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.
1,483
questions
0
votes
0
answers
26
views
Loss stops dropping in a convolutional classification network [closed]
I'm training my first CNN, but loss stops at 1, and its plot also shows spikes:
This is the code:
...
- 1
0
votes
0
answers
13
views
The accuracy of my validation set is always the same
I am training a CNN model which is used for a multi-label classification task. My training data set has 5000 data points, every data point is a 100000 long 1-D array. So the shape of my training set ...
James harden
0
votes
0
answers
14
views
Searching for a proper way to reduce the dimensionality of activations from a CNN
I am conducting an analysis to compare the similarities between different images across early and late layers in a CNN. The model I am working with is the pretrained DenseNet121 that comes with ...
- 33
1
vote
1
answer
26
views
Convolutional Neural Networks - Flattening with multiple feature maps
I have a very simple question about CNNs, which I unfortunately couldn't find an explanation for.
Imagine we have a CNN, that has four filters (eg right, left, top, bottom edges) each of those outputs ...
- 13
0
votes
0
answers
6
views
Fully Convolutional Networks: Fully Connected Layers as Convolutional Layers
I'm reading the paper "The Fully Convolutional Network" and I don't understand this passage in 3.1:
Typical recognition nets, including LeNet [21], AlexNet [19], and its deeper successors [...
- 1
0
votes
0
answers
46
views
How can I feed data files of different sizes to a CNN?
I have 5022 data files. I want to feed them into a CNN model.
However, the lengths of data files vary from, say, 50 to 1015 rows (the number of columns is constant).
In the case of image files, we can ...
- 2,126
0
votes
0
answers
16
views
Help with multimodal (hydra) CNN architecture
I am trying to learn hydra like CNN architecture based on publication
using AFLW dataset. I replicated results from repository and its fine. Moving forward, I added one another head/regression task ...
- 1
0
votes
0
answers
28
views
How can I reduce fluctuations in my validation accuracy?
I'm training a CNN with pictures data for binary classification and while my training accuracy increases, my validation accuracy keeps fluctuating between small and high values of accuracy. I have a ...
- 1
0
votes
0
answers
9
views
Backpropagation Derivation for CNNs
I’m looking for a good reference (be that textbook, website, article…) that goes through a derivation of backpropagation in a multi-channel convolutional neural network. I’m also looking for a ...
- 165
0
votes
0
answers
26
views
Formula/Proof: How many times must maxpooling (3x3 kernel, stride 2x2) be applied for a random binary mask (dxd size) to have only 1 values?
Given a d x d array, 1% of which contains the value 1 and all remaining locations contain the value 0. (e.g. a 128 x 128 array would have 164 values equal to 1 and 16220 values equal to 0). What would ...
0
votes
1
answer
35
views
Creating a CNN model for multi-output prediction where one target variable is categorical, and others are numeric
I want to create a simple CNN model for multi-output prediction. The predicted values are four numeric values (all between 0-1) and one categorical value (4 classes). When I try to create a model ...
- 1
0
votes
0
answers
31
views
How to Change Architecture of DCGans?
https://pytorch.org/tutorials/beginner/dcgan_faces_tutorial.html
I was refering this notebook but default size is 64*64
I want to change architecture to 256 or 512
Can anyone help me with training
...
0
votes
0
answers
18
views
VAE for Motion Sequence Generation - Convergence Issue with Scheduled Sampling
I implemented a Variational Autoencoder (VAE) in PyTorch for motion sequence generation using human pose data (joint angles and angular velocities in radians) from the CMU dataset. The VAE ...
- 1
2
votes
1
answer
208
views
Is batching needed for the test set?
I'm just starting to learn about CNN (convolutional neural networks). Does the test data also need to be divided into batches, similar to how it's done with the training data?
- 21
0
votes
0
answers
35
views
Image processing: Inferre rotation angle of tilted rectangle on noisy background
I have many "grayscale" images i.e. 2d-arrays like the following:
i.e. dark rectangles which are tilted by an angle $\alpha \in [-3^\circ, 3^\circ]$ and a bright but noisy background. I ...
- 133
1
vote
1
answer
69
views
Probability in a classification problem
In a classification problem (let's say of two categories, cat and dog) with a softmax output, does the probability have any physical meaning other than assigning the category to the input based on the ...
- 111
0
votes
2
answers
49
views
Can not understand a column in a paper about CNNs
I am reading the SqueezeNet paper and I do not get the parameter depth here:
There isn't a description under the table, and the only extra mention of the parameter is that it means the number of ...
- 135
0
votes
0
answers
13
views
CNN kernels similarity
I know some theory about deep neural network, cnn and back propagation in general. I am fascinated by the power of these technologies. I try to understand also the math aspects.
For example the fact ...
- 1
0
votes
0
answers
6
views
My CNN seems to reach a certain loss threshold, then stops learning from there [duplicate]
I wasn't sure how to phrase this question, but yeah.
My CNN (code here) keeps getting up to an accuracy of around 66% until it stops learning anymore. It just keeps fluctuating around the 66% range. I'...
0
votes
0
answers
14
views
Shape of the flattened vector in CNN
If I have a max-pooled convolution layer of dimension (5,5,4), means 4 no. of 5x5 feature maps, what will be the shape of the flattened vector after applying ...
2
votes
1
answer
75
views
How to choose the output vector size for metric learning? [closed]
In metric learning of e.g., MNIST images, a CNN projects a 28 x 28 image into a $d$-dimensional vector which gets passed to a metric learning loss function: minimize the Euclidean or cosine distance ...
- 23
0
votes
0
answers
62
views
Discriminator and Generator's losses in a GAN
I'm a little confused about the Generator's and Discriminator's losses while training a cGAN. I am aware that a stable GAN is one where the Discriminator's loss reaches 0.5. I know that we can read ...
- 15
1
vote
1
answer
75
views
Is it wrong to view convolution as template matching?
I am reading about the convolution operation but I can't see how it can be seen as template matching.
Suppose that we convolve the input $\mathbf{X}$:
$$
\begin{bmatrix}
1 & 0 & 0 & 0 \\
0 ...
- 452
0
votes
0
answers
15
views
How to improve the expressibility of a cGAN
I am training a cGAN on the problem of reconstructing a density matrix. My inputs to the network are matrices and expectation values. That is, I have a set $(A,x)$ where A are measurement operators, ...
- 15
1
vote
0
answers
9
views
CNN: adding a third class improves the overall ranking of the network
I am working on a Convolutional Neural Network for classifying two classes of images whose difference between them is very small. Running the CNN (using PyTorch), it was able to correctly classify ...
- 253
3
votes
1
answer
187
views
Can SVM and Decision Trees be seen as instances of neural networks?
We already know that neural networks with specific choices of activation function as well as connections can generalize large amount of ML models.
My question is: neural network also generalize SVM ...
- 1,430
3
votes
1
answer
81
views
Can translation invariance be achieved by just a global pooling layer?
I am trying to understand the purpose of the max pooling layers that are insterted between intermediate convolutional layers. As we know, the outputs of convolutional layers are translational ...
- 452
1
vote
0
answers
20
views
Fluctuating validation accuracy with steady accuracy increase
I have four layers of CNN to predict Javanese script letter data. The training accuracy and loss monotonically increase and decrease respectively. But, my test accuracy starts to fluctuate wildly. I ...
- 11
2
votes
0
answers
25
views
Does a 1x1 convolutional Layer have a bias (Inception Modules)?
This question is regarding to the 1x1 convolutional layer idea from the paper
"Going Deeper with Convolutions"(https://arxiv.org/abs/1409.4842) that describe the idea of so called inception ...
- 21
0
votes
0
answers
249
views
How is the backbone and head of YOLOv8 connected?
I have view this .
But C2f only has single output tensor.
What is the meaning of the connection between Backbone and Head in the diagram? Is it a copy of output ...
0
votes
0
answers
9
views
Does it makes sense to prune filters in convolutional layer when rank of filters is smaller than number of filters?
Let’s consider a convolutional layer, taking filters (matrices used in convolution), vectorizing them, putting them into matrix $A$ and computing it's rank. Does it makes sense to remove $|filters|-...
- 498
0
votes
0
answers
20
views
FastAI error rate still at 1.0 while validation and training loss are decreasing
I'm using fastai to run some experiment models. I've downloaded the Kaggle fingers dataset and with that ran the code below. What I don't understand is why, although ...
- 191
0
votes
0
answers
90
views
Modify normal TCN to seq2seq TCN for binary classification
I'm working in PyTorch and I would like to use a Temporal Convolutional Neural Network (TCN) to perform binary classification for each timestep of an input sequence, therefore a Sequence-to-Sequence (...
- 101
1
vote
0
answers
43
views
What is the best strategy to determine the final weights of the model after doing k-fold cross-validation?
Suppose we are building a CNN model. After we do 5-fold cross-validation and examine the accuracies and other statistical tests, how do we choose the final weights of the model?
Should we choose the ...
- 11
0
votes
0
answers
22
views
A simple neural network for classifying datasets
I want to train a neural network that is as simple as possible, that is able to classify datasets.
I have 5 MedMNIST datasets that each contain a number of labels, and with complex features.
I'm ...
0
votes
0
answers
26
views
What is a feature of an image?
Given an image to train on a CNN what would be considered the features that are the inputs of the model? Would each individual pixel be a feature or would the R,G,B channels be the features?
Lets say ...
0
votes
0
answers
19
views
The validation loss and training loss are low and close to each other but there exists high test loss?
I have been using CNN-LSTM for action recognition with dataset split 70% for training , 20% for validation , 10% for testing and after training , the validation loss and training loss were very close ...
0
votes
0
answers
9
views
How the skip connections occur in resnet50 architecture when input output layers are different? [duplicate]
I am having difficulty understanding how the skip connections occur in resnet50 when the input and output layers are different in shape.
For example, in the first residual block, a 56*56*64 size ...
1
vote
0
answers
61
views
How can I put GNN and CNN together?
I am trying to combine GNN and CNN in my model.
Every node of my graph has 2d space coordinates, so as a whole it's like an irregular 2d mesh.
I think we can't use deep GNNs due to oversmoothing. So I ...
- 111
1
vote
0
answers
119
views
Batch normalization before or after channel-wise concatenation?
I have a block in a CNN that splits the input channel-wise in half, and one half goes through a regular 3x3 2d convolutional layer, and the other goes through a dilated 3x3 2d convolutional layer. ...
- 11
0
votes
0
answers
78
views
Input Image to U Net with Pretrained ResNet34 Encoder
Most of the pretrained architecture accept the input image size 224x224, but do we have to always resize our images to 224x224?
I try to do both resize 224x224 and 512x512, the first resolution give ...
- 21
2
votes
1
answer
55
views
Augmenting Image Data on Small Dataset [duplicate]
I want to train image segmentation model using U-Net with pretrained ResNet34 as encoder. My dataset is really small, i separate it with
Train data : 57 images
Validation data : 16 images
Test data : ...
- 21
0
votes
0
answers
7
views
Are the two formulas for computing the output shape of a convolution equivalent? Computing the floor before or after adding $1$?
I've come across two different formulas in my studies to calculate the output shape of a convolution.
Below, $I$ is the input image size, $K$ the filter size and $S$ the stride.
$$
\lfloor \frac{I - K ...
- 290
0
votes
0
answers
18
views
How to make the generated machine learning model satisfied with specified proportion
If an image is composed of three categorical variables with fixed proportions (such as 0.2,0.3, and 0.5), how could I ensure the generated image (such as GAN model) is satisfied with this proportion?
- 61
0
votes
0
answers
119
views
Replacing a fully connected layer with a 1x1 convolution vs with fxf convolution
Suppose an input of shape (width x height x channel_num) = (10 x 10 x 15) is obtained from previous convolutional layers, and this input is about to be inserted into a fully connected (fc) layer of K=...
- 37
2
votes
1
answer
302
views
Do we lose information when we normalize an image? [closed]
Before training a machine learning algorithms, it is advisable to perform feature scaling. Suppose we have a "toy" dataset where each image is composed of two pixels $x_0$ and $x_1$. Lets ...
- 452
0
votes
0
answers
33
views
What can convolution network without activation function contribute to the DL model if it is the last layer?
I recently saw some paper about stereo matching :
End-to-End Learning of Geometry and Context for Deep Stereo Regression [1]
https://openaccess.thecvf.com/content_ICCV_2017/papers/Kendall_End-To-...
0
votes
0
answers
16
views
Performances of train/test split vs train/validation/test split
Despite there are multiple questions about it, I cannot figure a solution about my problem. I have built a simple neural network classifier on the MNIST database. I have divided it in training, ...
- 51
0
votes
2
answers
59
views
Model giving accuracy of 89% after training it by CT scan Images. and giving accuracy of 35% on testing. and low precision, recall, f1-score [closed]
I am training a model for cancer detection by using chest CT scan Image. training set is 70% testing set is 20% validation set is 10%.
Data contain 3 chest cancer types which are Adenocarcinoma, Large ...
0
votes
0
answers
41
views
How can I help my model's batch-normalization layers converge?
I have a very deep convolutional network that I am training and found the batch normalization layers to fail to learn a good representation of the data.
My dataset consists of several subsets of ...
- 290