Questions tagged [conv-neural-network]

Convolutional Neural Networks are a type of neural network in which only subsets of possible connections between layers exist to create overlapping regions. They are commonly used for visual tasks.

Filter by
Sorted by
Tagged with
1 vote
0 answers
18 views

Is downsampling necessary in CNN?

I am still trying to understand the effect of any downsampling (reducing the input height and weight by 2 for example) by pooling or strided convolution? Does downsampling improve accuracy? Because in ...
1 vote
0 answers
24 views

Why does downsampling improve acccuracy? (UNet)

I am trying to understand why my UNet replica model has lower accuracy when I did not downsample at all (Only changing the number of channels throughout the network) compare to when I use maxpooling ...
2 votes
0 answers
15 views

How can fully convolution neural networks handle images of different sizes?

I've read that if we want to use images of different sizes in a convolutional neural network without resizing the images to a default size, we can use Fully Convolutional Neural Networks. But I do not ...
  • 121
0 votes
2 answers
29 views

why is the receptive field size not depend on the input image?

Let say if I fixed the filter size as 3x3, then shouldn't an input of 128x128 image has a larger receptive field size than an image of 32x32? In addition, what would be the affect of stride and ...
1 vote
0 answers
9 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 ...
  • 378
1 vote
0 answers
17 views

Why not use two convolution layers in a row?

This question is about composing (aka stacking, repeating, successively applying) convolution layers with no nonlinearity between them. I first encountered this idea in Justin Johnson's Stanford/UMich ...
0 votes
0 answers
12 views

Algorithm to calculate the transposed convolutional [duplicate]

I'm really getting frustrated understanding the Transposed Convolution. In the previous question I got a good answer which contains the code for calculating the Transposed Convolution. But I'm looking ...
  • 171
0 votes
0 answers
17 views

What are the pros and cons of Skip Connections?

When I was reading about the ResNet, I came across with the Skip Connections. I do understand how they work but I don't quite understand their purpose or when they play a role. In other words, which ...
  • 171
0 votes
0 answers
40 views

The result of back propagation for a neural network [duplicate]

I have created a neural network that feeds an image into a convolutional neural net, then feeds the flattened output of this network into an artificial neural network. I have a feeling that my ...
  • 23
2 votes
0 answers
20 views

Multi-class image classification model has high test accuracy but low prediction confidence

TL;DR: Although my model is predicting correct classes, it is not very confident about those correct predictions. My dataset contains 32x32-sized images for 46 classes. I have trained a Deep ...
1 vote
0 answers
16 views

Standardizing data for neural network: High density at mean, large amount of outlier values

I have not been able to find much information on good practices for dealing with these two issues. The first is a high concentration around the mean, and the second is a large amount of values far ...
1 vote
1 answer
22 views

CNN result replication on tensorflow

I am trying to replicate the first Sequential example in this video: https://www.youtube.com/watch?v=WAciKiDP2bo using the following code ...
  • 161
1 vote
1 answer
27 views

Best reconstruction loss for RGB images?

Which loss works the best for pixel-wise RGB image (3, width, height)reconstruction loss? It seems there are several options Regression way. The input image has ...
1 vote
0 answers
58 views

Why does the CNN model accuracy vary too much when the dataset is the same?

I have been working on a project where I have a lot of time series data(3000 csv file) from 6 different devices and I am trying to convert those data to an image array so that I can use them in CNN to ...
0 votes
0 answers
81 views

cnn-lstm autoencoder for multivariate time series

Hello I’m new in deep learning,I have a multivariate time series dataset composed of 49 sensors .i’m trying to perform anomaly detection using cnn lstm autoencoder where 1d cnn is used to encode the ...
0 votes
0 answers
27 views

What does do 'Graph Convolution' in 'Point GCN' architecture based on Graph CNN

I am reading various research papers to understand the methods which takes Point cloud and convert into machine learning readable data for various purpose such as classification. I have one question ...
1 vote
0 answers
21 views

Can I use batchnorm in CNN + RNN, and where to place it exactly?

I have designed a following neural network that combines CNN, RNN and Dense layers. It aims to predict a positive or negative outcome for the time step t+1, given a ...
  • 181
0 votes
0 answers
24 views

How do i design my CNN so that i picks up on details like a logo [duplicate]

i'm quite new to machine learning, so i hope i can get som help on here. I try to classify pictures of soda-cans and bottles with size 299x299. So based on the shape of the object itself, my networks ...
  • 103
0 votes
0 answers
30 views

Training VGG16 architecture BGR vs RGB differences

I am training a VGG16 architecture (using Keras api) on a binary imbalanced dataset (around 10% of the samples are positive). I have noticed that in both cases (random weights train vs fine tuning an ...
  • 101
0 votes
0 answers
18 views

What's GAN's Input-size Limitations?

I am interested in GAN for generating synthetic data. I am studying the input limitations for GAN starting from which GAN is no longer usable. I have found many applications that use GANs for ...
  • 101
0 votes
0 answers
18 views

Cross Entropy with a Very Sparse Output

I'm interested in training a deep convolutional network for 2D gridded data that, instead of classifying an entire sample, will classify each pixel in the sample. I'd like it to find the center pixel ...
  • 111
1 vote
1 answer
22 views

How does finding the error work in the back propogation of a CNN when 4D tensors are used?

I am currently creating a CNN from scratch in C++ only using vectors, and Matrix classes I have created. Mainly doing this as an exercise so that I can make sure I fully understand the process. I am ...
1 vote
0 answers
34 views

How ResNet researchers achieve high accuracy?

I tried to recreate the original ResNet 50 model myself, using all the same layers, hyperparameters, and data augmentation methods mentioned in the paper. (I only trained on a few hundred iterations ...
1 vote
0 answers
16 views

Why didn't the authors of ResNet scale their dataset by the standard deviation?

In the original ResNet paper, when pre-processing their dataset, the authors subtract the mean, but they don't scale the images by STD (or by 255): The plain/residual architectures follow the form in ...
0 votes
0 answers
17 views

Emotion classifier: overfitting the training dataset [duplicate]

I'm working on a binary classification model over the RAVDESS dataset with a CNN model. These are the performances on the train and validation set and these are the performance on the test set for ...
0 votes
0 answers
50 views

Predict the missing pixel value by CNN

The data is images which resize to 256*256. And for per 8x8 area, we remove 4 pixels from the 8x8 block. Then iter this process to whole image. So the task is how to use per block pixels value(60 ...
1 vote
1 answer
36 views

Difference between Symmetrically normalized Laplacian matrix versus graph laplacian matrix

I am trying to understand the graph laplacian matrix in Graph Convolution networks. To get a basic understanding of graph laplacian matrix I am referring to this https://mbernste.github.io/posts/...
0 votes
1 answer
40 views

Derivate of Neural Network respect to input

I have a neural network like this $x=\text{input}$ $z_1=W_{1x}\cdot x+b_1$ $h_1=\text{relu}(z_1)$ $z_2=W_2\cdot h_1+W_{2x}\cdot x+b_2$ $h_2=\text{relu}(z_2)$ $y=W_3\cdot h_2+W_{3x}\cdot x+b_3$ input ...
0 votes
1 answer
24 views

Can bounding boxes be used in UNets?

I recently read a paper where the researchers used a UNet algorithm to localize/detect cyclones using a bounding box. However, my interpretation of a UNet is that it performs semantic segmentation and ...
  • 1
1 vote
1 answer
21 views

Masking pixels for a CNN

I'm trying to implement a CNN with RGB and depth images. But my depth images are a little sparse. So I would like to mask out those neighborhoods in the RGB image where the depth neighbors are empty. ...
0 votes
0 answers
31 views

What do flattening learning curves indicate and when to stop training of a ML model in that case?

I am training CNNs for image segmentation on a limited dataset and apply some on-the-fly data augmentation. I measure mean intersection over union (mean IoU) to evaluate the training and select models....
0 votes
1 answer
15 views

How to calculate the total number of inputs in CNN?

I search this kind of question for a while and I find many discussions involve on counting the number of parameters of a Convolutional Neural Network, but not on the inputs. Using the Fashion MNIST ...
1 vote
0 answers
16 views

Is there literature about how different neural network layers recognise certain features?

Let's consider a deep convolutional network. It seems that there is some consensus on the following notions: 1. Shallow layers tend to recognise more low-level features such as edges and curves 2. ...
  • 649
0 votes
0 answers
19 views

GoogleNet-LSTM, cross entropy loss does not decrease [duplicate]

...
  • 1
1 vote
1 answer
28 views

Where can I find pre-trained fully convolutional neural networks? [closed]

I know that fully convolutional neural networks can be used for classifying images of arbitrary sizes. I would like to use some pre-trained fully-convolution neural networks for extracting features in ...
1 vote
0 answers
26 views

When I do not add an activation function to my convolutional layer the model gets quickly stuck in a local optima, why?

I have model A: ...
0 votes
0 answers
23 views

What kind of neural networks’ architecture is employed by Playground Tensorflow?

As you may know, Playground tensorflow allows you to design in real time an artificial NN and have a visual understanding of how it works. How is it possible to have such a representation? Is that a ...
  • 289
1 vote
0 answers
417 views

CNN-LSTM or LSTM better for univariate Time series forecasting?

I am working with simulated univariate sequential data and the goal is to forecast that data. I was wondering which model CNN-LSTM or LSTM is better for predicting univariate time series data. Both ...
  • 121
0 votes
0 answers
55 views

Normalizing features for CNNs and out of distribution

As I was reading this question on another thread: Why normalize images by subtracting dataset's image mean, instead of the current image mean in deep learning? I realize that either one point is ...
0 votes
0 answers
11 views

amount of parameters of Conv1D vs Conv2D (Keras library) [duplicate]

imo, I should have the same amount of parameters if I construct Conv1D and Conv2D whereby in Conv2D one dimension is set to 1 (as if it was eliminated) here are snippets from the summary function of ...
  • 196
0 votes
0 answers
36 views

Is my model learning? - judging by the results on real dataset and random generated data

Ok this might seem like a repetitive question but trust me it's not - at least to the best of my knowledge. I'm working on a CNN regression project, where the input are 3D images structured as ...
  • 1
1 vote
0 answers
43 views

Sampling a Parameter Space with a Neural Network versus other Methods [closed]

I have heard a lot about neural networks as an approach to sampling a large parameter space in aim of finding the space of model parameters that match a dataset. I've done some research online, but ...
  • 111
1 vote
0 answers
49 views

Is deep double descent important in practical contemporary CNNs?

Deep double descent is an empirically observed phenomenon that happens with contemporary neural networks. Its essence is that often, increasing the model complexity first leads to the test loss ...
  • 152
0 votes
1 answer
54 views

What is the purpose of 2 fully connected hidden layers in VGG16?

Here is the architecture of VGG16: The first 18 convolution layers can be understood as feature extraction. How about the two fully connect hidden layers after them? What is their purpose?
  • 867
0 votes
0 answers
41 views

validation accuracy spikes on Unet Neural network

I'm a computer science student actually working on my graduation thesys based on a Unet ANN.I'm using Kolektor Surface-Defect Dataset 2 (i'll leave the link below) that is a dataset of annotated ...
1 vote
0 answers
51 views

Why does a neural network perform poorly in case of small loss?

Background I'm building a convolutional neural network (CNN) to predict the response factor (continuous variable) of organic molecules. As input, I use 86x86 onehot-encoded matrices that represent the ...
0 votes
0 answers
104 views

Can I train Mask R-CNN with very small images and detect objects in much bigger images?

I have lots of images depicting one object each, and the objects are labelled with one of 10 classes (image size 40x20). Is it possible to train a Mask-RCNN with those small images and then detect ...
  • 101
1 vote
0 answers
102 views

Does a CNN always learn a latent space?

In general, a latent space is a structure of reduced dimensionality than that of the input space where points on this space share resemblance the closer they are to each other. This article also ...
  • 649
1 vote
1 answer
16 views

How to determine the number of pixel from a warped image from a region proposal RCNN

I am learning R-CNN with this slides. On slide 63 one warped image region is specified by 224x224 pixel. Is this just a random value or from where is the value coming from? I cannot see any ...
1 vote
0 answers
35 views

How to deal with unknown classes with a convolution neural network classifier?

I'm quite new into the DL and ML field. I'm training a CNN able to classify 3 different classes, however I would like in the testing phase to make the CNN able to not misclassify images that do not ...
  • 11

1
2 3 4 5
29