Questions tagged [computer-vision]

Questions related to image representation, segmentation, visual object categorization and image processing algorithms in general.

Filter by
Sorted by
Tagged with
2
votes
1answer
37 views

Learning useful semantic representations of data

Training a neural network on its final task (e.g. classification) right from the beginning is not always the best way to go. I'd like to make a short list of recognized methods of motivating a NN to ...
2
votes
1answer
79 views

Detect visual attention area in an image [closed]

I'm trying to detect the visual attention area in a given image and crop the image into that area. For instance, given an image of any size and a rectangle of say LxW dimension as an input, I would ...
2
votes
1answer
438 views

How to extract vector representation from a comparison neural networks

From what I understand about embeddings in neural networks, the upper layers (fully-connected) from convolutional neural networks can serve as an image encoding (vector representation of an image), ...
2
votes
1answer
412 views

How is the stride calculated in the Faster RCNN paper?

I am trying to understand a paragraph in the Faster RCNN paper. We train and test both region proposal and object detection networks on single-scale images [7, 5]. We re-scale the images such that ...
2
votes
1answer
640 views

batch normalization

I've read a lot about batch normalization and how it is implemented. But I couldn't understand when to use batch normalization or not. What should we notice about the Convolutional Neural Network to ...
2
votes
2answers
742 views

Bag of Visual Words: the number of words is equal to the number of k-means centroids?

I was reading these slides about Bag of Features (BoF), in particular at slide 23: A visual vocabulary of 1M words is generated using an approximate K-means clustering method based on randomized ...
2
votes
1answer
422 views

Why are convolutions preferred to local receptive fields?

It is my intuition that unrestricted / unshared weights encourage the network to capture more invariances than in the case of convolutions, where the weights are shared. (These capture mostly ...
2
votes
0answers
46 views

What is the error for ImageNet “Object localization” challenge?

I have been reading some papers which use the ImageNet-LOC (ImageNet-Localization) dataset. I tried to read up on it to understand what the goal of this dataset is, and hence, what the error we are ...
2
votes
0answers
35 views

Can Graph Neural Networks be better than Convolutional Neural Networks for computer vision tasks? [closed]

Recently, a strong trend in deep learning is the adoption of Graph Neural Networks for computer vision tasks (https://github.com/thunlp/GNNPapers#computer-vision). But the main question is could this ...
2
votes
0answers
29 views

How to use machine learning to create combine of opposite images side by side [closed]

Inspired by: Two Worlds Pictures I just want to create a Machine Learning Model that can automatically combine the opposite images into 1 image. I am thinking about 2 possible solutions: Pose ...
2
votes
0answers
285 views

What is the relation between a loss function and an energy function?

A loss function is a function that measures the distance between the expected value and the actual value of a model (an example of a loss function is the cross entropy). An energy function can be ...
2
votes
1answer
13 views

Model to Recommend Ideal Parameter Changes for Best Performance of Industrial Machine

I'm trying to develop a machine learning model to solve this problem, and am unsure of where to start. We begin with some user-defined settings. The settings are used by a machine to create a product....
2
votes
0answers
11 views

What are the important methods that evolved in computing optical flow?

I have gone through various approaches to find optical flow. But I have a tad confusion between Horn and Shunck method and Lucas Kannede method. Where are these methods useful and where do these ...
2
votes
0answers
21 views

How can I lower the resources taken by an ANN while limiting data loss?

I designed a visual quality assurance system with a goal to determine whether a factory part during manufacturing is flawed or not. The need arose because it is rather difficult to tell if a part is ...
2
votes
0answers
108 views

clustering objects in point cloud

I am currently working on point cloud data analysis, trying to label objects which are not ground, vegetation, etc. So far I tried many clustering algorithms, with moderate success. In my best model ...
2
votes
0answers
53 views

CNN: Modifying VGG16 Architecture

I'm currently trying to modify the VGG16 network architecture so that it's able to accept 400x400 px images. Based on literature that I've read, the way to do it would be to covert the fully ...
2
votes
0answers
34 views

Developing algortihm/model to identify thin linear features in aerial imagery [closed]

I am exploring the possibility of identifying fencelines from NAIP aerial imagery (GSD = 0.6m). I have tried some basic processing in OpenCV using canny edge detection that was detailed in a question ...
2
votes
0answers
38 views

Machine Learning: mathematical verification of this text-to-image cross entropy loss function?

I'm implementing a research paper on GANs and have come across this rather convoluted text-image loss function which has these main components: $$P(D_i | Q_i) = \frac{\exp({\gamma_3 R(Q_i, D_i)})}{\...
2
votes
0answers
107 views

How to choose a good batch size for a powerful CPU [duplicate]

I am training a deep neural networks for self driving cars using Adam optimization, and I wonder how can I find a standard batch size value , currently I am using the value 1 and I can see that my ...
2
votes
0answers
242 views

What areas/applications that still use traditional computer vision algorithms instead of deep neural networks?

What areas/applications that still use traditional computer vision algorithms instead of deep neural networks?
2
votes
1answer
101 views

Any way to recognize pattern(such as Char and Number) from image without labeled data?

I am trying to build a captcha recognizer. I found CNN play very well if there are enough labeled data. For example, I use this https://github.com/lepture/captcha to generate 4 size char+number ...
2
votes
0answers
636 views

Precise localization and characterization of rudimentary shapes with neural networks

I understand that there are flavors of (convolutional) neural networks that are useful for object localization and detection tasks of reasonable difficulty. In all of the examples I have seen so far, ...
2
votes
0answers
289 views

Different sized inputs for batch training in fully CNNs

The idea of transfer learning is to use already trained networks for another purposes to the one it was initially trained for. Using fully convolutional networks, the activation maps can be used to ...
2
votes
0answers
578 views

Using Shape Context to classify OpenCV contours with KNN

I have a set of 2D polygons represented as OpenCV contours that I would like to use to train a KNN classifier on using shape context. I am using the OpenCV python "cv2....
2
votes
0answers
56 views

How to build a network that detects common object in images?

I have a group of images. All contain some common object (let's assume all at the same size and no other alterations). I want to train a CNN that will learn the filter for the common object and gives ...
2
votes
1answer
48 views

What's the state-of-the-art in learing human (motion) behaviour from videos? [closed]

Dear fellow colleagues, I’m currently working on a research project that involves predicting human behavior from video streams. Since I’m mostly working on other fields of research, and thus I’m not ...
2
votes
0answers
67 views

how can deep learning model to help us explain the underlying pattern for a given data set

As it is well known, deep learning can help to learn the features. However, are there any ways for us to either understand or visualize these learned features? Moreover, how to explain the patterns ...
2
votes
0answers
320 views

Right procedure for Leave-One-Out Cross-Validation in multi-class classification

I am intending to implement a existent method to recognize sign language. The model uses a data-set of N (375) classes but with only 3 examples per class. The authors used Leave-One-Out Cross-...
2
votes
0answers
69 views

Defining the probability distribution of a Random vector given the probability over a “sub-vector”

Suppose I want the probability distribution over a random vector $X={X_1 ,X_2 ... X_n }$. What I already have with me is the distribution over a subvector $X_i , X_{i+1}...X_m$, $m<n$ which I ...
2
votes
0answers
36 views

What image registration metric should be employed to account for Poisson Noise?

I have a pair of adjacent frames (volumes) Xt and Xt+1 and I want to obtain a non rigid registration field relating the pixels (voxels) of Xt+1 to Xt. The intensities among the images are similar. If ...
1
vote
1answer
1k views

100% accuracy on training, high accuracy on testing as well. What does this mean?

I was training a model to classify different traffic signs and decided to use a pre-trained alexnet model and redefining the last fully-connected layer to match the classes of the dataset. When I did ...
1
vote
1answer
432 views

Handling data without ground truth bounding boxes in SSD / RetinaNet?

In the paper SSD: Single Shot MultiBox Detector by Liu et al., 2015, the Matching strategy section reads: During training we need to determine which default ...
1
vote
2answers
872 views

HOG blocks: adding histograms or concatenating histograms?

In Histogram of Oriented Gradients, the edge orientations within rectangular patches are binned into the bins of a histogram. Each pixel adds the strength of its orientation in the bin corresponding ...
1
vote
1answer
485 views

Data normalization in k-means and svm

Generally if I want to normalize my data in which direction I should normalize (subtracting mean and dividing by std)? Lets say I have a data matrix D (...
1
vote
1answer
28 views

How does Stochastic Gradient Descent with momentum distinguish between local minima and global minima?

I have several questions regarding this. How does SGD momentum know to converge at global minina and skip over local minima? I read that "SGD momentum goes past the minima (due to its velocity build ...
1
vote
1answer
58 views

Object detection with just one image? [duplicate]

Deep neural networks require lots of examples to learn tasks like image classification, and object recognition. On the other hand, we humans can learn and identify object just by looking at it once. ...
1
vote
1answer
48 views

Simple but effective face detection algorithm using neural networks

I'd like to have my undergrad machine learning students have the option of doing a face detection project using neural networks (constructed by the students using Keras). The algorithm should ideally ...
1
vote
1answer
82 views

In weakly supervised learning for object detection and localization, how does the neural network associate a particular object with its label?

I understand how an object detection algorithm (in particular neural nets) can localize one object. The question is how they can associate a particular object with a particular label when there are ...
1
vote
1answer
189 views

HoG feature vector confusion

I am new in the machine learning world and currently I am working on Computer Vision project. I am confused about HoG feature vector. In my understanding, the feature vector should contain magnitude ...
1
vote
1answer
440 views

How do DenseNets work?

I implemented a DenseNet for class and wondered why it works well, when I noticed no one had posted on StackExchange specifically about DenseNet. Why are DenseNets good?
1
vote
1answer
2k views

Can you brief the training procedure of SSD in TF Object Detection API?

In the paper it has mentioned how they consider classification and localization loss . But it is not clear how the TF-OD API for SSD model does that. Can any one brief this ?
1
vote
1answer
1k views

Calculating joint distribution of two i.i.d geometric random variables, and proving their independence

In this question, we will further investigate the geometric distribution. Let X, Y be i.i.d. geometric random variables with parameter p. Let U = min{X,Y} and V = max{X,Y} − min{X,Y}. Compute the ...
1
vote
2answers
423 views

Which machine learning algorithim can I use for this kind of pattern recognition?

I have an interesting real world problem that can be abstracted and decomposed into a pattern recognition problem - specifically, recognising "known configurations" from within a 2D plane. The ...
1
vote
2answers
2k views

How to normalize filters in convolutional neural networks?

Usually, when convolving images, the elements in the filter sum to one. Is this criterion enforced in convolutional neural networks? If yes, how?
1
vote
3answers
9k views

How to calculate disparity of two images in matlab?

I have two images and I am trying to calculate the disparity between them using sum of squared distances and reconstruct disparity in 3D space. Do I need to rectify the image before calculating ...
1
vote
1answer
17 views

why is using a mixture of logstic distributions makes sense in pixelcnn++?

I went trough the paper and code of the pixelcnn++ model. From what I understand, they train the network in the following way for predicting the value of a single pixel: the inputs are the pixel ...
1
vote
1answer
30 views

Performance of MaskRCNN/YOLO as a function of object size in pixels

I am trying to find references on how the resolution of an object affects the ability of object detection systems such as MaskRCNN and YOLO to correctly identify the object. For example, if the ...
1
vote
1answer
27 views

Incorrect predictions on extracted images from text [closed]

I trained a model in PyTorch on the EMNIST data set - and got about 85% accuracy on the test set. Now, I have an image of handwritten text from which I have extracted individual letters, but I'm ...
1
vote
1answer
126 views

Understanding the weighted cross-entropy method of u-net

I am trying to implement the weight-cross entropy mentioned in unet paper to counter the class-imbalances. I am not really able to understand how they are exactly implementing the weight-cross entropy....
1
vote
1answer
206 views

Why are we interested in top-n accuracy?

I know the definition of top-n accuracy: [1] What is the definition of Top-n accuracy? My question is, why do we even care for this? A short answer could be to compare different models in various ...

1 2
3
4 5
7