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
9 views

Optical flow models (FlowNet) training/finetuning process

I'm reading about optical flow models, particularly FlowNet and PWC-Net. I thought I understood how training and finetuning are being done, but I don't believe so anymore after trying to understand ...
1
vote
0answers
11 views

Fine-tuning VGG-Face for Facial Expression Recognition on FER2013 - Grayscale vs RGB Images

I am experimenting with Facial Expression Recognition and want to use a pretrained CNN model and a multi-stage fine tuning strategy to deal with scarce data. I came across the work of Knyazev et al. (...
0
votes
0answers
9 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 ...
0
votes
0answers
70 views

Help with derivation of gradient for specific filter in CNN

I need help with . I have to compute gradient for the special type of filter in CNN and everytime it comes up to be 0. Either this si correct, or I have some fundamental problem there, any hints are ...
0
votes
0answers
15 views

How to model the probability of detecting an image, given it is seen multiple times

Are there any existing methods/models describing the probability of an object being detected by a computer vision algorithm given it is seen $n$ times at similar angles and orientations? I know that ...
1
vote
0answers
11 views

Setting up an optimization problem using image datasets

I've been working with computer vision for a few months now. I've read a lot of papers that introduce novel models that do well on imagenet. I recently started reading a larger variety of CV papers ...
1
vote
0answers
24 views

How to tell if a model is overfitting or underfitting or the problem is something entirely different

I'm a complete beginner and I'm trying to do a multi-label classification on the well known dataset ChestX-ray14, which contains about 112 thousand x-ray images from about 31 thousand patients, the ...
1
vote
0answers
15 views

If I know specific pair of characters that model confuses in OCR task how can I fixe it?

I train OCR model to recognize cyrillic handwritten text. I know, for example, that it confuses very often 'Б' with '6'. How can I use this information to fine tune the model ? Just in case, my ...
0
votes
0answers
12 views

Resources for Computer Vision Algorithms and Applications

Are there any videos or other books/notes/slides that anyone has come across that follow Computer Vision Algorithms and Applications by Richard Szeliski? We are using this book in class and I have ...
0
votes
0answers
24 views

Class token in ViT and BERT

I'm trying to understand the architecture of the ViT Paper, and noticed they use a CLASS token like in BERT. To the best of my understanding this token is used to gather knowledge of the entire class, ...
2
votes
2answers
47 views

VGG19 vs Resnet18. When does VGG win?

I have always been under the impression that resnet is more technologically advanced than vgg and so you will always get better performance on resnet. I trained and tested my model on a sample data ...
0
votes
0answers
17 views

What will be the Precision and Recall value for Faster RCNN?

I am using TensorFlow object detection API for Faster RCNN object detector. Now I want to measure the performance of my model, so I have evaluated it using the code below for getting the mAP, ...
1
vote
0answers
13 views

Bilinear Interpolation Algorithm for up-sampling 2D images

In keras it is possible to use UpSampling2D layer to up-sample an image. You can use Bilinear Interpolation. Given an image ${h\times w}$ it is possible to increase its size in ${h*k\times w*l}$, ...
1
vote
0answers
9 views

Can I use SIFT feature detector other than images?

I know how to use SIFT algorithm for images but I never use it for other kinds of data. I have RGBD data (x,y,z,time) where x,y,z is the joint position along x,y,z coordinates. Now can I use SIFT ...
0
votes
0answers
12 views

Convolution formulation with central element of the Kernel matrix is superimposed on the pixel

Suppose we perform the convolution operation with a Kernel of odd size. Suppose that the central element of the Kernel matrix is superimposed on the p-th pixel of the image being processed. Suppose: ...
1
vote
1answer
66 views

Confusion about prior used in Recursive Bayes Filter

I'm currently using this thesis to understand key concepts about probabilistic inference in computer vision which is being a great source. The frame of the question is the following: Let us assume we ...
0
votes
0answers
3 views

Captions are better, yet validation loss is increasing

I am training and validating an image captioning model with the following architecture: Encoder: ResNet-101 Pre-trained on ImageNet Decoder: GRU (1-Layer) Embeddings: Last BERT hidden state I ...
1
vote
0answers
11 views

Creating a MRZ system on contracts

I'd like to create a MRZ detection system using a CNN : inputs are images of contracts and outputs are zones where to read. All the contracts have the same format but pictures can differ (light, angle....
0
votes
0answers
60 views

What is the difference between FPN(Feature Pyramid Network), FPNlite and SSDlite?

I came across this when I used MobileNet v2 from tensorflow hub. I know that FPN means feature pyramid network and it's better at identifying smaller objects in the frame. However I still don't know ...
0
votes
1answer
21 views

Is there any statistically meaningful definition for object confidence in object detection?

Most modern object detection algorithms rely on neural networks and output a bounding box and confidence for each object (or more accurately, a confidence for each possible object class considered, ...
1
vote
0answers
20 views

Are CRF/MRF/GRF still used widely in computer vision?

I've tried to find recent (the year 2020) popular works that use Markov/Gibbs/Conditional Random Fields. My approach was: go to Google Scholar and find the works, citing a few relevant works on this ...
0
votes
0answers
6 views

What is the intuition behind Dynamic Image Networks or Rank Pooling?

I am not quite understand why the parameter of ranking machine can be used as a representation, is there an intuitive way to explain it? Rank Pooling for Action Recognition: Ranking machines trained ...
0
votes
0answers
8 views

How to calculate labels and scores for a feature detector without classification or segmentation?

I am using SIFT feature detector to detect features from ground truth image and test image. I am using the location and scale of ground truth feature to define the predicted features on test image. I ...
2
votes
1answer
94 views

What are “Grids” and Detection at different scales" in YOLOV3?

I've recently started working with Yolov3 and the more I go in depth, the more confused I get. In the simplest terms what I think about YOLOV3 ...
0
votes
0answers
4 views

Marking the location of classes on an image with keypoints given a dataset of images, classes and keypoints

I have a dataset of 1000 images and 1000 JSON files. For example, the JSON file for the first image looks like this: index class aspect keypoints 0 roof new [(22, 24), (23, 2323)] 1 awning old [(76,...
0
votes
0answers
9 views

Denoising Prior to Image Classification

From what I have read, Denoising during preprocessing for image classification tasks seems to be a bit controversial. While on one hand it might improve classification accuracy, the computational ...
0
votes
0answers
7 views

Does TensorFlow's Object Detection API models look at the whole image or only the bounded target?

I was wondering if CNNs, specifically the models/feature extractors offered in Tensorflow's Object Detection API, only train on the bounded box of the target image or if it considers the entire image ...
1
vote
1answer
51 views

When is my Wasserstein GAN-GP overfitting?

I have a hard time interpreting the WGAN-GP losses attached. At which epoch is D and/or G overfitting? The quality improves a lot overtime, yet the generator loss at later epochs does not appear on ...
0
votes
0answers
9 views

Dealing with large input image size for CNNs

I'm using CNNs to implement a defect detection system for quality control. Since the dataset is not extremely large, I have decided to use transfer learning and take the low level features of another ...
0
votes
0answers
10 views

Can the scale Struct2depth be recovered given the real focal length?

Recently I read a great paper, struct2depth. But as I noticed that the scale is normalized in this paper, I wonder: if one can recover the actual ...
1
vote
1answer
150 views

Darknet and Data Augmentation

In the darknet deep learning framework .cfg files we see parameters like angle, saturation, exposure These parameters are used ...
1
vote
0answers
16 views

What does the determinant of a homography matrix represent?

I am trying to find one image (needle) within another (haystack). I am using the following OpenCV method, which first matches keypoints with SIFT and then applies homography: https://opencv-python-...
1
vote
0answers
37 views

SIFT: why Gaussian blur is performed iteratively?

SIFT is the feature detector I am trying to implement for self-study purposes. But my question concerns the Gaussian blurring done as part of detecting the keypoints. Gaussian pyramid is constructed. ...
2
votes
1answer
54 views

When computing mAP for an object detection model, how many detections should one consider?

I am trying to write some code to evaluate the MS COCO style mAP (mean average precision, average computed at the category level) at different IOU levels in the context of object detection with a ...
1
vote
0answers
30 views

huge neural networks for small datasets

In this period my colleague is working on a computer vision task involving a dataset very small (it's a classification task with a number of examples for class ranging from 20 to few hundreds). She ...
1
vote
0answers
89 views

what are the main differences between parametric and non-parametric machine learning algorithms?

I am interested in parametric and non-parametric machine learning algorithms, their advantages and disadvantages and also their main differences regarding computational complexities. In particular I ...
0
votes
0answers
6 views

Difference in features generated by same filters for color and grayscale images?

Would there be ay difference between the features generated by CNNs if they are fed with same image in color and grayscale format. If I am performing classification with same network for let's say ...
0
votes
1answer
13 views

What are some existing techniques for pose estimation angle normalization?

So I am currently building a model which does a certain type of action recognition, which I am implementing as a two-stage, end-to-end system. The first stage is a pose estimation model, and I want to ...
0
votes
1answer
361 views

Multiple object detection where a single image can have multiple objects

I did a project of object detection in which there was a single object in the whole image. My CNN network was taking an image and was outputting two things one is the class or category of the object (...
0
votes
0answers
11 views

N-to-1 frame CNN prediction model: how to represent the data?

I’m relatively new to vision modeling and I have a PyTorch CNN model that can predict the next frame of image based on 1 input frame relatively well. Now I want this model to output 1 image based on a ...
1
vote
0answers
20 views

Why conditional GAN does not need “negative sample”?

In conditional GAN's implementation, the inputs and labels of the discriminator are (fake_image, corresponding_condition, 0) and (real_image, corresponding_condition, 1). For example, (fake_dog, dog, ...
1
vote
0answers
42 views

Enemy detection in 3D video games without labeled dataset

I'm working on a tool for single-class object detection in video games, which should be able to detect enemies in 3D environments. The main problem is, that I cannot use any labeled dataset. Instead I ...
0
votes
1answer
32 views

Resize image in object detection task of computer vision

In object detection, they usually resize by keeping the ratio the same as the original image, which usually names "letterbox" resize. My question is: Why we need to do that? As I see with ...
0
votes
0answers
129 views

How to create one hot segmentation masks from rgb mask image for multiclass segmentation?

I am trying to train Deeplabv3 for semantic segmentation on BDD100k dataset. It contains 20 classes for segmentation task. In the Dataset labels are provided as RGB mask image (3 channels). How do I ...
0
votes
1answer
50 views

How to make use of ground truth data in image anomaly detection?

If I have an image dataset that consists of "normal", anomalous and ground truth image data, how do I make use of the ground truth data? To my understanding if I train an unsupervised ...
0
votes
2answers
258 views

Different input size for training and prediction in CNN for image segmentation?

I’m relatively unexperienced when it comes to deep learning and I’m trying to reimplement a CNN architecture for segmentation of medical images based on a paper. In the paper they state that they use ...
3
votes
1answer
90 views

Why use multiple anchor boxes with the same positions in a multi-box detector?

What is the benefit of using multiple anchor boxes with the same positions in a single-shot multi-box detector model, like YOLO? In particular, I notice Google's BlazeFace model does this. If the ...
0
votes
0answers
16 views

Instance Segmentation: How to separate incomplete instances?

My task is to train instance segmentation model for documents segmentation (there are multiple documents on the image). After segmentation, we are going to run OCR algorithms for each deteted document....
0
votes
0answers
7 views

Data augmentations to mimic natural variations present satellite imagery

I'd like to apply some machine learning algorithms to satellite imagery that we've collected, but I want to encourage invariance to factors such as sunlight intensity, time of day, atmospheric ...
0
votes
0answers
5 views

How to implement a transfer learning like training process?

I've been working on a UNet and I've been advised to try a transfer learning style approach. My issue is that I can't visualise the training procedure, I've got myself confused by overthinking the ...

1
2 3 4 5
8