Stack Exchange Network

Stack Exchange network consists of 175 Q&A communities including Stack Overflow, the largest, most trusted online community for developers to learn, share their knowledge, and build their careers.

Visit Stack Exchange

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.

1
vote
2answers
22 views

In CNN, how to map from fully connected layer to output image?

In CNN, where the last layers are fully connected, how to make pixel-wise prediction to output an image(binary matrix), if the number of neurons in the last layer is less than the size of the image?
0
votes
0answers
8 views

Predict Probability Distribution with Neural Network or Monte Carlo

Let's say we would like to predict price of Microsoft Stock. We have historical data and interested in predicting price distribution for future time t+1, like shown ...
0
votes
0answers
10 views

CNN, “squared” or "non-squared image?

I'm working on a project about image recognition. In my dataset I have images of different size, all rectangular image (the most 640x480 and 1280x640). I would like to build my classifier to ...
0
votes
0answers
17 views

why the validation accuracy of deep CNN is high but not stable

I'm training a CNN on some vibration data, and I get some somewhat strange results. I found the validation accuracy is unstable. In some epoch, the val_acc may be 90%+, but in the next epoch, it may ...
1
vote
0answers
13 views

What is parameter chaining in CNN?

I have read an article about the benefits of CNN. One of the points was "parameter chaining". What does it mean? And how does it make CNN more convenient?
0
votes
1answer
33 views

What are general practises used to divide the data into training / dev and test set?

Example: I have am building a dog vs cat classifier and I have collected data from 15 countries. Europe: 1. UK 2. France 3. Germany 4. Italy 5. Finland Asia: 1. India 2. China 3. Japan 4. Russia 5. ...
0
votes
0answers
10 views

how to take a subset of a dataset to fine-tune a neural network?

I would like to build a classifier with 80 000 images and 45 classes. As each epochs takes 1 hour to train, Is there a way to win time by training only a subset of the dataset without lowering the ...
1
vote
0answers
15 views

What could cause a flat loss function to suddenly decrease in a u-net used for denoising?

So I am trying to understand U-Nets better, and I built a very shallow U-Net and trained it to denoise MNIST images (training set is 90% of the whole dataset). The loss function evolution I obtained ...
0
votes
0answers
9 views

overfiting validation set increase model performance

I'm using CNN for image classification on an unbalance data set. (e.g class A = 1000 image, class C=50 image). I got 16 class. I'm using class weights and in total i have less than 3500 images. I do ...
0
votes
0answers
9 views

Confused about z-score image normalization output

I am trying to normalize my input data for a convolutional network, I applied the z-score normalization technique to my image dataset as follows: Formula: (image - mean(image)) / std(image) ...
0
votes
0answers
13 views

Meaning of kernel size 1 for 1-D convolution in Keras

The kernel size is the window size for 1D convolution. Can anyone explain what is meant by kernel size $1$ in Keras/TensorFlow?
0
votes
0answers
14 views

Implementing a graph convolutional layer, pixel2mesh example

I'm trying to read through some python code in order to understand how to implement a Graph Convolutional Layer. I was particularly interested in pixel2mesh, digging through the code I've found the ...
0
votes
1answer
21 views

1-D convolution neural network in Keras

I am exploring 1-D CNN with Keras. My data is $\mathit{k}\times\mathit{N}$ where $\mathit{k}$ is the number of time stamps and $\mathit{N}$ is the number of features. I want to apply CNN with 1-D ...
0
votes
1answer
34 views

Replacing CNNs with Random Forests

Suppose I have a sequence like "ADTGESW". Each character in this sequence can attain a number of possible values, let's say 10. I can then one-hot encode this sequence and obtain a matrix with shape ...
0
votes
0answers
16 views

Literature recommendation for convolutional neural nets

I am looking for a good book or an article concearning convolutional neural nets, especially their architecture. I like the http://deeplearningbook.org but it does not provide any information on the ...
0
votes
1answer
36 views

Which machine learning or deep learning model to use / make [closed]

I wanted to use machine learning / deep learning to predict profit of a , say, a shop using satellite imagery and previous data and I don't know what to use. I want to use close satellite imagery to ...
0
votes
0answers
16 views

Image augmentation on test set for CNN model, using SMOTE

I know it's uncommon to augment the test set, but i know we can augment test set and then do the average (using the same augmentation rules than for training set). here the thing, i made a mistake ...
0
votes
0answers
11 views

How many parameters do I have to train two layers with different filter sizes?

I have a $10\times10\times3$ colour image input and I want to stack two convolutional layers with kernel size $3\times3$ with $10$ and $20$ filters respectively. I have stacked on how to calculate ...
0
votes
0answers
15 views

what is the best approach in dealing with large dimension custom data for training and predicting deep learning models

i am trying to implement semantic segmentation for satellite images.My custom dataset has dimensions(height,width)in range (3000, 3000)what is the best approach for feeding(for training) and ...
0
votes
0answers
27 views

What receptive field do we have after stacking $n \times n$ CONV layers with kernel size $k \times k$?

What receptive field do we have after stacking $n \times n$ convolutional layers with kernel size $k \times k$ and stride $1$? Layers numeration starts with $1$. The resulting receptive field will be ...
2
votes
2answers
121 views

Can we use two convolutional layers without a activation layer in middle of them?

Recently I trained a model for more than 100,000+ images. I forgot to use a activation layer in between two convolutional layers, but model trained better with good accuracy (99%). So I want to know ...
1
vote
0answers
21 views

Is it better to use activation (ReLu) layer after Global Average pooling layer?

I have seen most have used non-linearity -> GAP -> Dense. But in some have used following order; GAP -> non-linearity -> Dense. Which is better? Is it good to use activation layer after GAP? (Here, ...
0
votes
1answer
24 views

CNN for image classification, new image representation as input

i'm trying to classify image pattern with CNN; I started to optimize a neural network with image represented in cartesian coordinate. If I use image represented in polar coordinate should i totaly ...
1
vote
0answers
17 views

Pixel-wise classification of black & white high resolution images

I know of methods using CNNs with fully connected conditional random fields as additional interpolation, or fully convolutional networks which seem to be brutally slow. So I do not want to use these ...
0
votes
0answers
8 views

When doing pca on a specific cnn feature map, is the pca substance always the same regardless of the input?

Lets say I have a pretrained CNN, and I extract the feature map from one of the layers for some input data x. I do the same thing for for a second set of input data y, which is very different from ...
0
votes
0answers
19 views

Dealing with images of variable resolution in CNN autoencoders

Let's suppose would like to build a CNN autoencoder that would be able to turn greyscale images into coloured ones. The final model should be able to accept images of any resolution. Also, note that ...
1
vote
0answers
21 views

The reason for multiple bounding boxes use for each grid cell - YOLO

I'm not quite sure what is the main reason why YOLO uses multiple bounding boxes for a grid cell. An answer I can find on web is for multiple aspect ratio in the prediction. However I don’t think it’s ...
0
votes
1answer
28 views

Determine clusters for the encodings of a Siamese Neural Network

I implemented a Siamese Neural Network that encodes images of different objects and outputs "coordinates" for each image in a lower dimension. My goal is to measure how good the network is clustering ...
0
votes
0answers
77 views

When to give up CNN? [duplicate]

i'm implementing my first CNN for image classification in a field with very few research. I'm aware i could extract feature and then try SVM, knn... But i want to be sure CNN is not a viable ...
2
votes
1answer
44 views

In Convolutional networks, how to do input data normalization? is it necessary?

I'm wondering about data normalization in CNN, how can we do it for the input images?, what can it add to the model's performance? and what are the main pre-processing techniques before doing the ...
1
vote
1answer
27 views

CNN - I know in advance which class my input cannot be

I have a CNN built with Keras, based on a multilayer perceptron. The last layer is softmax. ...
0
votes
0answers
12 views

Deep Learning Model for Complicated Pattern REcognition

I am using transfer learning using ResNet50 for snack packets recognition. They are one and another similar in dominant color and shape. Those like in images below. I have about 33 items to ...
0
votes
1answer
12 views

How does one use Convolutional Neural Nets (CNNs) on varying size sentences for NLP so that the final fully connected layer can remain fixed size?

I wanted to use CNNs to classify sentences. The sentences are varying length. I am going to use standard Word Embeddings (any sort of pre-trained vectors) as features for each word then concatenate ...
0
votes
2answers
41 views

How is RELU used on convolutional layer [duplicate]

I know that when dealing with artificial neural networks, RELU yields a value based on the weighted sum of the inputs plus a bias term. However, this logic does not seem to apply to convolutional ...
0
votes
0answers
41 views

Convolutional Conditional Variational Autoencoder Implementation

This may be a rather trivial question, but I am somewhat confused. I have been able to implement a convolutional variational autoencoder. I have also been able to implement a conditional variational ...
0
votes
0answers
10 views

Normalising predictions across datasets

I am currently training a model to predict a binary attribute. The model gives the output in range [0, 1]. The metric is TPR@FPR, e.g. I need to achieve maximum ...
0
votes
1answer
35 views

How can we determine the appropriate number of hidden layers, kernels in convolutional neural network (CNN)?

I have checked a lot of questions here and in other websites. What I concluded is that there is no rules for choosing the right number of hyper-parameters in CNN, all what can we do is just trying ...
0
votes
0answers
20 views

Why can we remove FC layers in VGGNet?

From https://cs231n.github.io/convolutional-networks/: VGGNet. The runner-up in ILSVRC 2014 was the network from Karen Simonyan and Andrew Zisserman that became known as the VGGNet. Its main ...
0
votes
1answer
16 views

downloadable weights of VGG-16 during ImageNet training

Does anybody know a place from where it is possible to download the weights of VGG-16 at different epochs, along a succesful training on ImageNet? The ideal situation would be to have downloadable ...
17
votes
4answers
4k views

Is it a good idea to use CNN to classify 1D signal?

I am working on the sleep stage classification. I read some research articles about this topic many of them used SVM or ensemble method. Is it a good idea to use convolutional neural network to ...
0
votes
0answers
25 views

How to handle maxpool layer backpropagation with recurring max values in same position

Say I have a layer a: 3 4 2 1 5 0 8 6 4 The maxpool using 2x2 filter is: <...
2
votes
1answer
26 views

Do fully connected layers in the middle of a network impede optimization?

I submitted a paper that uses an auto-encoder network with several convolutional layers in both the encoder and the decoder and a fully connected layer (FCL) in between. Besides the FCL being useful ...
1
vote
0answers
17 views

What autoencoder architectures are effective for large images?

I've had easy success in the past making autoencoders for small images using necked down dense networks, but my new application has images of ~1M pixels, which is impractical to address (I think) with ...
1
vote
1answer
45 views

How can we relate the concepts of GAN/cGAN in SRGAN? Is SRGAN a Conditional GAN?

I have been reading and looking at implementations of the SRGAN, from "Photo-realistic Single Image Super Resolution with Generative Adversarial Networks" paper. One thing that I noticed is that the ...
0
votes
0answers
84 views

Very High Training Accuracy and very low Testing Accuracy CNN

I'm using 3 layer CNN with 8, 16, and 32 filters, each of size 5 X 5. I'm getting an training accuracy of 99.97%. Testing accuracy of 41.11%. Total classes: 605 Train Set: Each class has 7 samples ...
0
votes
0answers
9 views

Handle loss while converting high dimensional image to specific size in VGG 16

I am training a VGG16 net using transfer learning. I have removed the fully connected layers and used fine tuning to classify objects into few categories but I have faced below problems: 1.I have ...
0
votes
0answers
14 views

Augmentation of data collected from single stationary source

How do one augment data that is being collected from single stationary sensor source. The orientation, color and size always remain same. Only the pattern in the dataset vary (example : sunspots are ...
0
votes
0answers
25 views

Solving an equation that involves convolution and batch normalization

Basically the equation is $$ x = \mathrm{BN}(\mathrm{Conv}(z)) $$ where the convolution operation uses SAME padding with stride $1$ ensuring that the input and output have the same size. The ...
1
vote
2answers
39 views

Extending a neural network to classify new objects

Suppose a model M classifies apples and oranges. Can M be extended to classify a third class of objects, e.g., pears, such that ...
0
votes
0answers
7 views

CNN analysis help

http://scs.ryerson.ca/~aharley/vis/conv/ I'm trying to better understand the architecture in this CNN. After reading the paper here: http://scs.ryerson.ca/~aharley/vis/ the author says This ...