Skip to main content

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).

Filter by
Sorted by
Tagged with
0 votes
0 answers
12 views

Quantification of Leaf Disease with Semantic Segmentation

I am trying to quantify leaf disease using dataset of original images and corresponding masks. I have two approaches in my mind: Train-Test model for Semantic Segmentation of Leaf and diseased region ...
Urwa Shanza's user avatar
2 votes
1 answer
22 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....
Shri's user avatar
  • 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....
Maxim's user avatar
  • 11
0 votes
0 answers
16 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 ...
Thao Nguyen's user avatar
0 votes
0 answers
6 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 ...
Aditya Kulkarni's user avatar
0 votes
0 answers
62 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 ...
KFkf's user avatar
  • 111
2 votes
1 answer
74 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}}{\...
sephlink's user avatar
0 votes
0 answers
28 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
11 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 [...
user0906's user avatar
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 ...
FarO's user avatar
  • 111
0 votes
0 answers
44 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 ...
Amir Vahdani's user avatar
1 vote
0 answers
183 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 ...
mutli-arm-bandit's user avatar
1 vote
0 answers
12 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 ...
donut's user avatar
  • 263
1 vote
0 answers
50 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 ...
ceebs's user avatar
  • 11
0 votes
1 answer
332 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 ...
Cork's user avatar
  • 3
1 vote
0 answers
39 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 ...
AMAS AL's user avatar
  • 33
0 votes
0 answers
19 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 ...
Anson Kao's user avatar
  • 101
1 vote
1 answer
34 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&...
jss367's user avatar
  • 408
1 vote
0 answers
67 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 ...
Mountassir El Moustaaid's user avatar
0 votes
1 answer
384 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 ...
Yiftach's user avatar
  • 101
1 vote
0 answers
30 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 ...
jss367's user avatar
  • 408
1 vote
0 answers
73 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 ...
explicitEllipticGroupAction's user avatar
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 ...
Aryaman Darda's user avatar
0 votes
0 answers
377 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 ...
Rafael Toledo's user avatar
0 votes
0 answers
911 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 ...
James Arten's user avatar
0 votes
0 answers
427 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 ...
Davi Magalhães's user avatar
1 vote
0 answers
160 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 ...
brian grey's user avatar
2 votes
1 answer
422 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. ...
Rafael Toledo's user avatar
1 vote
1 answer
291 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? ...
Maria's user avatar
  • 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{...
Dang Manh Truong's user avatar
-1 votes
1 answer
130 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. ...
maciek's user avatar
  • 171
1 vote
0 answers
84 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" ...
ava's user avatar
  • 100
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 ...
McM's user avatar
  • 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: $$Sensitivity=\frac{TP_{pixels}}{TP_{pixels}+FN_{pixels}}$$ ...
jeffalltogether's user avatar
1 vote
1 answer
140 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 ...
KDecker's user avatar
  • 251
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 ...
user avatar
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 ...
banikr's user avatar
  • 116
1 vote
0 answers
385 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 ...
gmedina-v's user avatar
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 (...
banikr's user avatar
  • 116
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 ...
Emre's user avatar
  • 11
2 votes
1 answer
339 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 ...
charzhar's user avatar
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 ...
spb's user avatar
  • 11
1 vote
0 answers
28 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 ...
vcucu's user avatar
  • 73
1 vote
0 answers
120 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 ...
Tom Dörr's user avatar
  • 371
2 votes
1 answer
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 ...
Prasanjit Rath's user avatar
1 vote
1 answer
603 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 ...
Петр Воротинцев's user avatar
-1 votes
1 answer
617 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 ...
fatpanda2049's user avatar
1 vote
1 answer
488 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 ...
Many's user avatar
  • 113
2 votes
0 answers
106 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) ...
shiva's user avatar
  • 123
2 votes
0 answers
327 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.
aayushe singh's user avatar