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).
106 questions
0
votes
0
answers
35
views
Heteroscedastic variance prediction constant where it should not be
I am trying to make predictions about the heterscedastic noise in my dataset. I have an FPN already set up, treating the variance as an additional class. My dataset is the aerial semantic segmentation ...
0
votes
0
answers
9
views
How to perform user assisted image segmentation using Gaussian Mixture Models?
I have a general idea of Gaussian Mixture Models. My understanding:
GMM is a way of clustering data points which, unlike K means clustering, soft assigns them under different distributions by ...
1
vote
0
answers
23
views
Train model with labels generated by similar model: overfit?
I train models to predict some linear features from aerial imagery. Because the reference data are just lines, I made a simple buffer so that labels resemble very approximately the width of the target ...
- 11
2
votes
1
answer
24
views
What would be the convolutional layer output by keras.layers.Conv2D when conv output is fractional?
I have input ($n=224$), strides ($s=4$), filter size ($k=11$) and no padding which gives me a fractional conv output:
$$\texttt{conv output} = (n-k+2p)/s + 1 = 54....
- 23
0
votes
0
answers
8
views
Segmentation 3D architectures
I would like to ask you for advice on the best solution for my problem. I have a task to segment an object based on some of its signal. That is, I have input objects of size T x H x W, where T is time....
- 11
0
votes
0
answers
20
views
How do I initialize a bias for the final layer of a CNN if my final output is logits and not probabilities?
I'm working on a medical image binary segmentation problem using a U-Net in tensorflow, and my classes are extremely unbalanced (about 1 in 10,000).
I want to initialize a good bias for the last layer ...
- 153
0
votes
0
answers
22
views
Detecting Object Removal in Images
The problem statement is as follows - Given an altered image (an image from which some object has been removed), generate a mask for the removed object.
For instance, say an original image contains ...
- 115
0
votes
0
answers
98
views
Instance segmentation using a discriminative loss?
I have been reading this paper and I was wondering if their discriminitive loss definition is correct for instance segmentation ?
From what I understand they map the image pixels into a higher ...
- 111
2
votes
1
answer
92
views
Calculation of the Generalized Dice Loss Gradient
I am trying to understand the gradient of the Generalized Dice Loss (GDL) shown here Link. It says that the GDL for two classes is:
$$
GDL = 1 - 2 \frac{\sum_{l=1}^2w_l \sum_{n=1}^{N} r_{ln}p_{ln}}{\...
- 21
0
votes
0
answers
29
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 ...
1
vote
0
answers
22
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
1
vote
0
answers
245
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 ...
- 138
1
vote
0
answers
18
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 ...
- 283
1
vote
0
answers
61
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
1
answer
383
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
1
vote
0
answers
43
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
22
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
2
votes
1
answer
42
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&...
- 438
1
vote
0
answers
71
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
416
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
37
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 ...
- 438
1
vote
0
answers
83
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 ...
0
votes
0
answers
15
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
391
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 ...
- 156
0
votes
0
answers
968
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 ...
- 659
0
votes
0
answers
447
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
163
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
464
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. ...
- 156
1
vote
1
answer
333
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:
$$Dice = \frac{...
- 347
-1
votes
1
answer
134
views
Proper approach for image recognition of ~1000 symbols
We have a dataset of black symbols in grey squares (like attached below).
The symbols are various letters (arabic, greek) as well as numbers in many distinct fonts; altogether ~1000 different images.
...
- 181
1
vote
0
answers
86
views
Does atlas-based imaging segmentation generally involve machine-learning [closed]
Segmentation is an important task in medical imaging analysis.
Many FDA approved medical device use "atlas-based" segmentation tasks. Newer device use "deep-learning based" ...
- 120
0
votes
0
answers
75
views
Spatial Verification Techniques for Image Prediction
I am working on an image prediction problem, where we use a U-Net to predict a real-valued image. I've found that conventional metrics like MSE, r^2, MAE, etc just don't really cut it. What are some ...
- 131
1
vote
0
answers
36
views
Estimating the mean and variance for a set of probabilities (bounded by 0 and 1) based on Image Segmentation results
Data
My data is from a set of images wherein I am computing the Sensitivity of an image segmentation algorithm. Sensitivity is computed as:
$$Sensitivity=\frac{TP_{pixels}}{TP_{pixels}+FN_{pixels}}$$
...
- 569
1
vote
1
answer
146
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 ...
- 251
0
votes
1
answer
3k
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 ...
- 116
1
vote
0
answers
415
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 ...
- 33
1
vote
0
answers
60
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 (...
- 116
1
vote
0
answers
40
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
370
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
31
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 ...
- 103
1
vote
0
answers
125
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 ...
- 371
2
votes
1
answer
3k
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 ...
- 121
1
vote
1
answer
630
views
Tversky Loss function for RGB masks
I have a very imbalanced dataset for my semantic segmentation problem (monitoring deforestation using setellite images) and I found Tversky Loss to be much better than categorical crossentropy (due to ...
-1
votes
1
answer
641
views
How to initialize an encoder-decoder type of neural network that is used in image segmentation?
In this example:
https://github.com/qubvel/segmentation_models.pytorch/blob/dcd19d676bdfbf73fc140d5b98d780f449b0a2f8/segmentation_models_pytorch/base/initialization.py
It only initializes the decoder ...
- 155
1
vote
1
answer
502
views
Using FCNN for multi-class semantic segmentation trained on single class labeled image data
I am working on project where main task is semantic segmentation of land cover and another objects in Sentinel 2 multi-spectral images. Currently I posses dataset ...
- 113
2
votes
0
answers
108
views
How to choose the best segmentation model using the area under the precision recall curve, IOU and Dice metrics?
I am using several U-Net variants for a brain tumor segmentation task. I get the following values for the performance measures including Dice, IOU, Area under receiver-operating characteristic (AUC) ...
- 123
2
votes
0
answers
328
views
How to measure the performance of Mask RCNN model. Given that there are two tasks , one object detection and another image segmentation
Mask RCNN is an instance image segmentation technique. It is based on Faster RCNN for object detection and an additional mask operation is performed by another CNN.
