-
Notifications
You must be signed in to change notification settings - Fork 0
/
segmentation-of-cellular-structures.html
679 lines (589 loc) · 34.3 KB
/
segmentation-of-cellular-structures.html
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
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
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
<!DOCTYPE html>
<html>
<head>
<meta charset="utf-8">
<!-- Meta tags for social media banners, these should be filled in appropriatly as they are your "business card" -->
<!-- Replace the content tag with appropriate information -->
<meta name="description" content="DESCRIPTION META TAG">
<meta property="og:title" content="DEEP-EM: Segmentation of Cellular Structures" />
<meta property="og:description"
content="Unlock the power of Deep Learning in Electron Microscopy with the DEEP-EM TOOLBOX standardized workflows for EM image analysis." />
<meta property="og:url" content="URL OF THE WEBSITE" />
<!-- Path to banner image, should be in the path listed below. Optimal dimenssions are 1200X630-->
<meta property="og:image" content="static/image/your_banner_image.png" />
<meta property="og:image:width" content="1200" />
<meta property="og:image:height" content="630" />
<meta name="twitter:title" content="DEEP-EM: Segmentation of Cellular Structures">
<meta name="twitter:description"
content="Unlock the power of Deep Learning in Electron Microscopy with the DEEP-EM TOOLBOX standardized workflows for EM image analysis.">
<!-- Path to banner image, should be in the path listed below. Optimal dimenssions are 1200X600-->
<meta name="twitter:image" content="static/images/your_twitter_banner_image.png">
<meta name="twitter:card" content="summary_large_image">
<!-- Keywords for your paper to be indexed by-->
<meta name="keywords" content="Deep Learning, Electron Microscopy, Data Analysis, Data Interpretation, Toolbox">
<meta name="viewport" content="width=device-width, initial-scale=1">
<title>DEEP-EM: Segmentation of Cellular Structures</title>
<link rel="icon" type="image/x-icon" href="static/images/icon.png">
<link href="https://fonts.googleapis.com/css?family=Google+Sans|Noto+Sans|Castoro" rel="stylesheet">
<link rel="stylesheet" href="static/css/bulma.min.css">
<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">
<link rel="stylesheet" href="static/css/index.css">
<style>
.hidden {
display: none;
}
.button-55 {
align-self: center;
background-color: #fff;
background-image: none;
background-position: 0 90%;
background-repeat: repeat no-repeat;
background-size: 4px 3px;
border-radius: 15px 225px 255px 15px 15px 255px 225px 15px;
border-style: solid;
border-width: 2px;
box-shadow: rgba(0, 0, 0, .2) 15px 28px 25px -18px;
box-sizing: border-box;
color: #41403e;
cursor: pointer;
display: inline-block;
font-family: Neucha, sans-serif;
font-size: 1rem;
line-height: 23px;
outline: none;
padding: .75rem;
text-decoration: none;
transition: all 235ms ease-in-out;
border-bottom-left-radius: 15px 255px;
border-bottom-right-radius: 225px 15px;
border-top-left-radius: 255px 15px;
border-top-right-radius: 15px 225px;
user-select: none;
-webkit-user-select: none;
touch-action: manipulation;
}
.button-55:hover {
box-shadow: rgba(0, 0, 0, .3) 2px 8px 8px -5px;
transform: translate3d(0, 2px, 0);
}
.button-55:focus {
box-shadow: rgba(0, 0, 0, .3) 2px 8px 4px -6px;
}
.green-background {
background-color: #dde9afff;
/* Green background */
padding: 20px;
/* Padding inside the element */
border-radius: 10px;
/* Rounded corners */
box-shadow: 0 4px 8px rgba(0, 0, 0, 0.2);
/* Subtle shadow for depth */
font-size: 16px;
/* Font size */
margin: 20px 0;
/* Margin outside the element */
}
.red-background {
background-color: #ffaaaaff;
/* Green background */
padding: 20px;
/* Padding inside the element */
border-radius: 10px;
/* Rounded corners */
box-shadow: 0 4px 8px rgba(0, 0, 0, 0.2);
/* Subtle shadow for depth */
font-size: 16px;
/* Font size */
margin: 20px 0;
/* Margin outside the element */
}
.orange-background {
background-color: #ffb380ff;
/* Green background */
padding: 20px;
/* Padding inside the element */
border-radius: 10px;
/* Rounded corners */
box-shadow: 0 4px 8px rgba(0, 0, 0, 0.2);
/* Subtle shadow for depth */
font-size: 16px;
/* Font size */
margin: 20px 0;
/* Margin outside the element */
}
</style>
<script src="https://ajax.googleapis.com/ajax/libs/jquery/3.5.1/jquery.min.js"></script>
<script src="https://documentcloud.adobe.com/view-sdk/main.js"></script>
<script defer src="static/js/fontawesome.all.min.js"></script>
<script src="static/js/bulma-carousel.min.js"></script>
<script src="static/js/bulma-slider.min.js"></script>
<script src="static/js/index.js"></script>
<script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-mml-chtml.js"></script>
<script>
function toggleVisibility(id_content, id_button) {
var content = document.getElementById(id_content);
var btn = document.getElementById(id_button);
if (content.classList.contains('hidden')) {
content.classList.remove('hidden');
btn.innerHTML = "Show Less"
} else {
content.classList.add('hidden');
btn.innerHTML = "Show More"
}
}
</script>
</head>
<body>
<section class="hero">
<div class="hero-body">
<div class="container is-max-desktop">
<div class="columns is-centered">
<div class="column has-text-centered">
<!--<img src="static/images/icon.png"
alt="Schematic showing 3 differnt types of task applicable for deep learning. (image to values, image to image & 2D to 3D)" />
-->
<h1 class="title is-1 publication-title">Image to Image</h1>
<!--Add use case category (one of: Image to Value(s), Image to Image, 2D to 3D)-->
<h2 class="title is-1 publication-title">Segmentation of Cellular Structures</h2>
<!-- Add title of the use case-->
<div class="is-size-5 publication-authors">
<!-- authors -->
<span class="author-block">
<a href="https://viscom.uni-ulm.de/members/hannah-kniesel/" target="_blank">Hannah
Kniesel</a><sup>1</sup>,</span>
<span class="author-block">
<a href="https://viscom.uni-ulm.de/members/tristan-payer/" target="_blank">Tristan
Payer</a><sup>1</sup>,</span>
<span class="author-block">
<a href="https://viscom.uni-ulm.de/members/poonam/" target="_blank">Poonam
Poonam</a><sup>1</sup>,
</span>
<span class="author-block">
<a href="" target="_blank">Tim Bergner</a><sup>2</sup>,
</span>
<span class="author-block">
<a href="https://phermosilla.github.io/" target="_blank">Pedro
Hermosilla</a><sup>3</sup>
</span>
<span class="author-block">
<a href="https://viscom.uni-ulm.de/members/timo-ropinski/" target="_blank">Timo
Ropinski</a><sup>1</sup>,
</span>
</div>
<div class="is-size-5 publication-authors">
<span class="author-block"><sup>1</sup>Visual Computing Group, Ulm
University<br><sup>2</sup>Central
Facility for Electron Microscopy, Ulm University<br><sup>3</sup>Computer Vision Lab, TU
Vienna</span>
<!-- <span class="eql-cntrb"><small><br><sup>*</sup>Indicates Equal Contribution</small></span> -->
</div>
<div class="column has-text-centered">
<div class="publication-links">
<!-- Arxiv PDF link
<span class="link-block">
<a href="https://arxiv.org/pdf/<ARXIV PAPER ID>.pdf" 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>
-->
<!-- Supplementary PDF link
<span class="link-block">
<a href="static/pdfs/supplementary_material.pdf" 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>-->
<!-- Github link
<span class="link-block">
<a href="https://github.com/YOUR REPO HERE" 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> -->
<!-- ArXiv abstract Link
<span class="link-block">
<a href="https://arxiv.org/abs/<ARXIV PAPER ID>" 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:/Link to Notebook" target="_blank"
class="external-link button is-normal is-rounded is-dark">
<span>notebook</span>
</a>
</span>-->
<a target="_blank"
href="https://lightning.ai/hannah-kniesel/studios/deep-em-toolbox-segmentation-of-cellular-structures">
<img src="https://pl-bolts-doc-images.s3.us-east-2.amazonaws.com/app-2/studio-badge.svg"
alt="Open Studio" />
</a>
</div>
</div>
</div>
</div>
</div>
</div>
</section>
<!-- Teaser image
<section class="hero teaser">
<div class="container is-max-desktop">
<div class="hero-body">
<img src="static/images/tasks.png" alt="Schematic showing 3 differnt types of task applicable for deep learning. (image to values, image to image & 2D to 3D)" />
<h2 class="subtitle has-text-centered">We propose to categorize tasks within the area of EM data analysis into Image to Value(s), Image to Image and 2D to 3D. We do so, based on their specific requirements for implementing a deep learning workflow. For more details, please see our paper.</h2>
</div>
</div>
</section>
End teaser image -->
<!-- Motivation -->
<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">
<div class="content has-text-justified">
<p>
We choose the segmentation of certain cell organelles as a relevant Image to Image task.
Segmentation is an important tool in EM image analysis as it contributes to a better
visualisation of
certain
organelles and complex cell structures,
which facilitates the interpretation of EM data. Segmentation allows for detailed analysis
of organelle
morphology, spatial relationships and distribution within cells.
This is crucial for understanding intracellular organisation and its relationship to cell
function.
Due to the small available dataset sizes we deploy data augmentation methods, make use of
pretrained weights
and train an ensemble model, which has been shown to provide
better generalizability even when trained on smaller dataset sizes [1].
</p>
<p><i>[1] Shaga Devan, Kavitha, et al. "Weighted average ensemble-based semantic segmentation in
biological
electron microscopy images." Histochemistry and Cell Biology 158.5 (2022): 447-462.</i>
</p>
</div>
</div>
</div>
</div>
</section>
<!-- End Motivation -->
<!--DEEP-EM TOOLBOX Workflow -->
<section class="section hero">
<div class="container is-max-desktop content">
<h2 class="title is-3">DEEP-EM TOOLBOX: Workflow</h2>
<div class="content has-text-justified">
<figure>
<img src="static/images/segmentation-of-cellular-structures/Teaser-ensemble.png"
alt="Teaser for segmentation of cellular structures">
<figcaption id="fig:teaser">
For the segmentation of cellular structures we follow [1] and train a so called "ensemble"
model.
An ensemble model is a set of models which are used to make multiple predictions for the same
input data.
The predictions are then combined to retrieve a more robust prediction.
</figcaption>
</figure>
<div class="orange-background">
<h3>Task</h3>
<p>
We categorize this segmentation as an <u>Image to Image</u> <u>task</u>, as we require an image
as input which will then
be translated into a semantic mask, also represented as an image. In the case of this notebook,
we follow the work of
[1] for defining the model.
As a <u>task-specific</u> architecture, we use an encoder-decoder network. More specifically, we
make use of the
U-Net [2]
architecture,
which was especially designed for biomedical image segmentation. Our goal is to develop a
so-called "ensemble" model.
An ensemble model is a set of models that are used to make multiple predictions for the same
input data. The predictions are then combined by a weighted average between all trained models.
The weights are computed by the performance of the model on the validation set. For the
ensemble, we use multiple
<abbr title="Convolutional Neural Network">CNN</abbr> <u>backbone</u> architectures and include
them in our
task-specific encoder-decoder architecture. Similar to our previous example notebook, we are
using
<abbr title="Convolutional Neural Network">CNN</abbr> architecture, as it is specially developed
for images and
performs well on small dataset sizes, compared to <abbr title="Vision Transformer">ViT</abbr>
backbones.
This allows us to get a more robust prediction, even though the training dataset is comparably
small.
</p>
</div>
<div class="green-background">
<h3>Data</h3>
<p>
Similar to the first example use case, we make use of pre-existing <u>real data</u> from
[3]
for <u>data acquisition</u>. Still, we encourage the research community to test the provided
script on their own data.
This requires gathering and annotating the data. We again propose the usage of the
CVAT [4] <u>annotation
tool</u>.
By doing so, the annotated data can simply be uploaded and plugged into the already provided
pipeline.
For details, we refer the reader to the "Data Annotation" section below and our corresponding notebook.
As <u>annotation type</u> we are working with pixel-level annotations of segmentation masks,
which are color-coded images grouping similar cellular structures, namely nucleus, cytoplasm,
and background. On your own data these classes can be adapted to your own needs.
</p>
<p>
For <u>data preprocessing</u>, we <u>split</u> the dataset into training, validation, and test
sets using 60%-20%-20% splits.
Lastly, we deploy different <u>augmentations</u> to the input images to increase the variance in
the rather small datasets.
The applied augmentations include vertical flipping, which flips the images along the vertical
axis, and random 90-degree
rotations that rotate the images by 90 degrees in any direction. We also apply horizontal
flipping to mirror the images
along the horizontal axis, and transposing to swap the image's width and height. Additionally,
grid distortion is used to
introduce slight warps and distortions across the image grid. These transformations are applied
on-the-fly with a probability
of one, which means that each training sample is slightly altered every time before it is
presented to the model.
This ensures a large variance in the training data, thus improving the model's ability to
generalize from the data.
</p>
<p>
We <u>reformat</u> our data from single-channel grayscale images to 3-channel RGB images so that
we are able to use
pretrained model weights where the model was trained on natural images. We do so because the
number of pretrained models
for <abbr title="Electron Microscopy">EM</abbr> is very limited, but we require multiple
pretrained backbones to initialize
the ensemble model. To address the difference between the pretraining data and our EM data, we
<u>normalize</u> the EM pixel
intensities similarly to z-score normalization but match the mean and standard deviation of the
pretraining data.
This helps reduce the gap between natural images used during pretraining and the <abbr
title="Electron Microscopy">EM</abbr> images.
</p>
</div>
<div class="red-background">
<h3> Model</h3>
<p>
As mentioned earlier, we are falling back on pretrained models of natural images for the
<u>initialization</u> of our model.
Despite the domain gap between natural images and <abbr title="Electron Microscopy">EM</abbr>,
pretrained weights from
ImageNet [5]
can deliver helpful prior information to the backbone.
</p>
<p>
We use the standard output as well as
<a href="https://wandb.ai/site" target="_blank">Weights & Biases</a>
for <u>logging</u> metrics for every model of the ensemble structure. Metrics include training
and validation loss as well as
<abbr title="Intersection over Union">IoU</abbr> on the validation set, which is a metric
specific for assessing the performance
of segmentation models.
</p>
<p>
Finally, during <u>evaluation</u> we accumulate the predictions of the models and pick the most
likely prediction, forming the
ensemble model. We assess the performance <u>quantitatively</u> by computing the <abbr
title="Intersection over Union">IoU</abbr>
for each model separately on the test set and finally combine their predictions to compute the
<abbr title="Intersection over Union">IoU</abbr> for the ensemble model.
</p>
<p>
We add <u>qualitative</u> evaluation on a subset of the test data by visualizing the model's
input, output, and corresponding
ground truth. Again, we assess this for each model separately as well as on the combined
ensemble model.
</p>
</div>
<p>[1] Shaga Devan, Kavitha, et al. "Weighted average ensemble-based semantic segmentation in
biological
electron microscopy images." Histochemistry and Cell Biology 158.5 (2022): 447-462.
</p>
<p>[2] Ronneberger, Olaf, Philipp Fischer, and Thomas Brox. "U-net: Convolutional networks for
biomedical
image segmentation." Medical image computing and computer-assisted intervention–MICCAI 2015: 18th
international conference, Munich, Germany, October 5-9, 2015, proceedings, part III 18. Springer
International Publishing, 2015.</p>
<p>[3] Morath, Volker, et al. "Semi-automatic determination of cell surface areas used in systems
biology." Front Biosci (Elite Ed) 5 (2013): 533-45.</p>
<p>[4] B. Sekachev et al., Opencv/cvat: V1.1.0, version v1.1.0, Aug. 2020. doi: 10 . 5281 / zenodo .
4009388. [Online]. Available: https://doi.org/10.5281/zenodo.4009388.</p>
<p>[5] Deng, Jia, et al. "Imagenet: A large-scale hierarchical image database." 2009 IEEE conference on
computer vision and pattern recognition. Ieee, 2009.</p>
</div>
</div>
</section>
<!--End DEEP-EM TOOLBOX Workflow-->
<!--Use your own data -->
<section class="section hero is-light">
<div class="container is-max-desktop content">
<h2 class="title is-3">Use Your Own Data</h2>
<div class="content has-text-justified">
<p>
Here we explain how you need to preprocess and annotate your data to apply the model for your use
case. We require that all data for training, testing and validation is annotated
and collected and then uploaded to the pytorch lightning studio.</p>
<h3>Data Structuring</h3>
<p>We require you to split the data in a test and train set. Our provided notebook will automatically
pick a validation split (20%) from the provided training data.</p>
<p>You should provide one folder containing the <b>training images</b> and one folder containing the
corresponding <b>training masks</b>, which were exported from the <a
href="https://www.cvat.ai/">CVAT</a> tool as described below.
</p>
<p>Similarly, you should provide one folder containing the <b>test images</b> and one folder containing
the
corresponding <b>test masks</b>, which were exported from the <a
href="https://www.cvat.ai/">CVAT</a> tool as described below.
</p>
<p>We recomment to use at least 20% of your dataset for testing.</p>
<p>For validation, we automatically pick a subset of 20% from the provided training images.</p>
<p>Within the provided notebook we descibe where you need to adapt the code in order to plug in your
data.</p>
<p></p>
<h3>Data Annotation</h3>
<p>
For data annotation we recomment using the <a href="https://www.cvat.ai/">CVAT</a> tool. It is
free to use
but requires a user account.
A quick user guide can be found <a href="https://docs.cvat.ai/docs/getting_started/">here</a>.
Multiple instruction videos can also be found on the <a
href="https://www.youtube.com/@LearnOpenCV">LearnOpenCV youtube channel</a> (for example <a
href="https://www.youtube.com/watch?v=yxX_0-zr-2U">this</a>).
For annotating the location labels within this use case we implement the use of point
annotations.
We export the data using the <i>Segmentation mask 1.1</i> format.
</p>
<button id="togglebtn-quan-ann" class="button-55"
onclick="toggleVisibility('content-quan-ann', 'togglebtn-quan-ann')">Show
More</button>
<div id="content-quan-ann" class="hidden">
<p></p>
<figure>
<img src="static/images/segmentation-of-cellular-structures/CVAT-1.png" alt="CVAT task 1">
</figure>
<p>When starting CVAT, you first need to create a new task. You can give it a name, add annotation
types and upload your data.</p>
<figure>
<img src="static/images/segmentation-of-cellular-structures/CVAT-2.png" alt="CVAT task 2">
</figure>
<p>Next, click on the <b>Add label</b> button. Name it based on the classes you want to annotate. In
our case these are "naked", "budding", "enveloped".
As annotation type choose <b>Points</b>. You should also pick a color, as this will simplify the
annotation process. For adding new labels click <b>Continue</b>.
Once you added all nessecary labels click <b>Cancel</b>.
</p>
<figure>
<img src="static/images/segmentation-of-cellular-structures/CVAT-3.png" alt="CVAT task 3">
</figure>
<p>Now you can upload the data you wish to annotate. Finally, click <b>Submit & Open</b> to continue
with the annotation of the uploaded data.
</p>
<figure>
<img src="static/images/segmentation-of-cellular-structures/CVAT-4.png" alt="CVAT job">
</figure>
<p>This will open following view. Click on the <b>job</b> (in this view the <b>Job #1150327</b>) to
start the annotation job.
</p>
<figure>
<img src="static/images/segmentation-of-cellular-structures/CVAT-5.png" alt="CVAT annotate">
</figure>
<p>To then annotate your data, select the <b>Draw new mask</b> tool. Select the <b>Label</b> you
wish to annotate from the dropdown menu. Then click
<b>Shape</b> to annotate
individual virus capsids with the label class.
</p>
<figure>
<img src="static/images/segmentation-of-cellular-structures/CVAT-6.png" alt="CVAT annotate">
</figure>
<p>When you finished annotating a single mask, click on the tick in the top left corner.
You can use the arrows on the top middle to navigate through all of your data and to see your
annotation progress.
</p>
<figure>
<img src="static/images/segmentation-of-cellular-structures/CVAT-7.png" alt="CVAT export 1">
</figure>
<p>Once you are done annotating data, click on the <b>Menu</b> and select <b>Export job dataset</b>.
</p>
<figure>
<img src="static/images/segmentation-of-cellular-structures/CVAT-8.png" alt="CVAT export 2">
</figure>
<p>During export select the <b>Segmentation mask 1.1</b> format and give the folder a name. It will
prepare the dataset for download.
</p>
<figure>
<img src="static/images/segmentation-of-cellular-structures/CVAT-9.png" alt="CVAT download">
</figure>
<p>In the horizontal menu bar at the top go to <b>Requests</b>. It will show a request <b>Export
Annotations</b>. On the right of this request click on the three dots to download your
annotated data. And you are done.
</p>
</div>
</div>
</div>
</section>
<!--End use your own data -->
<!--Links -->
<section class="section hero">
<div class="container is-max-desktop content">
<h2 class="title is-3">Contact</h2>
<div class="content has-text-justified">
<p>If you have any questions regarding this use case, please do not hesitate to contact <a
href="https://viscom.uni-ulm.de/members/poonam/">Poonam</a></p> <!--Add contact info here-->
</div>
</div>
</section>
<!--End Links -->
<!--BibTex citation -->
<section class="section is-light" id="BibTeX">
<div class="container is-max-desktop content">
<h2 class="title">BibTeX</h2>
<pre><code>BibTex Code Here</code></pre>
</div>
</section>
<!--End BibTex citation -->
<!-- Footer -->
<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 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>
<!-- End 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>