Questions tagged [convolution]

For use when discussing the commutative and linear, but not associative operator interpreted on functions and distributions.

Filter by
Sorted by
Tagged with
0
votes
0answers
35 views

TF/keras implement residual block

I read several papers, where they propose to implement residual blocks of ResNet as follows $$ u^{k+1} = u^k - \tau K^T \sigma(K u^k), $$ where $u^{k}$ denotes output on k-th layer, $\tau$ is ...
1
vote
1answer
39 views

Why is the kernel of a Convolutional layer a 4D-tensor and not a 3D one?

I am doing my final degree project on Convolutional Networks and trying to understand the explanation shown in Deep Learning book by Ian Goodfellow et al. When defining convolution for 2D images, the ...
0
votes
1answer
13 views

Performing 1D Depthwise conv using Keras 2D Depthwise conv

I would like to perform a 1D Depthwise convolution (ie the first step of the depthwise-separable convolution) for a machine learning model I am working on. This means that for an input activation ...
0
votes
0answers
13 views

Representing multi-channel input signals with a single signal

I am working on an EEG signal classification problem. My dataset consists of EEG signals stored as 19X30000 NumPy arrays. Each row represents a single channel. For now, I am converting each of the ...
0
votes
1answer
34 views

How to calculate convolution for 2nd conv Layer in CNN, Do we need to average across all feature maps?

I understand that for the first layer (assuming we have a grayscale image) we calculate the convolution of 3*3 receptive field as a weighted sum of receptive weights with pixels $ x1 · w1 + x2 · w2 + ...
1
vote
0answers
28 views

Understanding, visualizing and interpreting CNN activations

I am working with the first layer of a CNN and trying to understand how to interpret the activation output. My CNN takes input from 3 channels (RBG picture) and the first layer is ...
1
vote
0answers
149 views

Understanding scipy.signal.convolve2d full convolution and backpropagation between convolutional layers

I'm learning about convolutional neural networks. The convolution operation in order to extract features that is described in literature and posts used for this is quite intuitive and easy to ...
0
votes
1answer
91 views

What will be the input_shape of tf.keras.layers.Conv3D be for these inputs

I have many videos, and each video is made up of 37 images (there are 37 frames in the whole video). And the dimension of each image is (100, 100, 3).... So the ...
1
vote
0answers
14 views

Finding optimal time series using convolution [closed]

we logged sensor data while milling a workpiece. At several points, the workpiece was damaged and this induced a certain sensor data time series. Due to noise and since its a real world measurement, ...
0
votes
0answers
33 views

Pytorch: Reduce forward prediction dimensions of GRU network / Improving Network Architecture

I am currently working on a GRU network to predict a time series, please note that I am completely new to machine learning and pytorch. Also I have never had a formal education in programming. This ...
1
vote
0answers
22 views

Trade-off between number of channels and size of convolutional filters

As far as I understand, the common practice in the modern CNN architectures is to use a smaller convolutional filters, but deeper networks with more channels. One of the reason behind this is that one ...
1
vote
0answers
17 views

Are convolutions in deep learning associative?

Let's denote "convolution in deep learing" as "convolution-deep", and "convolution in math or signal processing" as "convolution-math". As we all know, ...
1
vote
0answers
25 views

Padding in Convolution Formula

Why is it that the formula for each element in a convolution between an image $I$ and a $k \times k$ sized kernel $K$ is $$ (I*K)_{ij}=\sum_{m=0}^{k-1}\sum_{n=0}^{k-1}I_{(i-m),(j-n)}K_{mn}=\sum_{m=0}^{...
0
votes
0answers
16 views

Understanding the convolution formula

According to several sources this formula, or the center originated version of it, is used to calculate an element of a convolution between an image $I$ and a kernel $K$ of size $k \times k$: $$ (I*K)...
3
votes
1answer
30 views

Does a Convolutional Layer in a Neural Network learn the correlation between its input signals via its kernel?

I am interested in the theory behing what a convolutional neural network learns with its convolutional operations. I think it learns (useful) kernels which measure the correlation between its input ...
0
votes
1answer
45 views

What features used by CNN model should a feature store actually store? [closed]

According to MLOPs principle, it is recommended to have a feature store. The question is in the context of doing image classification using deep learning models like convolutional neural networks ...
1
vote
0answers
18 views

Are 3D kernels in convolutions summed over their channels?

Say for example that I have a 28x28x1 grey scale image and I will perform two consecutive convolutions. The first convolution has 2 3x3x1 filters and the second has 3 3x3x2 filters. Each convolution ...
1
vote
0answers
14 views

Utilizing 1x1(x1) convolutions as a learned max pooling (3D)?

I have a semantic segmentation network that ingests 3D images (hyperspectral $(x, y, b)$) and predicts 2D images (semantic map $(x, y)$). This network takes the form of a classic UNet, though it ...
0
votes
0answers
17 views

Changing order of input dimension in Tensorflow 3-D Layers

According to the official documentation of tf.keras.layers.Conv3D 5+D tensor with shape: batch_shape + (channels, conv_dim1, conv_dim2, conv_dim3) if data_format='channels_first' or 5+D tensor with ...
1
vote
0answers
13 views

Can I say that a trained neural network model with less parameters requires less resources during real world inference?

Let us imagine that we have two trained neural network models with different architectures (e.g., type of layers). The first model (a) uses 1D convolutional layers with fully-connected layers and has ...
1
vote
0answers
15 views

Comparison of different ways of Upsampling in detection models

There are various ways to increase the resolution of tensor in (width, height) dimensions, frequently used in detection models like ...
0
votes
1answer
22 views

Should kernel size always be a prime number?

Should kernel size always be a prime number? E.g. (3,3) (5,5) (7,7). While tinkering with sklearn.preprocessing.KernelCenterer(), I noticed that I could only get it ...
0
votes
0answers
16 views

Notation of Transposed Convolution Operation in Equation

How do I notate a transposed convolution operation (as it is used in deep learning), in a math equation? A convolution operation for example is often notated as $\hat{y} = x \circledast W$ where $W$ ...
0
votes
0answers
21 views

About the relevance and interprertability of convolutional filters?

Convolution filters are known to perform very well in tasks, concerning some work with the image or video data, due to their ability to preserve some spatial information and equivariance property ...
0
votes
0answers
6 views

Autocorrelation of two functions multiplied and raised to arbitrary powers

Given a signal $A$ and a signal $B$ with autocorrelation times of $\tau_A$ and $\tau_B$, respectively, where $\tau_A > \tau_B$, is there any general statement that can be made about the ...
0
votes
0answers
17 views

Extended ResNet

The success of ResNet is mainly ascribed to the fact that it is very easy for deeper networks to learn the identity function hence there's little risk when the network becomes too deep. I was ...
0
votes
0answers
12 views

Will disparity in image format/quality between binary classifications affect training of Convolutional Neural Network?

I have an image dataset containing two classes. One of the classes has many images and they are all JPG images with the following format: ...
1
vote
1answer
32 views

Error after merging two Deep Learning models VGG16 and ResNet50

I have merged two different models namely VGG16 and ResNet50 and given the outputs of the two models as input to another model. I have checked the Layers graph is correct. Before merging the code was ...
1
vote
1answer
19 views

Adapting ZFNet on 2244x224 image using a filter 7X7

I am building a model based on ZFNet in Tensorflow 2.0. I am using the Petal images dataset. The images are of size 224x244x3. So my question is when implementing the first layer (conv2d) with filter ...
1
vote
0answers
31 views

There are 2 figures explaining transposed convolution. Which one is correct?

I have been struggling to understand transposed convolution. When I search for "transposed convolution", there are 2 figures explaining transposed convolution in which I think are not ...
0
votes
0answers
21 views

Why is everybody using `mobilenetv2` for mask detection?

I was looking for good pre-trained models to be used for mask detection and I found resnet50 and mobilenetv2 (lots of times). ...
1
vote
1answer
10k views

ValueError: Input 0 of layer sequential is incompatible with the layer: expected ndim=3, found ndim=4. Full shape received: [None, 25, 25, 1]

I am trying to use conv1D but getting that error. My dataset's is batched and has a shape of [None, 25, 25, 1] I am using input_shape=(25,25) I am not able to figure out what should I change so I can ...
0
votes
1answer
213 views

Can anyone recommend me a very good pre-trained model for face or head detection?

I really need to know the best pre-trained models to detect faces and/or peoples' head. Not a face recognition model, but only to classify whether an object is a person's head/face or not. I'm ...
1
vote
0answers
16 views

Can convolutional network learn structural properties of one feature w.r.t to other?

I'm going through the literature on pose-estimation ( DeeperCut, OpenPose, MultiPersonPosetrack). I'm interested in knowing whether these networks/ generally a CNN can learn properties (geometrical) ...
1
vote
0answers
39 views

What does it mean to say convolution implementation is based on GEMM (matrix multiply) or it is based on 1x1 kernels?

I have been trying to understand (but miserably failing) how convolutions on images (with height, width, channels) are implemented in software. I've heard people say their convolution implementation ...
1
vote
0answers
100 views

Understanding image size changes in DCGAN

I have been studying and trying to implement Generative Adversarial Networks using PyTorch. More precisely I tried to replicate the DCGAN PyTorch Tutorial tutorial using some custom dataset. My code ...
2
votes
0answers
15 views

Strategy for improving performance of 3D convolutional GAN

Others working with neural nets and GAN's might find this question interesting. Background: I've been working with data from Berkeleys PEER Ground Motion Database to generate new novel seismic traces. ...
2
votes
1answer
31 views

Can I tune a model after training it? (Convolutional Neural Network & Classification)

I am relatively new to Data Science and I've recently embarked on a project. Long story short, I've trained a CNN model to distinguish between Male and Female genders. However, I wish to tune my model....
0
votes
0answers
27 views

How to claim that a CNN model is lightweight?

CNN model has some parameters that can show the a model is lightweight compared to others. The parameters can be Size(after training), Trainable parameters or multiply add. Which parameters are ...
0
votes
1answer
56 views

How to perform upsampling (and NOT interpolation) process theoretically modelled?

As an example, I know that sampling a signal $s$ is modelled by multiplication of s by a dirac comb, which has the effect of convolving the Fourier Transform (FT) of $s$ by the FT of the dirac comb ...
0
votes
0answers
11 views

How to Convolve a High-Res Image by a (fully convolutional) CNN kernel?

My CNN is an extremely simple neural network. ...
0
votes
1answer
105 views

How to backpropogate Convolution layer padding inputs with respect to output derivative

I created a convolution network with 5 Conv blocks, let discuss the issue based on Conv block 4 & 5 Conv Block 4 Input Image size : 28 * 28, Padding size 1 : 30 * 30 (image size ...
0
votes
1answer
369 views

PyTorchs ConvTranspose2d padding parameter

Im confused about what PyTorchs padding parameter does when using torch.nn.ConvTranspose2d. The docs say that: "The padding argument effectively adds dilation * (kernel_size - 1) - padding ...
0
votes
0answers
25 views

Why is the accuracy on the test dataset very low when training a neural network on an IMU dataset?

I am trying to train an IMU (Inertial Measurement Unit) dataset. The dataset contain 6 features (3-gyro, 3-accelerometer) and 1 label column. I have build a neural network via Conv1D, LSTM and Dense ...
0
votes
1answer
65 views

weight sharing among neurons at same depth

I'm trying to understand some visual illustrations about the wight sharing in the Convolutional Neural Network as following: In this picture we see that for different outputs different inputs share ...
0
votes
1answer
21 views

Up to which layer can we consider the encoder to be?

I'm trying to extract the encoder from a U-Net network. Given its architecture: And its summary: ...
0
votes
0answers
29 views

Few shot learning in object detection

I am working on a project to detect buildings from satellite imagery in Tanzania using convolutional neural nets. I use a pre-trained model which I further train on a selected area in Tanzania. The ...
8
votes
4answers
1k views

Understanding how convolutional layers work

After working with a CNN using Keras and the Mnist dataset for the well-know hand written digit recognition problem, I came up with some questions about how the convolutional layer work. I can ...
2
votes
0answers
112 views

How is it possible to upsample 2x with a 3x3 convolution?

From the Pytorch docs on Conv2Transpose2d, the formula to compute the output of the upsampled convolution (assuming square input and no kernel dilation) is: $$H_{out} = (H_{in} - 1) \times S - 2P_{in}+...
0
votes
0answers
27 views

Majority of feature maps of CNN are black

Assuming we have a following CNN : Conv->MaxPool->Conv->Maxpool->Linear. What does it mean - intuitively - if the majority of the feature maps of the first convolutional layer are black i....

1
2 3 4 5 6