Questions tagged [convolutional-neural-networks]
For questions about convolutional neural networks, also known as CNN or ConvNet.
464
questions
1
vote
0answers
23 views
When and How to use a mix of Loss Functions for back propagation?
Trying to understand the best loss function to be used with CNN, I came to know that we can mix two loss functions. Can any body share in what case was it done and how?
1
vote
2answers
44 views
What are the standard problems for CNNs and LSTMs?
What are the standard (or baseline) problems (or at least common ones) for CNNs and LSTMs? As an example, for a feed-forward neural net, a common problem is the XOR problem.
Is there a standard ...
0
votes
1answer
56 views
Neural Network to estimate distance
I built a three layer neural network (first is 1D Convolutional and the remaining two are Linear). It takes an input of 5 angles in radians, and outputs two numbers from 0 to 1, which are respectively ...
0
votes
0answers
15 views
Computing layers dimensions for deep learning architectures
I have already a few projects in deep learning under my belt. However, there is one fundamental thing that has come to my mind recently while trying to implement my own architecture.
Looking at the ...
3
votes
1answer
30 views
CNN how can i reduce gpu memory usage with large image sizes?
I am trying to train a cnn-lstm model, my sample image sizes are 640x640.
I have a GTX 1080 ti 11GB.
I am using Keras with tensorflow backend.
Here is the model.
...
2
votes
0answers
37 views
CNN clasification model loss stuck at same value
I have CNN model to classify 2 classes. (Yes or No)
I use categorical_crossentropy loss and softmax activation at the end.
For input I use image with all 3 channels, for output I use One hot encoded ...
0
votes
0answers
58 views
What to do when an image classifier does good for a class but bad for another?
So I wrote a convolutional neural network for a binary image classification. I have around 5300 images for each class which I thought would be enough to at least give me a good accuracy on the ...
3
votes
1answer
42 views
Can a model, retrained on images classified previously by itself, increase its accuracy?
Let's assume I have a CNN model trained to categorize some objects on the images. By using this model I find more categorized images. If I now retrain this model on data set that consists old set and ...
3
votes
1answer
32 views
GPU/TPU acceleration for neural networks with various network topologies
I was thinking about different neural network topologies for some applications. However, I am not sure how this would affect the efficiency of hardware acceleration using GPU/TPU/some other chip.
If, ...
0
votes
0answers
13 views
Numbers to image regression
I would like to create a machine learnig framework that could predict the 3D heat distribution of a room(of size 120x120x120) , given multiple parameters(position of the heater, orientation, power of ...
1
vote
0answers
24 views
How to get same accuracy with identical models in Keras and Tensorflow?
As we all know Keras backend uses Tensorflow and so it should give out same kind of results when we provide same parameters, hyper-parameters, weights and biases initialisation at each layer, but ...
4
votes
0answers
29 views
Get the position of an object, out of an image
I have some images with a fixed background and a single object on them which is placed, in each image, at a different position on that background. I want to find a way to extract, in an unsupervised ...
0
votes
0answers
7 views
Looking for the right type of 1D-Convolution that only considers one column/attribute
My input has the shape of n rows (time steps) and m columns (attributes). I want to train a convolutional neural network on it to predict a class.
I am currently using 1D-Convolutions. I got a good ...
0
votes
0answers
10 views
How much data is needed to train a deep learning model to detect instance masks?
I am trying to get an idea of how much data is needed to train a deep convolutional neural network to detect instance masks from images. I am interested in both papers that have been written on the ...
1
vote
0answers
21 views
Threshold selection for Siamese network hyper-parameter tuning
I'm interested in modeling a Siamese network for facial verification. I've already written a simple working model that inputs feature vectors generated from two CNNs with shared weights then outputs a ...
-1
votes
0answers
33 views
Which opinions are available to slow down the DexNet robotic grasping project?
The first version of a robotic arm who is able to grasp objects dexterously was announced three years ago. [1] It is working with convolutional neural networks who are trained on thousands of 3d ...
3
votes
1answer
22 views
Image classification with an associated matrix
I have a dataset of images with 9 different classes. However, there are different categories with the same type of associated image and only can be differentiated with an associated matrix in my ...
1
vote
1answer
30 views
How well can ConvNET distinguish an object from its class?
ConvNET can easily predict class of an object in an image. My question is, can ConvNET distinguish Pisa Tower from other buildings or Hagia Sophia from other mosquoes easily? If it can, how many ...
0
votes
0answers
14 views
Matterport MRCNN and multiclass classification
I want to create a model which solve a multiclass classification problem.
The main concept is:
every picture contain only one object
the background is very simple
all object is coming from the same ...
1
vote
0answers
16 views
Sliding Window Detection
Suppose that we have a labeled training set of $n$ closely cropped images of cars $(x_1, y_1) , \dots, (x_n, y_n)$. We then train a CNN on this. Let's say we have $m$ test images. Then for each of the ...
0
votes
1answer
44 views
Generate credit cards dataset for locating number region
Currently I'm working on a project for scanning credit card and text extraction from cards. So first of all I decided to preprocess my images with some filters like thresholding, dilation and some ...
1
vote
0answers
37 views
Which approaches are best suited for text deblurring?
I want to deblur text images using deep learning. Which approaches are best suited for the task? Any example networks? Is unsupervised network the best approach? GAN or cycle GAN for these purposes?
...
2
votes
0answers
19 views
Confidence Maps and Non-Linearity
I am currently trying to improve a CNN architecture that was proposed for generating depth images.
The architecture was originally proposed for autonomous driving and it looks like following :
The ...
1
vote
0answers
21 views
Reference request: one-hot encoding outperforming random orthogonal encoding
I experimented with a CNN operating on texts encoded as sequences of character vectors, where characters are encoded as one-hot vectors in one embedding and as random unit length pairwise orthogonal ...
3
votes
1answer
51 views
Is it useful to eliminate the less relevant filters from a trained CNN?
Imagine I have a tensorflow CNN model with good accuracy but maybe too many filters:
Is there a way to determine which filters have more impact in output? I think it should be possible. At least, if ...
0
votes
1answer
63 views
Should I use single or double view for gender recognition?
My project requires gender recognition of people shown on the given images, with more than one person per image. However, these people can be positioned in frontal or side view(passing by ...
1
vote
1answer
29 views
What does an oscillating validation error curve represent?
I have been training my CNN for a bit now and, while both the training loss and the training error curves are going down during training, both my validation loss and my validations error curves are ...
0
votes
0answers
12 views
Training, validation loss and accuracy yolov3?
This is a version of Yolo V3 implemented in PyTorch – YOLOv3 in PyTorch
I am trying to use transfer learning to train this yolov3 implementation following the directions given in this post. This is ...
1
vote
2answers
26 views
How does one create a non-classifying CNN in order to gain information from images?
How do I program a neural network such that, when an image is inputted, the output is a numerical value that is not the probability of the image being a certain class? In other words, a CNN that doesn'...
1
vote
1answer
14 views
What should load_mask() return if an image doesn't have any objects? (Mask RCNN)
I want to use Mask RCNN to do image segmentation. I need to override the load_mask function for the dataset class. I know this function should return mask tensors and class ids of objects in an image. ...
2
votes
1answer
39 views
Understanding the intuition behind Content Loss (Neural Style Transfer)
I'm trying to understand the intuition behind how the Content Loss is calculated in a Neural Style Transfer. I'm reading from an articles: https://medium.com/mlreview/making-ai-art-with-style-transfer-...
1
vote
1answer
43 views
What are the differences between Bytenet and Wavenet?
I recently read Bytenet and Wavenet and I was curious why the first model is not as popular as the second. From my understanding, Bytenet can be seen as a seq2seq model where the encoder and the ...
2
votes
1answer
24 views
Transfer learning to train only for a new class while not affecting the predictions of the other class
I am basically interested in vehicle on the road.
YoloV3 pytorch is giving a decent result.
So my interested Vehicles Car ...
0
votes
0answers
18 views
Loss reduction, but constant performance with CNN
I made a CNN with a reasonable loss curve, but the performance of the model does not improve. I have tried making the model larger, I am using three convolutional layers with batch norms.
Thanks for ...
1
vote
0answers
42 views
Torch CNN not training
I am completely new to CNN's, and I do not quite know how to design or use them efficiently. That being said, I am attempting to build a CNN that learns to play Pac-man with reinforcement learning. I ...
4
votes
1answer
29 views
Combining mean pooling and max pooling
Is it popular or effective to concatenate the results of mean-pooling and max-pooling? To get the invariance of the latter and the expressivity of the former.
2
votes
0answers
27 views
3D geometry and similarity with a reference model
I am looking for a CNN method, or any other machine learning method, to recognize 3D natural geometries that are similar to each others, and compare these geometries with a reference 3D model. To ...
1
vote
0answers
40 views
CNN - Visualizing images near decision boundary - Pixels inexplicably tend to edges
We are exploring the images classified by a CNN at its decision boundary, using Genetic Algorithms to generate them. We have created a fine-tuned binary grayscale image classifier for cats. As the ...
2
votes
0answers
20 views
What is the difference between Squeeze-and-excite and bottleneck modules from Mobilenet v2?
Squezee-and-excite networks introduced SE blocks, while MobileNet v2 introduced linear bottlenecks.
What is the effective difference between these two concepts?
Is it only implementation (depth-wise ...
4
votes
1answer
80 views
What is the difference between graph convolution in the spatial vs spectral domain?
I've been reading different papers regarding graph convolution and it seems that they come into two flavors: spatial and spectral. From what I can see the main difference between the two approaches is ...
2
votes
2answers
52 views
Why do we get a three-dimensional output after a convolutional layer?
In a convolutional neural network, when we apply the convolution on a $5 \times 5$ image with $3 \times 3$ kernel, with stride $1$, we should get only one $4 \times 4$ as output. In most of the CNN ...
6
votes
1answer
34 views
Is there any use of using 3D convolutions for traditional images (like cifar10, imagenet)?
I am curious if there is any advantage of using 3D convolutions on Images like Cifar10/100 or Imagenet. I know that they are not usually used on this data set, though they could because the channel ...
1
vote
0answers
49 views
How to train and update weights of filters
I have some problems with training CNN :(
For example:
Input 6x6x3, 1 core 3x3x3, output = 4x4x1 => pool: 2x2x1
By backpropagation I calculated deltas for output.
This tutor and other tutors are ...
1
vote
1answer
37 views
What is the difference between 2d vs 3d convolutions?
I was trying to understand the definition of 2d convolutions vs 3d convolutions. I saw the "simplest definition" according to Pytorch and it seems the following:
2d convolutions map $(N,C_{in},H,W) \...
1
vote
0answers
14 views
How do I recover the 3D structure of a layer after a fully-connected layer?
I want to implement a CNN, but I want to explore what happens when my first layer is a fully-connected one. I still want to use convolutions, of course, but I want to apply them after the first layer. ...
0
votes
1answer
28 views
Tweaking a CNN for large number of input channels
I am using a CNN for function approximation using geospatial data.
The input of the function I am trying to approximate consists of all the spatial distances between N location on a grid and all the ...
0
votes
0answers
20 views
Super Resolution CNN generates black dots on output images
I have been trying to train a CNN for the super-resolution task based on the work of Dong et al., 2015 [1]. The network structure built in PyTorch is as follows:
...
0
votes
1answer
56 views
What is the correct way to read and analyse images in machine learning?
I am trying to understand the best practice to read and analyze images. If your image has 10,000 pixels, your input layers will have 10,000 inputs?
It sounds that my neural network will have too many ...
1
vote
1answer
49 views
What do the numbers in this CNN architecture stand for?
So I've got a neural net model (ResNet-18) and made a diagram according to the literature (https://arxiv.org/abs/1512.03385).
I think I understand most of the format of the convolutional layers:
...
2
votes
2answers
49 views
How should we pad an image to be fed in a CNN?
As everyone experienced in deep learning might know, in an image classification problem we normally add borders to images then resize it to the input size of a CNN network. The reason of doing this is ...
