Questions tagged [convolutional-neural-networks]

For questions about convolutional neural networks, also known as CNN or ConvNet.

Filter by
Sorted by
Tagged with
1
vote
1answer
34 views

Given an input of shape $(3, 32, 32)$, which is convolved with a $(3 \times 3)$ kernel, how do I calculate the FLOPS?

I have an input tensor of shape $\mathbf{(3, 32, 32)}$ consisting of 3 channels, 16 rows, and 16 columns. I want to convolve the input tensor using $\mathbf{(3 \times 3)}$ kernel/filter. How can I ...
1
vote
0answers
22 views

How to pass variable length data as feature to a neural network?

I am working on building a model to classify the type of touch the user makes(Long Press, Left Swipe, Right swipe and so on). I have data with features that characterise the user's touch, like ...
0
votes
0answers
15 views

What is the significance behind having small kernel sizes over having one large kernel size that covers the entire input in a CNN?

I have hardly ever seen anyone cover the entire input image with a filter of the same dimensions. I was wondering why that is the case, and if the performance in say, an image detection application ...
0
votes
1answer
15 views

Can we change bias and control the output of neural network?

I have read the use of Targeted Adversarial Attacks for making the model perform better. But can we change the bias of the neural networks and control the outcome of the network rather than changing ...
0
votes
2answers
55 views

Why do we lose detail of an image as we go deeper into a ConvNet?

I was reading this research paper titled 'Image Style Transfer using Convolutional Neural Networks' which as the title suggests was based on Neural Style Transfer. I came across this line which didn't ...
1
vote
0answers
16 views

Non-determinism with mixed precision?

Currently, we're trying to improve failure analysis capability when using neural nets. One thing we want to resolve is output variation between batched runs and non-batched runs. For example, we wrote ...
0
votes
0answers
20 views

How to improve detection of wide objects?

I am working on a project where part of it is to detect PV module arrays, I trained few object detection models through TensorFlow Object Detection API and the problem I got is that the trained models ...
1
vote
1answer
258 views

Are the reports on Stanford website are credible(authentic) enough to study? [closed]

I found a bunch of reports of Stanford students available in their website. The following is the link http://cs231n.stanford.edu/reports/ I am aware that materials from Stanford are credible. But, I ...
0
votes
0answers
9 views

Why do we transform feature vectors in attention modules for CNNs

If we have a set of feature maps with dimensions [B, C, H, W] (batch, channel, height, width), why do we transform our feature maps before we calculate their affinity/correlation in attention ...
0
votes
1answer
41 views

Attention mechanism: Why apply multiple different transformations to obtain query, key, value

I have two questions about the structure of attention modules: Since I work with imagery I will be talking about using convolutions on feature maps in order to obtain attention maps. If we have a set ...
1
vote
1answer
57 views

Why do we add 1 in the formula to calculate the shape of the output of the convolution?

In the formula to calculate output shape of tensor after convolution operation $$ W_2 = (W_1-F+2P)/S + 1, $$ where: $W_2$ is the output shape of the tensor $W_1$ is the input shape $F$ is the filter ...
2
votes
0answers
26 views

Is there a systematic way of conducting deep learning experiments?

I have been working on a computer vision problem with the use of cnns, but quite frustratingly I'm often in the situation of not knowing what to do to improve my results. It seems to me that most of ...
0
votes
1answer
32 views

Is a convolutional layer capable of converting, for example, a binary image into an RGBA image?

I am asking this question for a better understanding of the concept of channels in images. I am aware that a convolutional layer generates feature maps from a given image. We can adjust the size of ...
0
votes
0answers
12 views

Deep NN architecture for predicting a matrix from a matrix and list of floats

I am trying to predict a matrix (size RxC) based on an input matrix (size RxC) and a list of floats ...
1
vote
1answer
20 views

Is reconciling shape discrepancies the only purpose of padding?

Padding is a technique used in some of the domains of artificial intelligence. Data is generally available in different shapes. But in order to pass the data as input to a model in deep learning, the ...
0
votes
0answers
13 views

Feeding the output back to input in 3D CNN model

I am currently designing a Model which takes Input 3D Grid and Model Output at $t-1$. The model figure is described below I have two thoughts in training the model for above situation. Feed output $...
0
votes
0answers
20 views

How Tesla and other companies use outputs from neural networks to drive the car?

Here is the short description of Tesla Autopilot AI: https://www.tesla.com/autopilotAI And here are some videos about how Tesla uses neural networks: Andrej Karpathy - AI for Full-Self Driving at ...
0
votes
1answer
47 views

Is it possible to use deep learning to generate a 2D image from a few numerical values?

Is it possible to train a DL model that will generate a full resolution 2D image based on few numbers describing this image and what type of model or architecture would that be? What I want to ...
0
votes
0answers
17 views

Scrabble-MuZero: combine observation planes of different shape

I'm working on an implementation of Scrabble with MuZero. The board state is represented by a matrix with shape $15 \times15 \times 27$ ($26$ letters $+ 1$ wildcard, value $0/1$) and the rack state $...
1
vote
0answers
30 views

Is there any animation that illustrates the "fold" and "unfold" operations of convolutional layers?

There are fourteen convolution layers in PyTorch. Among them six are related to convolution, another six are related to transposed convolution. The remaining two are fold and unfold operations. The ...
1
vote
0answers
40 views

Is there any gain by lazy initialization of weights, biases and number of input channels for a convolution operation?

The basic layers for performing convolution operation in PyTorch are ...
0
votes
0answers
17 views

Prune Neural Networks layers for f% sparsity - TensorFlow2

I am using TensorFlow 2.5 and Python3.8 where I have a simple TF2 CNN having one conv layer and an output layer for binary classification as follows: ...
0
votes
0answers
6 views

The ratio between number of units in multi-input model

I have the model that accepts two inputs: Image from camera Speed of the car I can create some CNN layers to process the image input and some MLP layers to process other type of data (for example ...
1
vote
0answers
33 views

How can I compute a mathematical formula for my CNN?

Let's say, for example, I have built the following CNN model using Keras: ...
1
vote
1answer
42 views

How is the convolution operation connected to neural networks?

I've been reading up on the convolution operation and neural networks. I understand that the convolution operation is defined as: $$(f * g)(t)=\int_{-\infty}^{\infty} f(\tau) g(t-\tau) d \tau$$ The ...
1
vote
0answers
27 views

Using a rectified Tanh to train a CNN?

I have been experimenting with activation functions on CNN, and it occurred to me to use a rectified tanh function. So that is basically if z > 0 tanh(z) else 0. ...
0
votes
0answers
12 views

Comparing results of different image splicing methods on a part of the CASIA 2.0 dataset

So I am working on an image splicing detection algorithm using ResNet-50 model. I am using the CASIA 2.0 dataset which consists of 7491 Authentic images and 5123 Fake images. However out of the fake ...
0
votes
0answers
45 views

Does randomly adding hand-engineered features increase the CNN's sample efficiency/performance?

It is a known fact that preprocessing images using CV techniques will improve CNN performance (see this answer). But what happens when you feed in the entire image and the filtered image randomly to ...
2
votes
1answer
45 views

What does "differentiable architecture" mean?

I'm currently reading a paper that uses CNN's as a base approach to solving some image classification issues and I've found that they kept mentioning the term "Differentiable Architecture", ...
0
votes
0answers
19 views

What image augmentations can help a neural network identify the smallest pixels within an image?

I am training a CNN to identify objects and I believe the network will learn much faster if it can learn to focus on the smallest pixels. One way to go about this would be to augment the images before ...
1
vote
0answers
50 views

What causes high differences in neural network accuracy each run?

I trained a CNN using Keras in R to multi-dimensional image data for image classification of five classes. I realized that each run (I retrained the network on the same data for ten times), although I ...
0
votes
0answers
76 views

Computational complexity of a CNN network

In the following network, the convolution operations of convolutional blocks are performed by three 1-D kernels with the sizes 8, 5, and 3 respectively along with stride equal to 1. The final network ...
2
votes
1answer
51 views

Are there any advantages of the local attention against convolutions?

Transformer architectures, based on the self-attention mechanism, have achieved outstanding performance in a variety of applications. The main advantage of this approach is that the given token can ...
3
votes
1answer
49 views

Can some of the weights be fixed during the training of a neural network?

Is it possible to exclude specific layers from the optimization? For example, let's say I have an input layer, 2 hidden layers, and the output layer. I know there is a perfect solution for my problem ...
0
votes
0answers
11 views

How to scale Computer Vision? How to implement Emotion detection from live video feed of N different video simultaneously?

I have a pipeline based on Scaled Yolov4 detection algorithm for faces which extract faces of users and then uses a CNN to ...
0
votes
0answers
7 views

Image classification distributed inference (mobile/server)

I'd like to learn some stuff about distributed DNN inference and how it works in practice. So, let's consider the example of image classification and assume we have a mobile device which utilizes the ...
1
vote
0answers
25 views

Convolutional Neural Network (CNN) with Tree architecture to organize the number of classes

At the moment, I have around 1.000 classes with accuracy and loss that are acceptable. In the long term, there could be more than 100.000 classes. The main problem is that every time a new class is ...
1
vote
0answers
20 views

What is a Silhouette Neural Network

I was going through a study in which I found something called a dilated Silhouette Neural Network. I want to know what it is, what it can do, and how it is better from a CNN? Link to the journal: Link
1
vote
0answers
9 views

Vector input to CNN for object detection

I am training a 3D object detection network (Retinanet-based as of the moment) for re-detecting tracked objects. I would like to be able to add the velocity vector of the tracked object as an input to ...
0
votes
2answers
35 views

Is "Pruning" only applicable to CNNs?

What Is Neural Network Pruning And Why Is It Important Today? The above article only talks about Convolutional Neural Networks: One of the first methods of pruning is pruning entire convolutional ...
0
votes
0answers
15 views

Looking for advice on how to train an AI object detection algorithm to recognize smaller objects than what it has been trained on

For some academic work, I am training an AI object detection algorithm (TensorFlow models) to look for specific objects (plants, in my case). I am taking photos with a hand-held camera, and am having ...
0
votes
0answers
23 views

Regress values inside the bounding boxes to predict a value in Object Detection

I am currently working on an object detection task. I have a dataset of Grayscale and Depth Images. The annotation format is x1, y1, x2, y2, class, depth. I have calculated this depth (of each object/...
1
vote
0answers
15 views

More channels vs multiple inputs in neural network

Suppose I want to train the model for playing chess. I found that existing models use as input the grid with dimensions 8x8x20 (so we have 20 channels). Some channels may represent how different kind ...
0
votes
0answers
12 views

Loss function to Push response value towards extremes

I have a feature map whose values are in the range of [0,1]. I want to push these values either towards extreme 0 or 1 using some loss function. Since I don't have any target value so it had to be in ...
0
votes
0answers
34 views

AI model to predict/generate person's image

I want to make a model that predicts person's shape depending on his son's image. My plan is to create a dataset and each data point in it consists of two images; One for the father or mother and one ...
2
votes
1answer
66 views

What gets optimized in convolutional neural network?

In a convolutional neural network, the hyperparameters such as number of kernels and stride, kernel size, etc are determined. After some combination of convolutions, ReLU and pooling layer there is ...
0
votes
0answers
26 views

How to change number of trained layers in object detection TensorFlow models?

Training custom object detection models with TensorFlow usually means a transfer learning of pre-trained models and, if I understand it correctly, it means only training the few last layers, with ...
0
votes
0answers
28 views

How can a CNN be used in machine translation?

How can a convolutional neural net (CNN) be used in machine translation? Convolution is a mathematical operation, so how are natural languages translated into matrices? e.g., DeepL_Translator#...
0
votes
0answers
14 views

In CNN, how the conversion of convolution layer to fully connected layer decides the no. of kernel

I am trying to understand the shape of the activation map after every operation. Here is the model summary . All is clear, but from the point labeled 1, how 7x7x512 turns out to be 4096 specifically ...
0
votes
0answers
17 views

How is the size of the class activation maps related to the size of the input images?

According to the images of the block diagram of class activation maps on the Internet, it seems that these images are a weighted sum of feature maps of the last convolutional layer as shown in the ...

1
2
3 4 5
21