Convolution - Science topic

Please see a convoluted discussion that might change our perception of world literature. Could the heralded literature revolution be what we need to stop the wars and achieve global unity? Please explore the FaceBook comments and tell me your thoughts:

How do we typically choose between Convolutional Networks and Visual Language Models when it comes to Supervised Learning tasks for Images and Videos ?

What are the design consideration we need to make ?

To cite my article, I need some reference papers based on only CNN models using 5 or 7 convolution layers for the breast cancer binary classification in the BreakHis histopathology image dataset.

For smoothening of noisy CD spectra, several fitting models are available in Jasco spectra manager software, like Savitzky Golay, binomial, means movement and adaptive smoothening. Each one asks for convolution width as an input ranging from 5 to 25. Based on the model that we choose and the convolution width, the output smoothed spectra changes and accordingly the secondary structure content varies. Thus, my question is which model is ideal and widely used in industry and why?

In CNN(convolution neural network), can the feature map obtained determinately by a random initialization convolution kernel? if not, how to decide the weights in convolution kernel to obtain the feature maps we need? By trial and rerror, are we shotting if our eyes closed?

Are the weights of convolution kernel in CNN(convolution neural network) pre-desighable before trainning?

How can attention mechanisms be integrated with convolutional neural networks to enhance performance in image classification tasks?

Why is EfficientNet better than other CNN architectures ?

I found these soft-sediment sedimentary structures a month ago working in the Colombian Pacific coast and I think we might be looking at actual seismites, but I´m not an expert. The above and below lithology is mudstone and the convoluted one in the middle is sandstone, all Neogene. The area is well known for its seismic activity. Any thoughts?

Explore the fundamental role of convolution in signal processing, specifically its significance in comprehending the behavior of linear time-invariant systems. Seeking insights on its applications and implications in system analysis.

Seeking insights on optimizing CNNs to meet low-latency demands in real-time image processing scenarios. Interested in efficient model architectures or algorithmic enhancements.

..

..

..

An application based on convolutional neural networks (CNNs) in oncology can be a powerful tool in diagnosing and treating cancer. CNNs are a type of deep learning algorithm that can be trained to recognize patterns and features in images, making them particularly useful in medical imaging applications.

One example of an application based on CNNs in oncology is the detection of lung nodules in computed tomography (CT) scans. Lung nodules are small masses in the lungs that can be a sign of cancer, and early detection is crucial for successful treatment. A CNN can be trained to identify the presence of nodules in CT scans with high accuracy, allowing for early detection and intervention.

Another potential application is in the classification of breast tumors using mammography images. CNNs can be trained to distinguish between benign and malignant tumors with high accuracy, reducing the need for unnecessary biopsies and improving patient outcomes.

Furthermore, CNNs can be used to analyze histopathology images, which are images of tissue samples that have been stained and examined under a microscope. These images can provide valuable information about the structure and composition of tumors, helping to guide treatment decisions. A CNN can be trained to recognize specific features of cancerous cells and tissues, enabling faster and more accurate diagnosis and treatment.

In conclusion, an application based on convolutional neural networks in oncology has the potential to revolutionize cancer diagnosis and treatment by providing faster, more accurate, and more personalized care.

I'm currently working on a project that involves combining LSTM (Long Short-Term Memory) and CNN (Convolutional Neural Network) architectures using PSO .

I'm seeking more guidance on the best practices and considerations for effectively combining these techniques

I would like to add some questions that made me stuck for sometimes. So i have an issue in imbalance segmentation data for landslide modelling using U-Net (my landslide data is way less than my non-landslide data). So my questions are:

1. Should i try to find a proper loss function for the imbalance problem? or should i focus on balancing data to improve my model?

2. Some suggest to use SMOTE (oversampling), but since my data are images (3D) i have found out that it is not suitable to use SMOTE for my data. So, any other suggestions?

Thank you,

Your suggestions will be appreciated.

Im trying to create an image classification model that classifies plants from an image dataset made up of 33 classes, the total amount of images is 41,808, the images are unbalanced but that is something me and my thesis team will work on using Kfold; but going back to the main problem.

The VGG16 model itself is from a pre-trained model from keras

My source code should be attached in this question (paste_1292099)

The results of a 15-epoch run is also attached as well

what I have done so far is changing the optimizers from SGD to Adam, but the results are generally the same.

Am I doing something wrong or is there anything I can do to improve on this model to get it to atleast be in a "working" state, regardless if its overfitting or the like as that can be fixed later.

This is also the link to our dataset:

It is specifically a dataset consisting of Medicinal Plants and Herbs in our region with their augmentations. The are not yet resized and normalized in the dataset.

Monitoring of fabric material is essential for the creation of textiles. Defective textiles have cost the apparel business a lot of money. In less developed nations like Bangladesh, most of the production faults in synthetic fibers are often discovered through manual examination. The work of an investigator is difficult and time-consuming. At the moment, AI is increasingly applied. since AI already has triggered a shift in many sectors. It is unquestionably a wider foundation in terms of our technological advancements. For picture data, convolutional neural networks are helpful. AI is now a growing industry, and we all are well aware of how challenging it is to identify Cloth Defects today. We ultimately decided to use a new approach to discover textile defects using AI. For our nation, it is crucial. since the clothing business is well known in our country. In our nation, many individuals rely on it to alleviate unemployment. Therefore, it is crucial for us to ship quality cloth. However, it might be difficult for individuals to verify the material and find flaws in all of these clothes during quality control. Even if earlier generations had to examine the cloth’s quality, there were still a lot of errors. However, today's technology has simplified this process with its many speculative models, such as CNN.

Convolutional and recurrent architechtures made it possible to build effective models working on new data types such as images and sequential data. What is next?

In the context of Convolutional Neural Networks (CNNs) used for image recognition tasks, what is the primary advantage of incorporating convolutional layers into the architecture? How do these layers help capture meaningful features and patterns from images effectively?

I would like to know how you deal with the problem of flocculus’ fMRI signal asymmetry due to the convolution effect. Because the location of the flocculus is relatively deep and close to the base of the skull. Could you please give me some advices about this question and data analysis?

Thank you very much

I want to learn about Solving Differential Equation by using "Discrete Singular Convolution"method. I want to learn this method, if someone has hand notes it would be great to share with me. I need to learn that in 2 weeks. Thanks in advance.

Is it a good idea to extract features from pre-trained, the last 1x1 convolution removed U-NET/Convolutional Autoencoder? Data will be similar and the model will be trained for image segmentation. I know everybody suggests freezing the encoder is the best option but I think there is feature extraction in the decoder part too(In both convolutional autoencoder and U-NET). They are high-level feature extractors, if my data was different, the frozen decoder part wouldn't be a good idea. But what if my data is very similar?

I have seen the scale of at least 1000 s for cnn. I know it depends on many factors like the image and its details but is there roughly any estimate that can determine the number of samples is required to apply CNN reliably?

Have you seen 100 of images applied for CNN?

Although Convolutional Neural Networks (CNNs) are widely used for plant disease detection, they require a large number of training samples while dealing with wide variety of heterogeneous background. In this paper, a CNN based dual phase method has been proposed which can work effectively on small rice grain disease dataset with heterogeneity. At the first phase, Faster RCNN method is applied for cropping out the significant portion (rice grain) from an image. This initial phase results in a secondary dataset of rice grains devoid of heterogeneous background. Disease classification is performed on such derived and simplified samples using CNN architectures. Comparison of the dual phase approach with straight forward application of CNN or Faster RCNN on the small grain dataset shows the effectiveness of the proposed method which provides a five fold cross validation accuracy of 88.11%.

How do semantic segmentation methods (UNet), and instance segmentation methods (mask R-CNN) rely on convolutional operations to learn spatial contextual information?

..

سؤال مهم

In my research, I need to compare the neural networks I have built, consisting mainly of perceptron and normalization layers, with networks from other publications that have convolution, pulling, normalization and perceptron layers in terms of computational complexity. I have the ability to calculate the number of parameters a given neural network has on a given layer, but I don't know how I should compare it.

Should I take only convolution layers as the most taxing, or sum the number of parameters from all of layers?

How should I compare neural networks that have computationally stressful convolution layers with others that do not have them, but perform feature extraction in a different way?

Python code:

class HGNN(nn.Module):

def __init__(self, H, in_size, out_size, hidden_dims=hidden_dim):

super().__init__()

self.Theta1 = nn.Linear(in_size, hidden_dims)

self.Theta2 = nn.Linear(hidden_dims, out_size)

self.dropout = nn.Dropout(0)

# Node degree

d_V = H.sum(1).to_dense().double()#torch.sparse.sum(H, dim=1).to_dense()#

# Edge degree

d_E = H.sum(0).to_dense().double()#torch.sparse.sum(H, dim=0).to_dense()#

n_edges = d_E.shape[0]

# D_V ** (-1/2)

D_V_invsqrt = torch.diag(torch.pow(d_V,-0.5))

# D_E ** (-1)

D_E_inv = torch.diag(1./d_E)

# W

W = torch.eye(n_edges)

# Compute the Laplacian

self.laplacian = D_V_invsqrt.double() @ H.double() @ W.double() @ D_E_inv.double() @ H.T.double() @ D_V_invsqrt.double()

def forward(self, X):

Dr_float= self.dropout(self.Theta1(X)).to(self.Theta1.weight.dtype)

#lap_float = torch.tensor(self.laplacian, dtype=torch.float64)

lap_float= self.laplacian.to(self.Theta1.weight.dtype)

X = lap_float @ Dr_float

#X = X.to(self.Theta1.weight.dtype)

X = F.relu(X)

X= X.to(self.Theta1.weight.dtype)

Dr2_float= self.dropout(self.Theta2(X)).to(self.Theta1.weight.dtype)

X = lap_float @ Dr2_float

print(10)

# Add an activation function here

#X = torch.sigmoid(X)

return X

def compute_node_representations(self, X):

return self.forward(X)

I need the data for experiments of super-resolution. That is, low-resolution data can be get by convoluted Ricker wavelet with low-frequency, high-resolution data can be get by convoluted Ricker wavelet with high-frequency.

I am using a transfer learning approach for facial expression recognition. I wanted to know which pre-trained convolutional bases can provide better performance in classifying facial expressions.

Is it possible to classify PlanetScope Imagery using Convolutional Neural Networks in Google Earth Engine Code Editor? I would be highly grateful for any help in this regards.

What is the the main advantage of Autoencoder networks vs CNN?

Is the Autencoder network better than the convolution neural network in terms of execution time and computational complexity?

Four layer parameters of CNN are given in the table (figure attached). The first convolution layer has 1x9x9 input. After applying the 20 filters of size 4x4 how does the output 5x5x20?

can anybody help me to understand why the 5x5 output is?

Hi everyone, I'm attempting to code the Tacotron speech synthesis system from scratch to make sure I understand it. I'm done implementing the first convolutional filterbank layer and have implemented the max pooling layer, but I don't understand why the authors of chose a max-pooling over time with stride 1. They claim it's to keep the temporal resolution, but my problem is that I think using a stride of 1 is equivalent to just doing nothing and keeping the data as is.

As an example, say we have a matrix in which every time step corresponds to one column:

A= [1,2,3,4;

5,6,7,8;

1,2,3,4];

If we max pool over time with stride 2, we'll have:

B = [2,4;

6,8;

2,4]

Max-pooling with stride one will keep the time resolution but also result in B=A (keep every column). So what's the point of even saying that max-pooling was applied?

I hope my question was clear enough, thank you for reading.

hello

if i have a signal for example

x=[1 0 1 0 0 1]

and i want to convolve it with a filter that have for example three points and i want the result to be the signal without the leakage how can i perform convolution in MATLAB and take the correct signal points

I have a model, where in some part of which I have used 3 convolutional layers each having 3*3 kernel size parallelly like this

(in_) # My InstantiateModel

x1 = Conv2D(32, 3, padding="same")(in_)

x2 = Conv2D(32, 3, padding="same")(in_)

x3 = Conv2D(32, 3, padding="same")(in_)

So basically a 3*3 kernel is a 3*3 matrix, so I want to know what are the values of those kernels/matrix. In a convolutional layer we are directly specifying the size of kernel like above, where 3 is the kernel size , but what are the values of those kernel ?

For my case where I am using 3 layers with same kernel size parallelly, so are the values of each kernel same for each layer or different. Please provide me some information on this thing.

I am currently working on video files and my task is to classify whether a given video has raciness (not safe for work) or nudity. So far my approach to solve this problem was using video classification models or 3D convolutions. However the results were not promising.

Would be great if anyone has an experience on this task. Any suggestions, papers, articles are all welcome and respected.

I am currently working on an AR project and our team used the pre-trained SuperPoint model to detect key points and match different pictures. However, its behavior on matching the pictures with rotation of degree 90, 180, and 270 is not that good on our self-created testing sets. I am looking for ways to improve the SuperPoint model's ability in matching pictures that has been rotated 90 degrees, 180 degrees, and 270 degrees.

I found on the internet and Skydes suggested that "If you don't mind about compute and storage, a quick fix is to rotate the image by multiples of 90°, extract features for the 4 resulting images, match each of them against your other images, and keep the set of matches with the largest number of RANSAC inliers." However, we do care about compute and storage, so we are seeking ways to maybe improve the current structure of the SuperPoint model or any other ways that are friendly in compute and storage resource.

Could someone give some suggestion for me? Thank you very much!

Hello,

In neural network pruning, we first train the network. Then, we identify redundant parts and remove them. Usually, network pruning deteriorates the model's performance, so we need to fine-tune our network. If we determine redundant parts correctly, the performance will not deteriorate.

What happens if we consider a smaller network from scratch (instead of removing redundant parts of a large network)?

Hello dear researchers

Which deep network along with a convolution neural network(CNN & LSTM or CNN& GAN or CNN& CNN or etc), do you recommend for a binary classification problem? Please introduce the appropriate reference because I really need it?

Thank you for your support

Hello,

I've pruned my CNN layer-by-layer in two steps.

First, I removed a fourth of the filters of some desired layers, which led to performance degradation of around 1%. Then, I selected two layers that had the highest number of filters and caused the least performance deterioration. Next, I removed half of their filters. The second model performed even better than the original model. What is the reason?

HOW TO use Convolutional Neural Networks on google earth engine

Can you help me find a nice solution to plot different CNN architectures automatically?

At the moment, I have a 3 head 1D-CNN, with 2 convolutional layers, 2 max-pooling layers, and 2 fully connected layers.

I used 3 heads to have different resolutions (kernel size) on the same signals.

Hello, does anyone know how to use XRD data and CMWP (Convolutional Multiple Whole Profile) to calculate the mean square microstrain ?

I am quite confused with the setting of Instrumental Profile. The second column sholuld be intensity. However, the example provided on the website are all small numbers, and some even are negative.

Besides, what do the background spline's base points file mean?

Thank you for the answers.

Is there any post-hoc easily applicable explainable AI tool for detections carried out by complex CNN-base architectures e.g. Mask R-CNN?

I was reading chapter from Fundamental Concepts of Convolutional Neural Network and could not get through explanation about how weight sharing between layers works in Convolution Neural Networks?

What is the complexity to run Deep Convolutional Neural Network (DCNN) for 500 epochs?

Who has used one-dimensional convolutional neural network (CNN)?

Is a small change in the accuracy of the results using the convolutional neural network algorithm important?

Transformer models, originally developed for natural language processing, now seem to be growing in popularity for computer vision (e.g., Vision Transformers and variants). I am curious whether researchers/engineers in the deep learning and computer vision communities believe that Vision Transformers will eventually dominate over convolutional models, or not? I am referring to actual deployments, in particular in mobile, embedded and low-power applications.

Recently, MLP-based models such as MLP-Mixer, ResMLP, and ConvMixer have suddenly appeared as a potential alternative to Transformers. MLP-based models are much simpler than Transformers and seem to become competitive in accuracy. A similar question as above arises: do you believe MLP-based models for vision will overtake Vision Transformers and convolutional models, in actual deployments? What's your opinion, and why?

I want to develop an ensemble approach where the final layer of a CNN model(Flatten layer in this case) will be followed by a K-Means Clustering algorithm where I want to cluster inputs into a number of categories same as required number of categories in a task. I want help regarding how to apply K-Means Clustering with a CNN.

I have 400 input features and 1 output value for every record and 1000 records. I want to apply CNN, please anyone tell how I fit the convolution layer? Should I use 1D or 2D convolution?

SVM can be added as the last layer of the CNN model. How to improve the performance of CNN-based SVM?

I'm trying to make a multichannel neural network that has 2 input models that pipeline into a single model (see image: https://i.stack.imgur.com/b6V7x.png ).

I need the first (top/left) channel to take in one tensor, and the second channel to take in three tensors. Of course, in doing so, I'm running into the issue of ambiguous data cardinality, because I'm comparing the output to the y_train set which is only 1 tensor.

Here's the error I'm getting:

ValueError: Data cardinality is ambiguous: x sizes: 1, 3 y sizes: 2

What's the best way to make this work?

Here's essentially what I have at the moment for fitting the data to the model:

model_history = trmodel.fit((np.array([model_images[0]], dtype=np.float32), np.array([model_images[1], model_images[2], model_images[3]], dtype=np.float32)), np.array(labels_seconds, dtype=np.float32), batch_size=32, epochs=2000, validation_data=(labels_seconds,), callbacks=[checkpoint])

It's been some time since I've worked with Keras, and I've never needed a multichannel network until now, so my apologies for my rustiness in the process. I can post the full code if that would help, also.

I am working on EEG prediction classification task. I apply segmentation sliding window techniques and I use CNN for prediction and classification task, most of the researches in this task used CNN but I don't know why CNN is better than RNN ?

How to classify large binary image dataset using Neural Network or Convolutional Neural Network in MATLAB?

What configuration need to be done (layers configuration)?

or

Setting / Preprocessing of data

how to configure:

[ _,_, 3] to [ _, _,1] image from RGB to Binary only in MATLAB tool

I'm interested to know what mathematical theory is behind using SpaCy / Python to calculate semantic similarity.

I had seen that SpaCy mentions a Convolutional Neural Network and a backpropagation process, it's probably combined with the Cosine Distance to get the distance value, but I want to know more details.

Does anyone know and can explain a little bit how it works?

Thanks!

Hello;

I know how to train image classifiers (CNN) for classification of single cross-sectional CT or MRI image. However, I don't know how to send all images from one patient to the model. I need to stack all images from a patients, add the label, and send them to the model to be used for training. It should be a sort of 3D CNN.

Anybody knows how to do it?

Thank you very much.

I’m defining a number system where numbers form a polynomial ring with cyclic convolution as multiplication.

My DSP math is a bit rusty so I’m asking when does inverse circular convolution exist? You can easily calculate it using FFT but I’m uncertain when does it exist? I would like to have a full number system where each number only has a single well defined inverse. Another part of my problem is derivation. Let c be number in my number system C[X] where coefficients are complex numbers. Linear functions can be derivated easily but I’m struggling to minimize mean squared error (i = 0..degree(C[X]), s_i(c) selects i:th coefficient of the number, s_i(x): C[x]->C):

error(W) = Exy{sum(i)(0.5|s_i(Wx - y)|^2)}

I can solve problem in case of complex numbers W E C but not in case of W E C[X] where multiplication is circular convolution. In practice my linear neural network code diverges to infinity when I try to minimize squared error.

Pointing any online materials that would help would be great.

How to create a deep (with more than 10 or 20 layers) convolutional neural network in MATLAB?

how many layers should we include in custom neural network to make it as deep network?

-Convolutional Layer

-Pooling Layer

-Fully Connected Layer

-Dropout

I am working on a project in which I need to advance my skills in neural network programming using Python. I am currently attending Andrew Ng's courses on Neural Networks. (I reached the fourth course on Convolutional Neural Networks).

After completing this course, I plan to enhance my skills as a Pythonista. I am wondering, shall I start learning Tensorflow or PyTorch?

Are Convolutional Neural Networks not Deep? If "yes" then there is no point in proposing Deep Convolutional Neural Networks!!

Isn't it?

If "no" then how CNN is defined as a deep learning method?

I require to differentiate these two models.

I would like some help in using VLAD Encoding Convolutional Features

Since I have 63,010 training images and 6761 testing images, I extracted the characteristics of 1,024 dim and the clustering number of 374

What are the steps required to get Encoded Global Image Features use VLAD encoding and Fisher encoding.

Thank you very much

Suppose CNN has 2 convolutional layers.

Case 1. CNN training 10 times，epoch=10.Get the accuracy, i.e. Acc_1=90%

Case 2.

Question 1 ：How many times does the neural network train, in the Case 2 ?

Question 2 ：Can we say that the case 1 has better results than the case 2 in a comparison experiment?

Hello,

Is anyone already worked with MR image data set??? If so, Is there any model to remove the motion artifacts in the MR image data set if contains??? What should we do if we have an MR image with motion artifacts??? Please give me your suggestions if it is possible to remove artifacts once the scan is produced.

Thanks in advance,

Dhanunjaya, Mitta

i know the data is available under this website

but i am unable yo understand how to use in the coding?

Dear, I have an issue with MobileNet v1

From the architecture table of the first MobileNet paper, a depthwise convolution with stride 2 and an input of 7x7x1024 is followed by a pointwise convolution with the same input dimensions, 7x7x1024.

Shouldn't the pointwise layer's input be 4x4x1024 if the depthwise Conv layer was stride 2? (Assuming is the padding of 1)

Is this an error on the author's side? Or are there something that I've missed between these layers? I've checked implementations of MobileNet V1 and it seems that everyone just treated this depthwise layer's stride as 1.