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
0
votes
0answers
6 views

Non-linearity (ReLU, Batch Norm) before final sigmoid convolution in image segmentation

When building an image segmentation model, standard building blocks consist of Conv -> RELU -> Batch Norm -> DropBlock. I understand the importance of having a non linear activation, and the role that ...
0
votes
0answers
4 views

Optical flow equation understanding [closed]

Optical flow: I have been experimenting to extract flow data from a set of video files. The video files have all kinds of motions ( local, global and translation and rotation and slow and fast). Based ...
3
votes
0answers
12 views

Extreme Image Noise Removal

I've been trying to solve a noise removal (from images) problem using deep learning and I've tried a lot of the newer architectures for noise removal including FFDNet, NLRN and MWCNN. The problem is, ...
1
vote
1answer
22 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
0answers
16 views

What happens when the filter size is same as that of image size in a CNN?

In Convolutional Neural Networks (CNNs), theoretically speaking, if the filter dimensions are same as that of the image (training example) dimensions, will CNNs boil down to normal Fully Connected ...
0
votes
0answers
12 views

Intersection vs Chi-Square for comparing histograms

I've seen lots of examples where chi-square works better. But is there any example of for ex. 3 non-similar histograms where Intersection results in the similarity of histograms but Chi-square results ...
0
votes
0answers
6 views

Techniques for estimating homography

Is there any good comparison of methods (SIFT+RANSAC, SURF, MSER, ConvNets...) for estimating homographies / finding correspondencies? Or, if you have experience in this topic, what would be the best (...
0
votes
0answers
21 views

Why Resnets Converge Faster?

According to the Research Paper: Deep Residual learning for Image Recognition, the 18-layer plain/residual nets are comparably accurate , but the 18-layer ResNet converges faster. What is the reason ...
0
votes
0answers
7 views

Annotation of dataset for object detection

I am annotating a dataset for object detection of UI elements (buttons, text, edit text, etc..) in images of phone screens and I am wondering what is a better approach. If I want to detect buttons (...
1
vote
0answers
28 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 ...
1
vote
0answers
10 views

How to visualize 3d joints of a SMPL model based on pose params

I am trying to use demo.py in https://github.com/nkolot/GraphCMR. I am interested in obtaining joints from the inferred SMPL image and visualize it similar to ...
4
votes
1answer
625 views

IID in real life /Machine Learning - When is data truly IID? [duplicate]

In a course I am studying at Berkeley, some student said about a particular Dataset "Data is not iid" and the lecturer agreed with him. https://youtu.be/kl_G95uKTHw?list=...
1
vote
0answers
19 views

Pre-training a network on significant landmarks

Let's say there is a dataset consisting of photos of cars from various brands, and we're trying to train a ConvNet to identify the brand from a photo, just like these ones: One approach I was ...
2
votes
0answers
30 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 ...
1
vote
2answers
15 views

How can a given conv neural net layer handle filters of different size?

traditional method is to use multiple filters of same dimensions but with different weights and stack the output (basically concatenate them) that is then to be fed into the next conv layer. If I ...
1
vote
1answer
129 views

Training CNN - 3 different training sets, different noise

I am doing a road segmentation task for high resolution images. I have three different data sets: Around 100 with extra high resolution with the ground truth. Around 500 images with slightly worse ...
1
vote
2answers
369 views

To provide dimensionality reduction, 1x1 convolutions are used, before passsing them through a 3x3, or 5x5 convolution in an Inception module.

To my understading what a 1x1 convolution does is gives an embedding of the (i,j)th entry of the feature map along its depth. Besides here some dimensionality reduction is also done. How will the ...
2
votes
1answer
21 views

Are there any Non Neural Network models to do face detection in constrained domains?

In some constrained domains(eg: car driver), the camera is stationary which means the background will not change much. And we can sure when the car is running, there must be a driver. In this kind of ...
0
votes
1answer
31 views

How to develop a Deep learning model for only two Imges …?

I am trying to build Deep learning model for only two Images. I have a two images of doors , one image is some what good and another is some what bad, I want to identify the one is good and another ...
0
votes
1answer
275 views

Binary classification using GPML toolbox;

Using the demo given in the demo_classification file, I am trying to do a binary classification where each of my class contains 10 samples of 73 dimensions. Following is the code where I try to '...
2
votes
1answer
179 views

Custom Loss Function - Inducing sparsity

From the comments, I realized that my question wasn't clear enough, so I'll start with a short background. I am trying to construct an attention model that performs classification based on just a ...
0
votes
1answer
27 views

How to use additional non-visual data for image classification?

I've got a Resnet network that classifies images into n classes and is working fine. I want to boost its performance buy using additional information I have regarding the images. This info comes in ...
0
votes
0answers
33 views

Is object detection the right approach for this problem

I'm trying to build a model which, given a picture of someone's face, is able to identify all the following features, as well as others. The model would output, for each picture, a list of all the ...
1
vote
1answer
352 views

How best to combine object detection and tracking

I am trying to make a computer vision system which will be able to detect and track objects of interest. This will require (1) detection functionality to notice the object when it appears (2) ...
0
votes
0answers
17 views

Computer vision with smaller data set

I hope this is the right forum to ask. I had a client approach me with a demand for a vision system for their assembly line. The problem they are facing is that the operator sometimes forgets to put ...
1
vote
1answer
19 views

Looseness of the definition of domain adaption

I am a bit intrigued about “domain shift” concept. Specifically, in part 5 of the paper “Coupled Generative Adversarial Networks”, it reads We studied the problem of adapting a digit classifier ...
0
votes
1answer
22 views

Analysing the impact of each CNN layers

I want to analyse the impact of each layer of CNN. I have trained the CNN model with a dataset. After that, weights of first convolutional layer are fixed and remaining layers are initialise to zero ...
7
votes
1answer
2k views

Why do people use Zero-Padding in Convolutional Neural Networks?

I am wondering why people usually pad with zeros instead of e.g., using the min-value. Zero-padding, in my opinion, makes sense if you have input images with a pixel range [0, 255] or [0, 1] (after ...
0
votes
0answers
20 views

Reading a pressure gauge with a CNN

Using standard Computer Vision pipelines to read pressure gauges is well established, and not overly accurate or generalizable: For various reasons, I would like to use a CNN to do this. Since the ...
3
votes
1answer
144 views

Training an Object Detection Model Using with Artificial Data from Video Games

I had an interesting idea of using artificial data gathered from screen shots of a high-resolution video game as a cheap substitute for labeled real data, which can be quite expensive or difficult to ...
0
votes
0answers
27 views

Retraining of object detection CNN

I am working on an object detection system that should detect UI elements (such as button, checkbox, radio button, etc..) in the photo of a touch screen of printer (not screenshots, but literally a ...
17
votes
4answers
10k views

Is it possible to give variable sized images as input to a convolutional neural network?

Can we give images with variable size as input to a convolutional neural network for object detection? If possible, how can we do that? But if we try to crop the image, we will be loosing some ...
0
votes
1answer
166 views

Combining custom YOLO network for face detection with another CNN

I am looking for a way to build and train an end-to-end CNN that contains two steps: 1) a CNN for finding a face and hands in the image and 2) CNN that works on the crops of the face and hands. To ...
0
votes
0answers
17 views

Semantic Segmentation Multi-Class Single Channel Output Math

For semantic segmentation problems, I understand that it's a pixel-wise classification problem. At the last layer of the neural network, I would basically have a 1x1x1 convolution layer with a softmax ...
1
vote
1answer
27 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
2answers
159 views

How to think about the architecture of the Convolutional Neural Network?

Recently, I've started to learn more about CNNs to use them in some computer vision tasks. At the moment, I have roughly good knowledge about different parts of a CNN such as layers, solvers, loss ...
1
vote
1answer
26 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 ...
2
votes
1answer
171 views

Benchmarks and state-of-the-art methods for semantic segmentation of 3D meshes?

I'm wondering what benchmarks there exist for semantic segmentation of 3D meshes? I have already found "A Benchmark for 3D Mesh Segmentation"; is this currently the only benchmark that exists for 3D ...
11
votes
5answers
34k views

What loss function should I use for binary detection in face/non-face detection in CNN?

I want to use deep learning to train a face/non-face binary detection, what loss should I use, I think it is SigmoidCrossEntropyLoss or Hinge-loss. Is that right, but I also wonder should I use ...
0
votes
0answers
16 views

video normalization using skvideo

I'm building a model that would take as input a video of 25 frames and would (ideally) output the next 25 frames. My question is when we use images we usually normalize by dividing the X by 255. ...
3
votes
1answer
889 views

cross channel parametric pooling layer in the architecture of Network in Network

While reading the paper of Network in Network, I feel confusing about some points. The following figure shows the network architecture, looks like to me the two layers with red circle are just fully ...
1
vote
1answer
879 views

SegNet CamVid dataset training classes mismatch?

This is with reference to the CamVid dataset and one of its tutorial here: http://mi.eng.cam.ac.uk/projects/segnet/tutorial.html I'm quite confused by how the model is supposed to be trained on 11 ...
0
votes
0answers
13 views

Image classifier from text files

I have a data set of thousands of images of hundreds of pixels in gray scale ranging from -1 to 1. The labels represent 0 to 9. The issue is that the data set is in .txt format. How can one make an ...
3
votes
1answer
85 views

Same value of min and max in min-max normalisation

By the definition of min-max normalisation, the value is divided by max - min, what if the max and ...
2
votes
1answer
71 views

What would be the ideal dataset to train a model to detect advertisements in an image?

I am thinking of the requirements for training a model that would be able to detect if there is any kind of ad in an image. I know that this sound too broad not just for a question on CV but for ...
1
vote
0answers
46 views

Karush-Kuhn-Tucker Conditions in a Biometrics Research Paper

For one of my course works I'm supposed to find a scientific paper from my field of research that involves an optimisation problem solution based on the Karush-Kuhn-Tucker conditions. My research ...
0
votes
1answer
95 views

Object distortion after ROI Align in Mask R-CNN

In Mask R-CNN, if there are 2 proposed ROIs which cover 2 objects that looks like below: #1 A square object #2 A rectangular object So my question is: After ROI Align, is the #2 feature map ...
2
votes
3answers
679 views

Which image format is better for machine learning .png .jpg or other?

I'm trying to train a neural network with images. Since I'm extracting images from a video feed I can convert them either to .png or .jpg. Which format is preferred for machine learning and deep ...
1
vote
0answers
19 views

Reconstruction using low-light images

Let's say we have a regular photo and three low-light photos illuminated in different colors. Each pixel is a three-component vector $q=(R,G,B)$. Then $q_k^{A}$ is the $k$-th pixel of the regular ...
1
vote
1answer
84 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
2 3 4 5
7