Questions tagged [image-segmentation]
Image segmentation arises in computer vision and digital signal processing. The goal of image segmentation is to partition a digital image into pieces, where each piece corresponds to some semantically important concept. Usually, this means that each pixel is assigned to one of the concepts. An simple example is dividing a picture of a person into the subject (the person in the foreground) and the background (whatever is behind and around the subject).
111
questions
0
votes
0
answers
10
views
Calculation of the Generalized Dice Loss Gradient
I am trying to understand the gradient of the Generalized Dice Loss (GDL) shown here Link. It says that the GDL for two classes is:
$$
GDL = 1 - 2 \frac{\sum_{l=1}^2w_l \sum_{n=1}^{N} r_{ln}p_{ln}}{\...
- 1
0
votes
0
answers
25
views
Theory and mathematical background of active contour and segmentation models
I am looking for reference books/online material for self-learning the mathematical background in active contours and segmentation models (e.g., snakes, level set, geodesic). Can someone help me with ...
0
votes
0
answers
7
views
Fully Convolutional Networks: Fully Connected Layers as Convolutional Layers
I'm reading the paper "The Fully Convolutional Network" and I don't understand this passage in 3.1:
Typical recognition nets, including LeNet [21], AlexNet [19], and its deeper successors [...
- 1
1
vote
0
answers
19
views
How could I use machine learning to search for components in technical drawings?
I have a very large library of technical drawings (in image only format) specifically of mechanical parts, grouped in bundles.
The drawings have various levels of cleanliness since some were hand ...
- 111
0
votes
0
answers
25
views
How do i apply conformal prediction to achieve voxel-wise uncertainty quantification in a 3D binary segmentation problem?
Context: Typically in an image segmentation problem we go from the model's output logits to sigmoids to discrete labels (0 or 1 in this case). Akin to a binary classification problem, we can set up a ...
1
vote
0
answers
84
views
Comparison of Squared Dice Loss vs. Standard Dice Loss
I've been diving into segmentation tasks and came across two variations of the Dice Loss that I'm considering for my neural network: the standard Dice Loss and the Squared Dice Loss.
The Standard Dice ...
- 128
1
vote
0
answers
9
views
CNN: adding a third class improves the overall ranking of the network
I am working on a Convolutional Neural Network for classifying two classes of images whose difference between them is very small. Running the CNN (using PyTorch), it was able to correctly classify ...
- 253
0
votes
0
answers
72
views
How to calculate Mean Average Precision at a fixed IoU threshold?
Papers such as COCO and the VisDrone often list MAP, MAP50, MAP75 in their evaluation criteria. To my understanding, MAP is the averaged AP for all classes across an IoU range (0.5 - 0.95 for example),...
0
votes
0
answers
43
views
Resizing mask file in semantic segmentation - implications
I have followed the tutorial here for Multiclass semantic segmentation using DeepLabV3+.
My question regards a preprocessing function. In the following block, is it not problematic that resizing the ...
- 1
1
vote
0
answers
32
views
How to aggregate semantic segmentation results from a ML model?
I'm working on a semantic segmentation system using aerial imagery. To make this question concise, I'll present a toy version of the problem I want to solve.
Assume we are trying to produce a semantic ...
- 11
0
votes
0
answers
24
views
Accounting for spatial correlations in ROC analysis for image segmentation
Typically, we assume independent samples when performing ROC analysis.
But for image data, e.g. in a segmentation problem, pixels in a neighborhood come from the same image and are spatially ...
- 1
0
votes
0
answers
19
views
How to measure uncertainty and account for spatial bias when conducting ROC analysis on image data?
I have a task where I need to perform ROC analysis and measure the AUC of the ROC curve, but the data is image data.
I have pairs of images (which contain real-valued pixels) and masks (which contain ...
0
votes
0
answers
52
views
find the optimal threshold for each class in model.predict (multiclass segmentation)
I have a unet segmentation model, which outputs 5 classes, I would like to find the optimal threshold value for each class using the precision-recall curve:
...
- 1
0
votes
0
answers
13
views
Research Problem about Weakly Supervised Learning for CT Image Semantic Segmentation
My issue is that Grad-CAM often highlights the wrong areas. My task is to perform weakly supervised semantic segmentation (WSSS) of lung malignant tumors using image-level labels. Although I achieved ...
- 1
0
votes
0
answers
14
views
Considering large annotator differences, is there an appropriate evaluation metric for segmentation?
I am doing a segmentation task. There are 5 annotators to annotate the images. However, the segmentation target is a small object whose boundary is hard to delineate. Therefore, the differences of the ...
- 151
0
votes
1
answer
228
views
Which would be a better approach for Multiclass segmentation?
I would like to perform multiclass segmentation on DRR(digitally reconstructed radiographs) by using Unet network. For binary segmentation, unet works well (dice score over 0.93)
What I’m trying to do ...
- 3
0
votes
0
answers
9
views
Semantic segmentation: document masking for personal data
here i am again with a pretty broad question.
I'm trying to put down ideas to build a model capable of locate and obfuscate personal informations such as name, codes, numbers, etc in a document.
The ...
- 111
1
vote
0
answers
28
views
Image Segmentation using MAML algorithm (same objects exist in all tasks)
I have an n-takes k-shots medical image segmentation problem.
-Tasks: Different human organs ex: liver, spleen, kindness etc...
-Shots: 10 CT scans NIFTI images, where all tasks(human organs) exist in ...
- 33
0
votes
0
answers
51
views
Implementing Patch-Based Segmentation for a Lung Lesion Segmentation task
Background:
I have two datasets, both with Lung CT Scan images as input. The labels for both are masks. The first dataset has Lungs as masks. The second dataset has lung lesions (17 classes like ...
- 61
0
votes
0
answers
31
views
Segmentation network for melanoma can't generalize well [duplicate]
I'm trying to replicate a model from a paper, and cannot make it generalize well. I'm using ISIC 2018 dataset for melanoma semantic segmentation. The masks are binary indicating in white what is ...
0
votes
0
answers
14
views
How to detect an unknown number of segments, each to be fitted with an unknown parametric curve/surface equation?
Let's say I have a set of points (possibly noisy) in an N-dimensional space that represent an arbitrary number of curved segments, each segment having an arbitrary type of curve. See the 2D sample ...
- 101
0
votes
0
answers
91
views
Classification of images of different sizes in a Variational Autoencoder
Context:
We are using deep learning in image analysis of cells for segmentation and classification in a research setting. So far, we can successfully segment the cells using the U-Net/CellPose ...
- 1
1
vote
1
answer
29
views
What are some best practices for labeling data that exists in a continuum?
I am building computer vision models on data that exists in a continuum. For example, imagine I'm trying to do semantic segmentation on cars. Some of the labels are distinct, like "chipped paint&...
- 408
1
vote
0
answers
56
views
Is it a good idea to have a category and its subcategories in the training set of an object segmentation model?
I hope you are doing great!
I am currently training an object segmentation model (detectron2 : mask rcnn)
The objective is to detect materials like wood, plastic, glass etc...
wood is one of the ...
0
votes
1
answer
291
views
Intersection over union vs sensitivity
In the context of segmentation, what is the difference between IoU and sensitivity? It sounds to me like they describe the same formula in different contexts but I might be wrong.
When true ...
- 101
1
vote
0
answers
25
views
How to get segmentation algorithms to identify regularly repeating patterns
I am looking for defects in a structure that can be difficult even for a human to detect. I'm using segmentation algorithms (e.g. Mask RCNN, UNet) to do this. Sometimes the structure will have ...
- 408
1
vote
0
answers
63
views
Best K-fold Segmentation Cross-validation
In image segmentation tasks, what validation metric is standard for model selection? I'm logging binary cross-entropy loss, Dice coefficient, and accuracy, but I don't know if I should use validation ...
- 111
0
votes
0
answers
14
views
Probability of sampling >50% "bright" pixels from an image
Suppose an image consists of 300,000 pixels. 51% of these pixels are bright and 49% are dark. We are trying to determine if the image is bright or dark through sampling (in this example it would be ...
0
votes
0
answers
367
views
Why most works on Cityscapes don't use weighted cross-entropy?
Weight Cross-Entroy (WCE) helps to handle an imbalanced dataset, and Cityscapes is quite imbalanced as seen below:
If we check the best benchmarks on this dataset, most of the works use bare CE as a ...
- 158
0
votes
0
answers
763
views
Dice Coefficient vs *Negative* Dice Coefficient..?
While reading this paper I've noticed that they use what they call negative dice coefficient. I know that in general the dice coefficient is a metric commonly used in image segmentation tasks when we ...
- 577
0
votes
0
answers
388
views
How to compute ROC AUC for a method that uses two models?
To compare my method with others I'm trying to compute its AUC but I got a bit confused on how to do this for my case. My method uses a model that classifies an image as class A or B, after that if ...
1
vote
0
answers
155
views
Kernel_size for rgb images in cnn?
I came across a cnn code of rgb images where kernel_size was mentioned only 3 not 3,3,3. So does 3 means 3,3,3. and for greyscale images kernel size was mentioned 3,3 so for grey scale images 3,3 I ...
- 11
2
votes
1
answer
342
views
Is there an official procedure to compute mIoU (mean intersection over union)?
Although it sounds silly, I'm not finding an official source to compute mean intersection over union (mIoU).
I'm realizing a semantic segmentation task, and I want to compute the mIoU over a dataset. ...
- 158
1
vote
1
answer
187
views
Two basic questions about icp (iterative closest point) algorithm
I am trying to learn shape analysis and a part is learning icp. I have many confusions but for now I have two basic questions:
Does the point clouds need to have the same number of points for icp?
...
- 43
1
vote
1
answer
2k
views
Proper way to calculate mean Dice coefficient on a dataset
I would like to ask a question about the proper way to calculate the Dice coefficient for an image dataset. We know that the Dice coefficient is calculated via the following equation:
$$Dice = \frac{...
- 337
-1
votes
1
answer
128
views
Proper approach for image recognition of ~1000 symbols
We have a dataset of black symbols in grey squares (like attached below).
The symbols are various letters (arabic, greek) as well as numbers in many distinct fonts; altogether ~1000 different images.
...
- 171
1
vote
0
answers
79
views
Does atlas-based imaging segmentation generally involve machine-learning [closed]
Segmentation is an important task in medical imaging analysis.
Many FDA approved medical device use "atlas-based" segmentation tasks. Newer device use "deep-learning based" ...
- 80
0
votes
0
answers
71
views
Spatial Verification Techniques for Image Prediction
I am working on an image prediction problem, where we use a U-Net to predict a real-valued image. I've found that conventional metrics like MSE, r^2, MAE, etc just don't really cut it. What are some ...
- 131
1
vote
0
answers
36
views
Estimating the mean and variance for a set of probabilities (bounded by 0 and 1) based on Image Segmentation results
Data
My data is from a set of images wherein I am computing the Sensitivity of an image segmentation algorithm. Sensitivity is computed as:
$$Sensitivity=\frac{TP_{pixels}}{TP_{pixels}+FN_{pixels}}$$
...
- 569
1
vote
1
answer
113
views
Correct or most common term for altering a loss function to ignore unlabelled pixels?
In my experience it is quite commonplace to alter the loss function used when training a neural network for segmentation to ignore the contribution to the loss of unlabelled pixels. There are a few ...
- 211
0
votes
1
answer
2k
views
About ROC curve in segmentation model
I know how to draw ROC curves about classification model for a one class. And I know how to plot ROC curves about classification model for many classes. But is there a way to plot ROC curves for a ...
김현정
1
vote
0
answers
25
views
Am I doing correct unit test before whole batch training?
I read somewhere that unit tests are important before jumping onto training for the whole batch. And for that reason, if one sample overfits on the model, can we decisively say that the training will ...
- 106
1
vote
0
answers
339
views
Mask for image padding in semantic segmentation
I'm using data augmentation for a semantic segmentation task, where some images are cropped or rotated. As a result, some padding is added to ensure that the image is always the same size. These ...
- 23
1
vote
0
answers
56
views
Fluctuating loss curve/ steady dice score. Why? And How to improve? [duplicate]
I am training 3D data with multi-class 3D target ground truths(9 tissue labels) for segmentation. Using dice Loss and focal dice loss as loss criterion. Updating optimizer every second batch (...
- 106
1
vote
0
answers
38
views
Do horizontal and vertical dead pixels on some of the bands of a hyper-spectral image affect the accuracy of the model?
I hope this is the right place to ask this question. It should be here according to questions I saw.
I am working of PRISMA Earth Observation images which consist of tens of spectral bands. My final ...
- 11
2
votes
1
answer
285
views
What does ADE20k (the scene segmentation benchmark) stand for?
ADE20k is a scene segmentation dataset created by MIT. It is a common benchmark for localization tasks in computer vision research. I cannot find anywhere what the name stands for!
This information ...
- 21
0
votes
1
answer
2k
views
Correct way of computing dice score for image segmentation?
In binary image segmentation, for given a set of images, it's true mask and predicted mask. How to compute dice score?, should I compute dice score for each image separately and then find mean across ...
- 11
1
vote
0
answers
27
views
MobileNetV2 in Keras: Adjusting depth
I am trying to reconstruct specific U-Net architecture with the MobileNetV2 backbone using Keras. It seems for MobilenetV1, there is a way to adjust the depth using the ...
- 73
1
vote
0
answers
114
views
Reusing Weights in Transposed Convolution
As far as I know it's possible to reuse the weights of a convolution in a transposed convolution to upsample an image. However when reusing the weights, the resulting restored images aren't even close ...
- 341
1
vote
0
answers
2k
views
Are F1 score and Dice coefficient computed in same way or different way in image segmentation (two class segmentation)?
On page 8 of the paper An automatic nuclei segmentationmethod based on deep convolutional neuralnetworks for histopathology images, the authors show performance of their deep model on test sets. They ...
- 111