-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathindex.html
More file actions
555 lines (533 loc) · 42 KB
/
Copy pathindex.html
File metadata and controls
555 lines (533 loc) · 42 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
<!DOCTYPE html>
<html lang="en-US">
<head>
<meta charset="UTF-8">
<meta http-equiv="X-UA-Compatible" content="IE=edge">
<meta name="viewport" content="width=device-width, initial-scale=1">
<title>Robin Chan, Dr. rer. nat.</title>
<meta name="description" content="Robin Chan | Dr. rer. nat. Mathematics | Bielefeld University | Machine Learning Group"/>
<link href="https://fonts.googleapis.com/css?family=Montserrat:400,700,200" rel="stylesheet">
<link href="https://maxcdn.bootstrapcdn.com/font-awesome/latest/css/font-awesome.min.css" rel="stylesheet">
<link rel="stylesheet" href="https://cdn.jsdelivr.net/gh/jpswalsh/academicons@1/css/academicons.min.css">
<link href="css/aos.css" rel="stylesheet">
<link href="css/bootstrap.min.css" rel="stylesheet">
<link href="styles/main.css" rel="stylesheet">
<link rel="shortcut icon" type="image/x-icon" href="favicon.ico">
</head>
<body id="top">
<header>
<div class="profile-page sidebar-collapse">
<nav class="navbar navbar-expand-lg fixed-top navbar-transparent bg-primary" color-on-scroll="400">
<div class="container">
<div class="navbar-translate"><a class="navbar-brand smooth-scroll" href="#top" rel="tooltip"><img src="images/logo-rchan.png"></a>
<button class="navbar-toggler navbar-toggler" type="button" data-toggle="collapse" data-target="#navigation" aria-controls="navigation" aria-expanded="false" aria-label="Toggle navigation"><span class="navbar-toggler-bar bar1"></span><span class="navbar-toggler-bar bar2"></span><span class="navbar-toggler-bar bar3"></span></button>
</div>
<div class="collapse navbar-collapse justify-content-end" id="navigation">
<ul class="navbar-nav">
<li class="nav-item"><a class="nav-link smooth-scroll" href="#top">Top ↑</a></li>
<li class="nav-item"><a class="nav-link smooth-scroll" href="#about">About</a></li>
<li class="nav-item"><a class="nav-link smooth-scroll" href="#research">Research</a></li>
<li class="nav-item"><a class="nav-link smooth-scroll" href="#experience">Experience</a></li>
<li class="nav-item"><a class="nav-link smooth-scroll" href="#education">Education</a></li>
<li class="nav-item"><a class="nav-link smooth-scroll" href="https://www.user.tu-berlin.de/chan/" target="_blank">One Pager</a></li>
<li class="nav-item"><a class="nav-link smooth-scroll" href="docs/cv.pdf" target="_blank">CV</a></li>
</ul>
</div>
</div>
</nav>
</div>
</header>
<div class="profile-page" id="top">
<div class="wrapper">
<div class="page-header page-header-small">
<div class="page-header-image" data-parallax="true" style="background-image: url('images/pp-bg.jpg');"></div>
<div class="container">
<div class="content-center">
<div class="cc-profile-image"><a class="navbar-brand smooth-scroll" href="#top" rel="tooltip"><img src="images/profile.jpg" alt="Image"/></a></div>
<div class="h2 title">Dr. Robin Chan</div>
<!-- <p class="category text-white">Dr. rer. nat. Mathematics</p> -->
<a class="btn btn-primary smooth-scroll mr-2" href="#about" data-aos="zoom-in" data-aos-anchor="data-aos-anchor">About</a>
</div>
</div>
</div>
</div>
</div>
<div class="section" id="about">
<div class="container">
<div class="card" data-aos="fade-up" data-aos-offset="10">
<div class="row">
<div class="col-lg-6 col-md-12">
<div class="card-body">
<div class="h4 mt-0 title">About</div>
<p>Welcome! My name is Robin Chan and I am a machine learning researcher at the <a href="https://www.uni-bielefeld.de/" target="_blank">Bielefeld University</a> and the <a href="https://www.tu.berlin/en/" target="_blank">Technische Universität Berlin</a> in Germany.
Currently, I am also a visiting researcher at the <a href="https://www.epfl.ch/en/" target=_blank>École Polytechnique Fédérale de Lausanne.</a></p>
<p>My main research interest lies in developing methods for safe AI, in particular for computer vision in the context of automated driving.
I am mainly focussing on detection techniques of unknown objects in semantic segmentation.
</p>
</div>
</div>
<div class="col-lg-6 col-md-12">
<div class="card-body">
<div class="h4 mt-0 title">Basic Information</div>
<div class="row">
<div class="col-sm-4"><strong class="text-uppercase">Age:</strong></div>
<div class="col-sm-8">30</div>
</div>
<div class="row mt-3">
<div class="col-sm-4"><strong class="text-uppercase">Email:</strong></div>
<div class="col-sm-8">
<script>
var part1 = "rchan";
var part2 = Math.pow(2,6);
var part3 = String.fromCharCode(part2);
var part4 = "techfak.uni-bielefeld.de";
var part5 = part1 + String.fromCharCode(part2) + part4;
document.write("<a href= " + "mai" + "lto" + ":" + part5
+ " >" + part5 + "</a>");
</script>
</div>
</div>
<div class="row mt-3">
<div class="col-sm-4"><strong class="text-uppercase">Main Office:</strong></div>
<div class="col-sm-8">CITEC 2-224 at Inspiration 1, 33619 Bielefeld, Germany</div>
</div>
<div class="row mt-3">
<div class="col-sm-4"><strong class="text-uppercase">Second Office:</strong></div>
<div class="col-sm-8">MA-5-567 at Straße des 17. Juni 136, 10623 Berlin , Germany</div>
</div>
</div>
</div>
</div>
</div>
<div class="h4 text-center mb-4 title">Follow Me</div>
<div class="row justify-content-center">
<a class="cc-facebook btn btn-link" href="https://github.com/robin-chan"><i class="fa fa-github fa-5x" aria-hidden="true"></i></a>
<a class="cc-facebook btn btn-link" href="https://scholar.google.com/citations?hl=de&user=MRCF9PwAAAAJ"><i class="ai ai-google-scholar-square ai-5x" aria-hidden="true"></i></a>
<a class="cc-facebook btn btn-link" href="https://www.linkedin.com/in/robin-chan-5b95b9248"><i class="fa fa-linkedin fa-5x" aria-hidden="true"></i></a>
<a class="cc-facebook btn btn-link" href=" https://arxiv.org/search/?query=Chan%2C+Robin&searchtype=author&abstracts=show&order=-announced_date_first&size=50"><i class="ai ai-arxiv ai-5x" aria-hidden="true"></i></a>
<a class="cc-facebook btn btn-link" href="https://orcid.org/0000-0003-2935-3275"><i class="ai ai-orcid ai-5x" aria-hidden="true"></i></a>
<a class="cc-facebook btn btn-link" href="https://www.researchgate.net/profile/Robin-Chan"><i class="ai ai-researchgate ai-5x" aria-hidden="true"></i></a>
<a class="cc-facebook btn btn-link" href="https://www.semanticscholar.org/author/R.-Chan/4253866"><i class="ai ai-semantic-scholar-square ai-5x" aria-hidden="true"></i></a>
</div>
</div>
</div>
<div class="section" id="research">
<div class="container">
<div class="h4 text-center mb-4 title">Recent Research Projects</div>
<div class="row justify-content-center">
<div class="col-lg-5 col-md-6">
<br><img src="images/project9_ood_retrieval/Summary.png" alt="Image"/>
</div>
<div class="col-lg-5 col-md-6">
<br><img src="images/project9_ood_retrieval/OoD_Retrieval.png" alt="Image"/>
</div>
<div class="col-lg-12"><br>
<div class="card card-body" data-aos="fade-up" data-aos-anchor-placement="top-bottom">
<div class="h5">Have We Ever Encountered This Before? Retrieving Out-of-Distribution Road Obstacles from Driving Scenes</div>
<p><strong class="text-uppercase">Abstract. </strong>In the life cycle of highly automated systems operating in an open and dynamic environment, the ability to adjust to emerging challenges is crucial. For systems integrating data-driven AI-based components, rapid responses to deployment issues require fast access to related data for testing and reconfiguration. In the context of automated driving, this especially applies to road obstacles that were not included in the training data, commonly referred to as out-of-distribution (OoD) road obstacles. Given the availability of large uncurated recordings of driving scenes, a pragmatic approach is to query a database to retrieve similar scenarios featuring the same safety concerns due to OoD road obstacles. In this work, we extend beyond identifying OoD road obstacles in video streams and offer a comprehensive approach to extract sequences of OoD road obstacles using text queries, thereby proposing a way of curating a collection of OoD data for subsequent analysis. Our proposed method leverages the recent advances in OoD segmentation and multi-modal foundation models to identify and efficiently extract safety-relevant scenes from unlabeled videos. We present a first approach for the novel task of text-based OoD object retrieval, which addresses the question ''Have we ever encountered this before?''. </p>
<p><strong class="text-uppercase">Published </strong><i> in The IEEE Winter Conference on Applications of Computer Vision (WACV) 2024.</i> Available: <a href="https://arxiv.org/abs/2309.04302" target="_blank">https://arxiv.org/abs/2309.04302</a></p>
</div>
</div>
</div>
<br><br>
<div class="row justify-content-center">
<div class="col-lg-3 col-md-6 col-sm-6 col-6">
<br><img src="images/project8_lu_net/sampling_mnist.png" alt="Image"/>
</div>
<div class="col-lg-3 col-md-6 col-sm-6 col-6">
<br><img src="images/project8_lu_net/sampling_fashion.png" alt="Image"/>
</div>
<div class="col-lg-3 col-md-6 col-sm-6 col-6">
<br><img src="images/project8_lu_net/interpolation_mnist.png" alt="Image"/>
</div>
<div class="col-lg-3 col-md-6 col-sm-6 col-6">
<br><img src="images/project8_lu_net/interpolation_fashion.png" alt="Image"/>
</div>
<div class="col-lg-12"><br>
<div class="card card-body" data-aos="fade-up" data-aos-anchor-placement="top-bottom">
<div class="h5">LU-Net: Invertible Neural Networks Based on Matrix Factorization</div>
<p><strong class="text-uppercase">Abstract. </strong>LU-Net is a simple and fast architecture for invertible neural networks (INN) that is based on the factorization of quadratic weight matrices A=LU, where L is a lower triangular matrix with ones on the diagonal and U an upper triangular matrix. Instead of learning a fully occupied matrix A, we learn L and U separately. If combined with an invertible activation function, such layers can easily be inverted whenever the diagonal entries of U are different from zero. Also, the computation of the determinant of the Jacobian matrix of such layers is cheap. Consequently, the LU architecture allows for cheap computation of the likelihood via the change of variables formula and can be trained according to the maximum likelihood principle. In our numerical experiments, we test the LU-Net architecture as generative model on several academic datasets. We also provide a detailed comparison with conventional invertible neural networks in terms of performance, training as well as run time.</p>
<p><strong class="text-uppercase">Published </strong><i> in The IEEE International Joint Conference on Neural Networks (IJCNN) 2023.</i> Available: <a href="https://arxiv.org/abs/2302.10524" target="_blank">https://arxiv.org/abs/2302.10524</a></p>
</div>
</div>
</div>
<br><br>
<div class="row justify-content-center">
<div class="col-lg-6 col-md-6">
<br><img src="images/project7_what_should_ai_see/ID_009811_IMGID_0080_SEGID_0017_GT_11_ROI_1_.jpg" alt="Image"/>
</div>
<div class="col-lg-6 col-md-6">
<br><img src="images/project7_what_should_ai_see/survey_slider.png" alt="Image"/>
</div>
<div class="col-lg-12"><br>
<div class="card card-body" data-aos="fade-up" data-aos-anchor-placement="top-bottom">
<div class="h5">What should AI see? Using the Public's Opinion to Determine the Perception of an AI</div>
<p><strong class="text-uppercase">Abstract. </strong>Deep neural networks (DNN) have made impressive progress in the interpretation of image data, so that it is conceivable and to some degree realistic to use them in safety critical applications like automated driving. From an ethical standpoint, the AI algorithm should take into account the vulnerability of objects or subjects on the street that ranges from "not at all", e.g. the road itself, to "high vulnerability" of pedestrians. One way to take this into account is to define the cost of confusion of one semantic category with another and use cost-based decision rules for the interpretation of probabilities, which are the output of DNNs. However, it is an open problem how to define the cost structure, who should be in charge to do that, and thereby define what AI-algorithms will actually "see". As one possible answer, we follow a participatory approach and set up an online survey to ask the public to define the cost structure. We present the survey design and the data acquired along with an evaluation that also distinguishes between perspective (car passenger vs. external traffic participant) and gender. Using simulation based F-tests, we find highly significant differences between the groups. These differences have consequences on the reliable detection of pedestrians in a safety critical distance to the self-driving car. We discuss the ethical problems that are related to this approach and also discuss the problems emerging from human-machine interaction through the survey from a psychological point of view. Finally, we include comments from industry leaders in the field of AI safety on the applicability of survey based elements in the design of AI functionalities in automated driving.</p>
<p><strong class="text-uppercase">Published </strong><i> in AI and Ethics.</i> Available: <a href="https://arxiv.org/abs/2206.04776" target="_blank">https://arxiv.org/abs/2206.04776</a> Website: <a href="http://www-ai.math.uni-wuppertal.de/~chan/poll/" target="_blank">https://uni-w.de/1hx</a></p>
</div>
</div>
</div>
<br><br>
<div class="row justify-content-center">
<div class="col-lg-3 col-md-6 col-sm-6 col-6">
<br><img src="images/project6_segment_me/leopard0000_overlay.jpg" alt="Image"/>
</div>
<div class="col-lg-3 col-md-6 col-sm-6 col-6">
<br><img src="images/project6_segment_me/leopard0000_uncertainty_image_resynthesis.jpg" alt="Image"/>
</div>
<div class="col-lg-3 col-md-6 col-sm-6 col-6">
<br><img src="images/project6_segment_me/leopard0000_uncertainty_synboost.jpg" alt="Image"/>
</div>
<div class="col-lg-3 col-md-6 col-sm-6 col-6">
<br><img src="images/project6_segment_me/leopard0000_uncertainty_max_entropy.jpg" alt="Image"/>
</div>
<div class="col-lg-12"><br>
<div class="card card-body" data-aos="fade-up" data-aos-anchor-placement="top-bottom">
<div class="h5">SegmentMeIfYouCan: A Benchmark for Anomaly Segmentation</div>
<p><strong class="text-uppercase">Abstract. </strong>State-of-the-art semantic or instance segmentation deep neural networks (DNNs) are usually trained on a closed set of semantic classes. As such, they are ill-equipped to handle previously-unseen objects. However, detecting and localizing such objects is crucial for safety-critical applications such as perception for automated driving, especially if they appear on the road ahead. While some methods have tackled the tasks of anomalous or out-of-distribution object segmentation, progress remains slow, in large part due to the lack of solid benchmarks; existing datasets either consist of synthetic data, or suffer from label inconsistencies. In this paper, we bridge this gap by introducing the "SegmentMeIfYouCan" benchmark. Our benchmark addresses two tasks: Anomalous object segmentation, which considers any previously-unseen object category; and road obstacle segmentation, which focuses on any object on the road, may it be known or unknown. We provide two corresponding datasets together with a test suite performing an in-depth method analysis, considering both established pixel-wise performance metrics and recent component-wise ones, which are insensitive to object sizes. We empirically evaluate multiple state-of-the-art baseline methods, including several specifically designed for anomaly / obstacle segmentation, on our datasets as well as on public ones, using our benchmark suite. The anomaly and obstacle segmentation results show that our datasets contribute to the diversity and challengingness of both dataset landscapes.</p>
<p><strong class="text-uppercase">Published </strong><i> in The Thirty-fifth Conference on Neural Information Processing Systems (NeurIPS) Datasets and Benchmarks Track 2021.</i> Available: <a href="https://arxiv.org/abs/2104.14812" target="_blank">https://arxiv.org/abs/2104.14812</a> Website: <a href="https://www.segmentmeifyoucan.com/" target="_blank">https://segmentmeifyoucan.com</a> Video: <a href="https://slideslive.com/38969495/" target="_blank">https://slideslive.com/38969495</a></p>
</div>
</div>
</div>
<br><br>
<div class="row justify-content-center">
<div class="col-lg-3 col-md-6 col-sm-6 col-6">
<br><img src="images/project5_meta_ood/1.jpg" alt="Image"/>
</div>
<div class="col-lg-3 col-md-6 col-sm-6 col-6">
<br><img src="images/project5_meta_ood/2.jpg" alt="Image"/>
</div>
<div class="col-lg-3 col-md-6 col-sm-6 col-6">
<br><img src="images/project5_meta_ood/3.jpg" alt="Image"/>
</div>
<div class="col-lg-3 col-md-6 col-sm-6 col-6">
<br><img src="images/project5_meta_ood/4.jpg" alt="Image"/>
</div>
<div class="col-lg-12"><br>
<div class="card card-body" data-aos="fade-up" data-aos-anchor-placement="top-bottom">
<div class="h5">Entropy Maximization and Meta Classification for Out-Of-Distribution Detection in Semantic Segmentation</div>
<p><strong class="text-uppercase">Abstract. </strong>Deep neural networks (DNNs) for the semantic segmentation of images are usually trained to operate on a predefined closed set of object classes. This is in contrast to the "open world" setting where DNNs are envisioned to be deployed to. From a functional safety point of view, the ability to detect so-called "out-of-distribution" (OoD) samples, i.e., objects outside of a DNN's semantic space, is crucial for many applications such as automated driving. A natural baseline approach to OoD detection is to threshold on the pixel-wise softmax entropy. We present a two-step procedure that significantly improves that approach. Firstly, we utilize samples from the COCO dataset as OoD proxy and introduce a second training objective to maximize the softmax entropy on these samples. Starting from pretrained semantic segmentation networks we re-train a number of DNNs on different in-distribution datasets and consistently observe improved OoD detection performance when evaluating on completely disjoint OoD datasets. Secondly, we perform a transparent post-processing step to discard false positive OoD samples by so-called "meta classification". To this end, we apply linear models to a set of hand-crafted metrics derived from the DNN's softmax probabilities. In our experiments we consistently observe a clear additional gain in OoD detection performance, cutting down the number of detection errors by up to 52% when comparing the best baseline with our results. We achieve this improvement sacrificing only marginally in original segmentation performance. Therefore, our method contributes to safer DNNs with more reliable overall system performance.</p>
<p>
<strong class="text-uppercase">Published </strong><i>in The IEEE/CVF International Conference on Computer Vision (ICCV) 2021.</i>
Available: <a href="https://arxiv.org/abs/2012.06575" target="_blank">https://arxiv.org/abs/2012.06575</a>.
Video: <a href="https://youtu.be/jy1i5nKH3ps" target="_blank">https://youtu.be/jy1i5nKH3ps</a>.
Slides: <a href="http://www-ai.math.uni-wuppertal.de/~chan/talks/ICCV_Entropy_max/" target="_blank">https://uni-w.de/alf7s</a>
</p>
</div>
</div>
</div>
<br><br>
<div class="row justify-content-center">
<div class="col-lg-2 col-md-3 col-sm-5 col-6">
<br><img src="images/project3_metafusion/1_patch.png" alt="Image"/>
</div>
<div class="col-lg-2 col-md-3 col-sm-5 col-6">
<br><img src="images/project3_metafusion/2_patch.png" alt="Image"/>
</div>
<div class="col-lg-2 col-md-3 col-sm-5 col-6">
<br><img src="images/project3_metafusion/3_patch.png" alt="Image"/>
</div>
<div class="col-lg-2 col-md-3 col-sm-5 col-6">
<br><img src="images/project3_metafusion/4_patch.png" alt="Image"/>
</div>
<div class="col-lg-12"><br>
<div class="card card-body" data-aos="fade-up" data-aos-anchor-placement="top-bottom">
<div class="h5">Controlled False-Negative Reduction of Minority Classes in Semantic Segmentation</div>
<p><strong class="text-uppercase">Abstract. </strong>In semantic segmentation datasets, classes of high importance are oftentimes underrepresented, e.g., humans in street scenes. Neural networks are usually trained to reduce the overall number of errors, attaching identical loss to errors of all kinds. However, this is not necessarily aligned with human intuition. For instance, an overlooked pedestrian seems more severe than an incorrectly detected one. One possible remedy is to deploy different decision rules by introducing class priors that assign more weight to underrepresented classes. While reducing the false negatives of the underrepresented class, at the same time this leads to a considerable increase of false positive indications. In this work, we combine decision rules with methods for false positive detection. We therefore fuse false negative detection with uncertainty based false positive meta classification. We present the efficiency of our method for the semantic segmentation of street scenes on the Cityscapes dataset based on predicted instances of the 'human' class. In the latter we employ an advanced false positive detection method using uncertainty measures aggregated over instances. We thereby achieve improved trade-offs between false negative and false positive samples of the underrepresented classes.</p>
<p><strong class="text-uppercase">Published </strong><i>in The IEEE International Joint Conference on Neural Networks (IJCNN) 2020.</i> Available: <a href="https://arxiv.org/abs/1912.07420" target="_blank">https://arxiv.org/abs/1912.07420</a></p>
</div>
</div>
</div>
<br><br>
<div class="row justify-content-center">
<div class="col-lg-6 col-md-6">
<br><img src="images/project4_metaseg/metaseg1.jpg" alt="Image"/>
</div>
<div class="col-lg-6 col-md-6">
<br><img src="images/project4_metaseg/metaseg2.jpg" alt="Image"/>
</div>
<div class="col-lg-12"><br>
<div class="card card-body" data-aos="fade-up" data-aos-anchor-placement="top-bottom">
<div class="h5">Prediction Error Meta Classification in Semantic Segmentation: Detection via Aggregated Dispersion Measures of Softmax Probabilities</div>
<p><strong class="text-uppercase">Abstract. </strong>We present a method that "meta" classifies whether seg-ments predicted by a semantic segmentation neural network intersect with the ground truth. For this purpose, we employ measures of dispersion for predicted pixel-wise class probability distributions, like classification entropy, that yield heat maps of the input scene's size. We aggregate these dispersion measures segment-wise and derive metrics that are well-correlated with the segment-wise IoU of prediction and ground truth. This procedure yields an almost plug and play post-processing tool to rate the prediction quality of semantic segmentation networks on segment level. This is especially relevant for monitoring neural networks in online applications like automated driving or medical imaging where reliability is of utmost importance. In our tests, we use publicly available state-of-the-art networks trained on the Cityscapes dataset and the BraTS2017 dataset and analyze the predictive power of different metrics as well as different sets of metrics. To this end, we compute logistic LASSO regression fits for the task of classifying IoU=0 vs. IoU>0 per segment and obtain AUROC values of up to 91.55%. We complement these tests with linear regression fits to predict the segment-wise IoU and obtain prediction standard deviations of down to 0.130 as well as R^2 values of up to 84.15%. We show that these results clearly outperform standard approaches.</p>
<p><strong class="text-uppercase">Published </strong><i>in The IEEE International Joint Conference on Neural Networks (IJCNN) 2020.</i> Available: <a href="https://arxiv.org/abs/1811.00648" target="_blank">https://arxiv.org/abs/1811.00648</a></p>
</div>
</div>
</div>
<br><br>
<div class="row justify-content-center">
<div class="col-lg-6 col-md-6">
<br><img src="images/project1_max_like/INFO.png" alt="Image"/>
</div>
<div class="col-lg-6 col-md-6">
<br><img src="images/project1_max_like/PERSON.png" alt="Image"/>
</div>
<div class="col-lg-12"><br>
<div class="card card-body" data-aos="fade-up" data-aos-anchor-placement="top-bottom">
<div class="h5">Application of Decision Rules for Handling Class Imbalance in Semantic Segmentation</div>
<p><strong class="text-uppercase">Abstract. </strong>As part of autonomous car driving systems, semantic segmentation is an essential component to obtain a full understanding of the car's environment. One difficulty, that occurs while training neural networks for this purpose, is class imbalance of training data. Consequently, a neural network trained on unbalanced data in combination with maximum a-posteriori classification may easily ignore classes that are rare in terms of their frequency in the dataset. However, these classes are often of highest interest. We approach such potential misclassifications by weighting the posterior class probabilities with the prior class probabilities which in our case are the inverse frequencies of the corresponding classes in the training dataset. More precisely, we adopt a localized method by computing the priors pixel-wise such that the impact can be analyzed at pixel level as well. In our experiments, we train one network from scratch using a proprietary dataset containing 20,000 annotated frames of video sequences recorded from street scenes. The evaluation on our test set shows an increase of average recall with regard to instances of pedestrians and info signs by 25% and 23.4%, respectively. In addition, we significantly reduce the non-detection rate for instances of the same classes by 61% and 38%.</p>
<p><strong class="text-uppercase">Published </strong><i>in The 30th European Safety and Reliability Conference (ESREL) 2020. </i>Available: <a href="https://arxiv.org/abs/1901.08394" target="_blank">https://arxiv.org/abs/1901.08394</a></p>
</div>
</div>
</div>
<br><br>
<div class="row justify-content-center">
<div class="col-lg-6 col-md-6">
<br><img src="images/project2_ethical_dilemma/alt.jpg" alt="Image"/>
</div>
<div class="col-lg-6 col-md-6">
<br><img src="images/project2_ethical_dilemma/ego.jpg" alt="Image"/>
</div>
<div class="col-lg-12"><br>
<div class="card card-body" data-aos="fade-up" data-aos-anchor-placement="top-bottom">
<div class="h5">The Ethical Dilemma when (not) Setting up Cost-based Decision Rules in Semantic Segmentation</div>
<p><strong class="text-uppercase">Abstract. </strong>Neural networks for semantic segmentation can be seen as statistical models that provide for each pixel of one image a probability distribution on predefined classes. The predicted class is then usually obtained by the maximum a-posteriori probability (MAP) which is known as Bayes rule in decision theory. From decision theory we also know that the Bayes rule is optimal regarding the simple symmetric cost function. Therefore, it weights each type of confusion between two different classes equally, e.g., given images of urban street scenes there is no distinction in the cost function if the network confuses a person with a street or a building with a tree. Intuitively, there might be confusions of classes that are more important to avoid than others. In this work, we want to raise awareness of the possibility of explicitly defining confusion costs and the associated ethical difficulties if it comes down to providing numbers. We define two cost functions from different extreme perspectives, an egoistic and an altruistic one, and show how safety relevant quantities like precision / recall and (segment-wise) false positive / negative rate change when interpolating between MAP, egoistic and altruistic decision rules.</p>
<p><strong class="text-uppercase">Published </strong><i>in The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops 2019.</i> Available: <a href="https://arxiv.org/abs/1907.01342" target="_blank">https://arxiv.org/abs/1907.01342</a></p>
</div>
</div>
</div>
</div>
</div>
<div class="section" id="experience">
<div class="container cc-experience">
<div class="h4 text-center mb-4 title">Work Experience</div>
<div class="card">
<div class="row">
<div class="col-md-3 bg-primary" data-aos="fade-right" data-aos-offset="50" data-aos-duration="500">
<div class="card-body cc-experience-header">
<div class="h5">Since Sep 23</div>
</div>
</div>
<div class="col-md-9" data-aos="fade-left" data-aos-offset="50" data-aos-duration="500">
<div class="card-body">
<div class="h5">Visiting Postdoc</div>
<p class="category">École Polytechnique Fédérale de Lausanne | School of Computer and Communication Sciences | Computer Vision Laboratory | Lausanne, Switzerland</p>
</div>
</div>
</div>
</div>
<div class="card">
<div class="row">
<div class="col-md-3 bg-primary" data-aos="fade-right" data-aos-offset="50" data-aos-duration="500">
<div class="card-body cc-experience-header">
<div class="h5">Since Apr 23</div>
</div>
</div>
<div class="col-md-9" data-aos="fade-left" data-aos-offset="50" data-aos-duration="500">
<div class="card-body">
<div class="h5">Postdoctoral Researcher</div>
<p class="category">Technische Universität Berlin | Institute of Mathematics | Mathematical Modeling of Industrial Life Cycles | Berlin, Germany</p>
</div>
</div>
</div>
</div>
<div class="card">
<div class="row">
<div class="col-md-3 bg-primary" data-aos="fade-right" data-aos-offset="50" data-aos-duration="500">
<div class="card-body cc-experience-header">
<div class="h5">Since Aug 22</div>
</div>
</div>
<div class="col-md-9" data-aos="fade-left" data-aos-offset="50" data-aos-duration="500">
<div class="card-body">
<div class="h5">Postdoctoral Researcher</div>
<p class="category">Bielefeld University | Faculty of Technology | Machine Learning Group | Funded by the Ministry of Culture and Science of the German
State of North Rhine-Westphalia | Out-of-Distribution Detection via Generative Modeling of Deep Latent Representations | Bielefeld, Germany</p>
</div>
</div>
</div>
</div>
<div class="card">
<div class="row">
<div class="col-md-3 bg-primary" data-aos="fade-right" data-aos-offset="50" data-aos-duration="500">
<div class="card-body cc-experience-header">
<div class="h5">Jan 20 - Jun 22</div>
</div>
</div>
<div class="col-md-9" data-aos="fade-left" data-aos-offset="50" data-aos-duration="500">
<div class="card-body">
<div class="h5">Research Assistant</div>
<p class="category">University of Wuppertal | School of Mathematics and Natural Sciences | Stochastics Group | Funded by German Federal Ministry for Economic Affairs and Energy | KI Absicherung – Safe AI for Automated Driving | Wuppertal, Germany</p>
</div>
</div>
</div>
</div>
<div class="card">
<div class="row">
<div class="col-md-3 bg-primary" data-aos="fade-right" data-aos-offset="50" data-aos-duration="500">
<div class="card-body cc-experience-header">
<div class="h5">Nov 18 - Dec 19</div>
</div>
</div>
<div class="col-md-9" data-aos="fade-left" data-aos-offset="50" data-aos-duration="500">
<div class="card-body">
<div class="h5">Research Assistant</div>
<p class="category">University of Wuppertal | School of Mathematics and Natural Sciences | Applied Computer Science Group | Funded by Volkswagen Group Research | Wuppertal, Germany</p>
</div>
</div>
</div>
</div>
<div class="card">
<div class="row">
<div class="col-md-3 bg-primary" data-aos="fade-right" data-aos-offset="50" data-aos-duration="500">
<div class="card-body cc-experience-header">
<div class="h5">sep 17 - Oct 18</div>
</div>
</div>
<div class="col-md-9" data-aos="fade-left" data-aos-offset="50" data-aos-duration="500">
<div class="card-body">
<div class="h5">Intern</div>
<p class="category">Volkswagen Group Research | Automated Driving | Architecture and AI Technologies | Wolfsburg, Germany</p>
</div>
</div>
</div>
</div>
<div class="card">
<div class="row">
<div class="col-md-3 bg-primary" data-aos="fade-right" data-aos-offset="50" data-aos-duration="500">
<div class="card-body cc-experience-header">
<div class="h5">Oct 16 - Sep 17</div>
</div>
</div>
<div class="col-md-9" data-aos="fade-left" data-aos-offset="50" data-aos-duration="500">
<div class="card-body">
<div class="h5">Tutor</div>
<p class="category">University of Wuppertal | Mathematics for Economy Scientists | Mathematics for Civil Engineers | Wuppertal, Germany</p>
</div>
</div>
</div>
</div>
<div class="card">
<div class="row">
<div class="col-md-3 bg-primary" data-aos="fade-right" data-aos-offset="50" data-aos-duration="500">
<div class="card-body cc-experience-header">
<div class="h5">Apr 16 - Oct 16</div>
</div>
</div>
<div class="col-md-9" data-aos="fade-left" data-aos-offset="50" data-aos-duration="500">
<div class="card-body">
<div class="h5">Intern</div>
<p class="category">Siemens AG | Energy Sector | Large Gas Turbine Engineering | Probabilistic Design | Berlin, Germany</p>
</div>
</div>
</div>
</div>
</div>
</div>
<div class="section" id="education">
<div class="container cc-experience">
<div class="h4 text-center mb-4 title">Education</div>
<div class="card">
<div class="row">
<div class="col-md-3 bg-primary" data-aos="fade-right" data-aos-offset="50" data-aos-duration="500">
<div class="card-body cc-experience-header">
<div class="h5">Nov 18 - Apr 22</div>
</div>
</div>
<div class="col-md-9" data-aos="fade-left" data-aos-offset="50" data-aos-duration="500">
<div class="card-body">
<div class="h5">Doctoral Degree</div>
<p class="category">Mathematics | Doctor of Natural Sciences | University of Wuppertal</p>
</div>
</div>
</div>
</div>
<div class="card">
<div class="row">
<div class="col-md-3 bg-primary" data-aos="fade-right" data-aos-offset="50" data-aos-duration="500">
<div class="card-body cc-experience-header">
<div class="h5">Oct 16 - Sep 18</div>
</div>
</div>
<div class="col-md-9" data-aos="fade-left" data-aos-offset="50" data-aos-duration="500">
<div class="card-body">
<div class="h5">Master's Degree</div>
<p class="category">Mathematics | Master of Science | University of Wuppertal</p>
</div>
</div>
</div>
</div>
<div class="card">
<div class="row">
<div class="col-md-3 bg-primary" data-aos="fade-right" data-aos-offset="50" data-aos-duration="500">
<div class="card-body cc-experience-header">
<div class="h5">Oct 12 - Jan 17</div>
</div>
</div>
<div class="col-md-9" data-aos="fade-left" data-aos-offset="50" data-aos-duration="500">
<div class="card-body">
<div class="h5">Bachelor's Degree</div>
<p class="category">Mathematical Economics | Bachelor of Science | University of Wuppertal</p>
</div>
</div>
</div>
</div>
</div>
</div>
<div class="section" id="links">
<div class="container">
<div class="h4 text-center mb-4 title">Links</div>
<div class="text-center">
<div class="row">
<div class="col-lg-3 col-md-3 col-sm-3 col-xs-3 col-6 cc-link-image">
<a href="https://hammer-lab.techfak.uni-bielefeld.de">
<img src="images/logos/hammerlab.svg">
</a><br><br>
</div>
<div class="col-lg-3 col-md-3 col-sm-3 col-xs-3 col-6 cc-link-image">
<a href="https://www.mkw.nrw/hochschule-und-forschung/foerderungen/foerderlinie-kuenstliche-intelligenz-maschinelles-lernen">
<img src="images/logos/nrw.svg">
</a><br><br>
</div>
<div class="col-lg-3 col-md-3 col-sm-3 col-xs-3 col-6 cc-link-image">
<a href="https://www.izmd.uni-wuppertal.de/en/">
<img src="images/logos/izmd.png">
</a><br><br>
</div>
<div class="col-lg-3 col-md-3 col-sm-3 col-xs-3 col-6 cc-link-image">
<a href="https://www.ki-absicherung-projekt.de/en/">
<img src="images/logos/kia.png">
</a><br><br>
</div>
</div>
</div>
</div>
</div>
<footer class="footer">
<div class="container text-center">
<!-- <a class="cc-facebook btn btn-link" href="https://github.com/robin-chan"><i class="fa fa-github fa-5x" aria-hidden="true"></i></a>
<a class="cc-facebook btn btn-link" href="https://scholar.google.com/citations?hl=de&user=MRCF9PwAAAAJ"><i class="ai ai-google-scholar-square ai-5x" aria-hidden="true"></i></a>
<a class="cc-facebook btn btn-link" href=" https://arxiv.org/search/?query=Chan%2C+Robin&searchtype=author&abstracts=show&order=-announced_date_first&size=50"><i class="ai ai-arxiv ai-5x" aria-hidden="true"></i></a>
<a class="cc-facebook btn btn-link" href="https://www.semanticscholar.org/author/R.-Chan/4253866"><i class="ai ai-semantic-scholar-square ai-5x" aria-hidden="true"></i></a>
<a class="cc-facebook btn btn-link" href="https://www.researchgate.net/profile/Robin-Chan"><i class="ai ai-researchgate ai-5x" aria-hidden="true"></i></a>
<a class="cc-facebook btn btn-link" href="https://orcid.org/0000-0003-2935-3275"><i class="ai ai-orcid ai-5x" aria-hidden="true"></i></a> -->
<div class="h4 title text-center">Last updated on 22nd November 2023 by Robin Chan</div>
<div class="text-center text-muted">
Copyright © 2023 -
<!--Design <a class="credit" href="https://templateflip.com" target="_blank">TemplateFlip</a> - -->
<a class="credit" data-toggle="modal" data-target=".bd-example-modal-lg">Disclaimer</a>
<div class="modal fade bd-example-modal-lg" tabindex="-1" role="dialog" aria-hidden="true">
<div class="modal-dialog modal-lg">
<div class="modal-content">
<div class="h4 title text-center">Disclaimer</div>
<div class="card-body text-left">
The author makes every effort to ensure that the material contained in this web site is current, complete and correct. Nevertheless, errors and mistakes cannot be completely ruled out.
The author does not accept liability for the relevance, accuracy or completeness of the information and material offered in her/his web site unless the mistake occurred intentionally or through gross negligence. This refers to any loss, additional costs or damage of any kind suffered as a result of any use of any material in this web site.
The site links to third party information providers. The author has no influence on the contents or reliability of the web sites of these third party information providers. The following therefore applies to all links:
"The author has no influence on the design and contents of third party internet web sites. The author therefore expressly distances himself from the material of all third party internet web sites, even if the web site links to these external sites."
The declaration is valid for all links given on the author's web site and the material of all sites reached via banners and links on the author's web site.
</div>
</div>
</div>
</div>
</div>
</div>
</footer>
<script src="js/core/jquery.3.2.1.min.js"></script>
<script src="js/core/popper.min.js"></script>
<script src="js/core/bootstrap.min.js"></script>
<script src="js/now-ui-kit.js?v=1.1.0"></script>
<script src="js/aos.js"></script>
<script src="scripts/main.js"></script>
</body>
</html>