Skip to main content

All Questions

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

How to resize the time-frequency spectrum of 1D signal so that image classification model can be used?

In case of 1D signal generally they tend to be around 1000s of sample points for each trial especially for biomedical signal. The time-frequency spectrum then can have shape like [256,1000] or [50,...
thinking_sapiens's user avatar
0 votes
0 answers
64 views

How to train a model to make predictions for larger sequences than those in the dataset?

I'm working on a project where I need my model to predict a sequence of n 3x3 matrices given an input sequence of n 3x3 matrices ...
Awwab Azam's user avatar
1 vote
1 answer
31 views

why CNN can still classify permuted images (originally multi-channel EEG data)?

I used multiple images to do classification. These images are multi-channel EEG data. Now I used the same random matrix to shuffle all elements on the rows and columns. I use CNN to do classification ...
Chris Lee's user avatar
0 votes
0 answers
53 views

How to overcome overfitting when you can't increase the training data due to constraint in time and resources?

I trained a complex cnn but when I look at the training and validation loss, it seems that the model is overfitting. But due to limitation on resources I can't increase the datasize. Is there any ...
ananya's user avatar
  • 21
1 vote
1 answer
42 views

Teaching a neural network when order of output does not matter

I am trying to train a neural network that predicts 2 3D points in space by looking at the hit counts from voxel data. I am currently doing this by making the network return 8 floats (x1,y1,z1,x2,y2,...
efemantarci's user avatar
1 vote
0 answers
17 views

Detect moving persons by analyzing pair of images

Consider a case where you want to detect moving persons. No animals exist at the place. The illumination at the place may vary a lot. You don't have to detect immobile humans. A camera produces a ...
Harry's user avatar
  • 11
2 votes
0 answers
18 views

How can I visualize pixel level information perceived by a model (VGG-16)?

As a fast.ai starter project I regressed a little model on movie stills from various eras to see if I could predict a year of release given an unseen image from a film. It works reasonably well! (feel ...
Regan 's user avatar
  • 121
2 votes
0 answers
29 views

How to analyze furniture for digital reconstruction?

My end-goal is to take a single photo of a piece of common furniture (couch, chair, table) and create a 3d model from that. I'm a novice with deep learning as I've only done basic CNN's and such with ...
Jacksonkr's user avatar
  • 119
0 votes
0 answers
19 views

My CNN validation Accuracy increases super slow?

im doing a retinopathy detection project with over 3500 images, 700 in each class. I've filtered the image like It seems that my model isn't learning from the data, or is having trouble because the ...
Rishhh's user avatar
  • 1
0 votes
0 answers
19 views

Efficient Net V2 M ONNX model infers significantly slower on small input

When I convert an Efficient net v2 m model from Pytorch to Onnx on differently sized inputs, I notice a strange and unexplained behavior. I was hoping to find an explanation to my observations from ...
Nitish Agarwal's user avatar
0 votes
2 answers
64 views

How do I improve my model accuracy and val_accuracy for my cnn model?

I'm using 3000+ retinopathy images in my CNN model. The accuracy remains around 77 to 80, how do i improve the accuracy value and reduce loss value? I've tried dropout and Adam optimizer to increase ...
Rishhh's user avatar
  • 1
0 votes
0 answers
28 views

How to Handle Masked Values in Neural Networks for Geospatial Data?

I am working on neural networks for oceanographic data and face challenges in dealing with masked values, which I set to NaN.. I can train a neural network model with 1D vertical profiles (e.g. ...
Giovanni's user avatar
0 votes
0 answers
46 views

Early divergence of YOLOv7-tiny train and val obj_loss plots

I am training a YOLOv7-tiny model and have the following observations from the training session: the train and val objectness loss plots diverged pretty early on in the training process the class and ...
fuse use's user avatar
0 votes
0 answers
35 views

What is the difference between 3 separate 5x5x1 convolutions, and one 5x5x3 convolution?

The actual numbers are just for the sake of clarifying my question, of course. What I mean is, since each channel in a multi-channel convolution has its own filter, what difference does it make if, ...
Idunnoanymore's user avatar
0 votes
0 answers
18 views

Monitoring validation loss with confident mispredictions

I am working on a binary classification task using a variant of ResNet. The dataset consists of medical recordings and is relatively small (N=2000), though I apply various validated data augmentation ...
Monotros's user avatar
  • 101
0 votes
0 answers
10 views

Train a decoder convolutional-based network with a non-differentiable classifier

Suppose you have a decoder network $D(\textbf{x})$ which outputs a latent representation $\textbf{z}$. My idea is to give such latent representation to a generic non-differentiable classifier $h(\...
King Powa's user avatar
  • 101
0 votes
2 answers
43 views

Adding one new class of images to the dataset and re-training the classifier model

I have a dataset of 10 types of images. I trained AlexNet to classify these images, and it gives decent results (around 80% accuracy). Now, I want to add the 11th class of images. Of course, I can ...
Yauhen Yakimenka's user avatar
0 votes
0 answers
26 views

Why isn't my CNN-RNN model learning despite the CNN performing well?

I'm working on a model that combines a CNN with an LSTM to process sequences of spectrograms and make per-time-step predictions. The CNN alone performs well on the task, but after adding an LSTM for ...
Leonardo Garofalo's user avatar
1 vote
1 answer
46 views

Where am I going wrong in my CNN approach to automate cropping images?

I have a dataset compiled of geological images. They often have unnecessary padding to the left, right, and bottom. I also have a folder containing cropped versions of these images where the padding ...
Vulcan's user avatar
  • 11
0 votes
1 answer
69 views

Fine-tune vs training from scratch

I'm training a 2 class Yolov8 Small detection model, and iterated through the model re-training over the previous best model a few times. Now, I added more data and the size of dataset is 2x (200,000 ...
Mary H's user avatar
  • 101
0 votes
1 answer
36 views

Why are video game upscaling models based on convolutional networks?

Why they use such networks, even through GANs and diffusion models can also be used for image upscaling? Examples of such models include Nvidia DLSS and AMD FidelityFX Super Resolution (FSR).
user avatar
1 vote
0 answers
9 views

Neural Networks that fit vector transforms

I have a CNN that is image to image and maps a binary image input to a binary image output. These are usually simple shapes, like a rectangle or a circle. Usually they become smoothed a bit (the ...
R S's user avatar
  • 11
1 vote
0 answers
27 views

Music simplifying modeling

I'm working on a machine learning task involving a unique dataset of piano tracks, represented as arrays with the shape (200, 10, 5000). Here, 200 is the number of tracks, 10 represents the number of ...
Michael's user avatar
  • 111
0 votes
0 answers
21 views

What is "Explicit Propagation" Image Inpainting - LatentPaint Paper - Generative AI

I am trying to implement "Explicit Propagation" method for Image Inpainting introduced in LatentPaint Research paper. This paper proposes the introduction of a module between VAE and ...
user3602374's user avatar
0 votes
1 answer
29 views

Deciphering terse CNN specifications

I'm reading a paper about three CNNs, described in a terse way that I cannot decipher (usually the case for space-efficient papers). Likely, this is due to my ignorance and I wonder if anyone can ...
Gaslight Deceive Subvert's user avatar
0 votes
0 answers
14 views

Why VGG fine tuning process keeps crashing in colab?

0 I am trying to fine-tune a VGG16 model for an ECG-image classification task with the resolution of 1700 * 2200 I haven't reshaped and reduced the size of the images because I thought some ...
rav2001's user avatar
  • 101
0 votes
0 answers
25 views

How to implement differentiable sinusoidal basis functions for a convolutional FFT module in torch

I currently have made a convolutional network module in torch that makes the basis functions in the usual way fourier_basis = np.fft.fft(np.eye(frame_size)) and I ...
lollercoaster's user avatar
0 votes
0 answers
15 views

Sinusoidal Kernel Initializer + 1D-CNN Implementation

I am trying to replicate the MSK+CNN model in this research paper: https://ieeexplore.ieee.org/document/9760210/authors#authors But I am not sure how to implement it correctly. It is suggesting that ...
L Z's user avatar
  • 11
1 vote
1 answer
42 views

How exactly is the dynamical unfolding implemented in ByteNet?

I am thinking about making use of ByteNet (https://arxiv.org/abs/1610.10099) architecture for a project, and would like to get a better understanding of how the model works. I've read through the ...
Philippa Richter's user avatar
0 votes
2 answers
164 views

How do multiple filters in a CNN work?

I understand how kernel size, stride, and the basics of CNNs. My question may look simple, but, despite my efforts, I haven't found the answer yet. In the following figure, let a $224\times224$ image ...
M a m a D's user avatar
  • 113
1 vote
1 answer
122 views

About the Matrix Multiplication in Fully Connected Neural Networks

I've learned some machine learning networks, and I think they're not that interesting. Sometimes I feel like they're just messing around with formulas and creating something that doesn't make sense. ...
shang's user avatar
  • 11
0 votes
0 answers
21 views

Federated CNN largely predicts only one (wrong) class

I have trained a CNN for the CIFAR dataset using federated averaging, adhering to this tutorial: https://www.tensorflow.org/federated/tutorials/federated_learning_for_image_classification. However, my ...
T.J.'s user avatar
  • 1
0 votes
0 answers
24 views

Why can't I replicate the validation loss from a Keras tuner (LSTM)

I feel like I'm doing a pretty straightforward sequence of tasks and must be making a simple mistake - I simply build a sequential model, tune it, build a clone of the optimal model by extracting the ...
Archetupon's user avatar
0 votes
0 answers
76 views

What's the simplest deep learning approach in signal processing? A 1D CNN?

I have a three datasets each containing one signal of a specific type (normal, periodically jammed, constantly jammed). I would like to experiment with this dataset to predict wether a signal is ...
AnneBlue's user avatar
0 votes
0 answers
24 views

What's the advantage of multi-GPU training in Alex-Net?

I was reading ImageNet Classification with Deep Convolutional Neural Networks(Alex et al) and they trained their model on two GPUs following fine-grained structure. Can you tell me why they chose that ...
Joseph Kasnoff's user avatar
0 votes
0 answers
13 views

Self-supervised learning weights initialization "after" projection head

For most Self-supervised learning algorithms: SimCLR, MoCo, BYOL, SimSiam, SwAV, etc., its common to have a projection head after the base encoder (which in most cases is a vanilla ResNet-50 CNN). An ...
Arun's user avatar
  • 235
0 votes
0 answers
15 views

What is the reason for the difference between the expected input tensor order for LSTM and Conv1d?

What is the reason for the difference between the expected input tensor order for LSTM and Conv1d? Say I have an input tensor for time series data of shape ...
Theta's user avatar
  • 1
0 votes
0 answers
17 views

Trying to understand backprop with convolution mathematically

I'm trying to understand how backpropagation works mathematically with convolutions. I have a VERY simple setup. X input to Y output through a linear filter: kernel = torch.ones((1, 1, 3, 3)) / 9 # ...
Shah's user avatar
  • 1
0 votes
0 answers
14 views

Why time based neuron still needed when the time-series data can be converted to time-freq domain (image) and use CNN for that?

LSTM, BI-LSTM, GRU, RNN are time-step based neuron. Why it still needed specifically for time-series data? I mean, we can just transform the time-series data into spectrogram and use CNN for that. For ...
Muhammad Ikhwan Perwira's user avatar
1 vote
1 answer
125 views

Kernels on a trained CNN seem random

I saw this question and am having the same issue: kernels on a trained CNN look random. I am using Pytorch to train a CNN, and based my code on examples that train the MNIST characters: ~50 ...
Tom Bensky's user avatar
2 votes
1 answer
64 views

Distinguishing between the fundamental structures of the convolutional neural network and the recurrent neural network: hierarchical vs sequential

I'm trying to distinguish between the fundamental structures of the convolutional neural network and the recurrent neural network. Convolutional neural networks build a hierarchical model from the ...
The Pointer's user avatar
0 votes
0 answers
24 views

Is it possible to have weightened (for every layer) gradient?

Imagine having pretrained network(resnet, ...) If we want to train for new classification task we change last layer and maybe freeze some first leayers. But what if we have another network that ...
Тима 's user avatar
0 votes
0 answers
12 views

Understanding matching of a CNN Layer's Output With the Receptive Field of Input Layer

I was trying to implement the following paper: https://arxiv.org/abs/1610.01563 and I came across something that seemed ambiguous to me. On page 4, second paragraph, it says After processing the ...
Redwanul Haque Sourave's user avatar
0 votes
0 answers
104 views

How to solve the exploding gradient problem in VAE training?

I was trying to implement VAE on the CelebA dataset inspired by the Tensorflow implementation of MNIST. I have tried varying batch size but there seems to be no effect from that. The image formed is ...
Vedant Bhardwaj's user avatar
0 votes
0 answers
17 views

How to get colored prediction output in multiclass UNET segmentation

i have the same data and annotation just like show in the provided link, https://stackoverflow.com/questions/60019869/u-net-how-to-improve-accuracy-of-multiclass-segmentation/78453876#78453876 query : ...
Muhammad Kamran Khan's user avatar
1 vote
1 answer
58 views

How to calculate the gradient for the output with respect to the input pixels

Hi for my project I'm using a somewhat simple CNN consisting of several convolution layers and pooling layers. Essentially the model is trained to perform a blur of sorts on an input image. For my ...
James Li's user avatar
0 votes
0 answers
30 views

CNN: Accuracy gap of 5-7 % between accuracy computed on-the-fly and separate model evaluation on the training set

I am training a CNN for some basic classification task. During training, I compute the training accuracy after every epoch. After the training has finished, I evaluate the model again on the entire ...
StrictlyStationaryPoster's user avatar
1 vote
0 answers
67 views

Visualization of Transposed Convolutions

After reading on Transposed Convolutions and Fully Convolutional Networks in the d2l book (14.10 and 14.11), I wondered about the visualization of transposed convolutions. As you probably know, ...
Mathy's user avatar
  • 153
0 votes
0 answers
57 views

I can’t pass a treshold no matter what I do

I am currently training an CNN for classification. My training data are 80x80 images, 3 channels, which I have grouped into 25% validation, 75% training, all evenly distributed. I have 3 classes into ...
will's user avatar
  • 1
0 votes
0 answers
27 views

Preparation of multivariate time series data

I am doing a university project on index/stock price prediction. I plan to use a combined cnn-lstm model, and I have several different types of data: Open High Low Close Volume, values, fundamental ...
Ivan's user avatar
  • 1

1
2 3 4 5
24