class="content has-text-justified">
<p>
Read our paper to learn more about:
</p>
<ul>
<li>Topology loss functions.</li>
<li>SCNP gradients.</li>
<li>SCNP's limitation on datasets with extremely small structures (ISLES24 and MSLesSeg datasets).</li>
<li>Complete description of the datasets and setup used in our experiments.</li>
<li>SCNP's performance compared to simple dilation/erosion.</li>
<li>Extra computational resources required by SCNP (in the order of ms and GPU's MiB).</li>
</ul>
<!--<table width="100%"><tr><td width="100%"><img src="static/images/motivation_topologyaccuracy.png"></td></tr></table>-->
</div>
</div>
</div>
</div>
</section>
<!--BibTex citation -->
<section class="section" id="BibTeX">
<div class="container is-max-desktop content">
<div class="bibtex-header">
<h2 class="title">BibTeX</h2>
<button class="copy-bibtex-btn" onclick="copyBibTeX()" title="Copy BibTeX to clipboard">
<i class="fas fa-copy"></i>
<span class="copy-text">Copy</span>
</button>
</div>
<pre id="bibtex-code"><code>@article{valverde2026scnp,
title={Towards High-Quality Image Segmentation: Improving Topology Accuracy by Penalizing Neighbor Pixels},
author={Valverde, Juan Miguel and Papadopoulos, Dim P. and Larsen, Rasmus and Dahl, Anders},
journal={Proceedings of the IEEE/CVF conference on Computer Vision and Pattern Recognition},
pages={XX--YY},
year={2026},
url={https://jmlipman.github.io/SCNP-SameClassNeighborPenalization}
}</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>
This page was built using the <a href="https://github.com/eliahuhorwitz/Academic-project-page-template" target="_blank">Academic Project Page Template</a> which was adopted from the <a href="https://nerfies.github.io" target="_blank">Nerfies</a> project page.
You are free to borrow the source code of this website, we just ask that you link back to this page in the footer. <br> This website is licensed under a <a rel="license" href="http://creativecommons.org/licenses/by-sa/4.0/" target="_blank">Creative
Commons Attribution-ShareAlike 4.0 International License</a>.
</p>
</div>
</div>
</div>
</div>
</footer>
<!-- Statcounter tracking code -->
<!-- You can add a tracker to track page visits by creating an account at statcounter.com -->
<!-- End of Statcounter Code -->
</body>
</html>