Questions tagged [vae]
The vae tag has no usage guidance.
41
questions
0
votes
0
answers
7
views
Variational Autoencoder Multi-Class Interpolation
I'm working on a variational autoencoder (VAE) with 20 different classes in my training data. I've successfully trained the VAE and can sample from the latent space to generate data points. However, I ...
0
votes
0
answers
31
views
Similiar reconstruction for Pytorch VAE
This is my first question here, so if I don't offer enough information for my question to be answered, please let me know.
I am currently working on my Bachelor Thesis, in which I aim to integrate ...
0
votes
1
answer
15
views
How does a VQ-VAE produce new images?
I'm implementing a VQ-VAE for a LDM for biological time series data. I trained the VQ-VAE, and reconstructions works somewhat reasonable, but I have an understanding problem with how a VQ-VAE works.
...
0
votes
0
answers
20
views
Is it a good idea to use attention in VAEs for image generation?
There are research papers and codebases on GitHub that deal with VAEs for image generation on popular datasets like CelebA, etc. While surfing through Google Scholar I found self-attention and other ...
0
votes
1
answer
39
views
What neural network architecture would help me model a spectrogram?
I'm really a novice working with these technologies and I'm struggling to design a neural network that is powerful enough to model a spectrogram. For a personal project, I'm working on a spectrogram ...
0
votes
0
answers
8
views
estimate p(z|D) in a VAE
I have a maybe naive question, but is it easy to estimate p(z|D) for a variational autoencoder where $D = (x_1, \dots, x_n)$ is an iid training dataset ?
If we write the derivation, we get :
$
p(z|D) =...
0
votes
0
answers
13
views
Binary latent representation
I've been working on a problem where I got stuck at encoding my data into a binary latent representation, most of the methods out there aren't really working for my case.
I have input for the encoder ...
3
votes
2
answers
152
views
How does variational autoencoders actually work in comparison to GAN?
I want to know about how variational autoencoders work. I am currently working in a company and we want to incorporate variational autoencoders for creating synthetic data. I have questions regarding ...
0
votes
1
answer
128
views
What is the dataset during testing a Variational auto-encoder?
I am getting confused in the testing dataset of a VAE. After training the VAE, what should be the testing data-set of the VAE?
I understand that during testing the VAE only has the decoder part. Hence,...
1
vote
1
answer
67
views
Does minimizing kl divergence (i.e. keep approximate posterior close to prior) contradict the goal of avoiding posterior collapse?
Posterior collapse means the variational distribution collapse towards the prior: $\exists i: s.t. \forall x: q_{\phi}(z_i|x) \approx p(z_i)$. $z$ becomes independent of $x$. We would like to avoid it ...
0
votes
0
answers
15
views
General analytical form of KL loss
The form of KL loss that I am familiar with only requires you to specify mu and sigma, which means it doesn't work particularly well when the target (prior) distribution is more complex in terms of ...
0
votes
0
answers
47
views
Sigmoid Activation Function (Output layer) Alternative
I have a Convolutional-VAE architecture where the target images are in the range [0, 1], their pixel values. To synthesize/reconstruct images in this scale, I am using a sigmoid activation function in ...
0
votes
0
answers
24
views
KL divergence between two multivariate gaussians where $p$ is $N(\mu, I)$
We know if we try to get $D_{KL}(q||p)$, where $p$ is a standard normal distribution, so mean is 0, variance is the identity matrix, and $q$ is a multivariate normal distribution, it can be calculated ...
0
votes
0
answers
13
views
VAE mean and Standard Deviations are input dependent?
The original presentation of variational autoencoders, VAE assumes the mean $\mu$ and the sd $\sigma$ are functions of the input variable, say $x$. I am studying this paper and came across the lines
...
1
vote
0
answers
13
views
Why only discrete labels are used for (semi-)supervised VAEs?
I've noticed all semi-supervised VAEs assume discrete (categorical) labels to encourage disentangled representation learning in VAEs.
e.g.,
Kingma, Durk P., et al. "Semi-supervised learning with ...
0
votes
1
answer
809
views
KL divergence loss first decreases and then increases in VAE training
I am training a VAE on CelebA HQ (resized to 256x256). The training is going well, the reconstruction loss is decreasing and reconstructions are also meaningful. But, the problem is with KL divergence ...
0
votes
1
answer
90
views
Generate new distribution from auto-encoder /variational autoencoder
I know that autoencoders can be used to generate new data.
From what I could understand..
The autoencoder uses the original distribution X to learn a random gaussian distribution described by mean and ...
1
vote
0
answers
76
views
How should I think when I want to compare mu and sigma for different images in VAE?
I'm searching for a way to compare mu and sigma values of the encoder network's output of variational autoencoders.
In detail, imagine I trained my VAE on the MNIST digits dataset using the official ...
1
vote
0
answers
98
views
What does it means (concretly) that a VAE encode inputs as distribution?
From this post we can read that VAEs encode inputs as distributions instead of simple points ?
What does it mean concretely ? If the encoder consists of the weights between the input image and the ...
1
vote
1
answer
103
views
Latent variable graph in Variational Autoencoder
I followed this Keras documentation guide about Auto Encoders.
At the end of the documentation there is the graph of the latent variable z:
But I can not understand and how to interpret the plot, ...
0
votes
1
answer
720
views
Controlling the sampling from Variational AutoEncoder (VAE)
Suppose a Variational Autoencoder (VAE) is trained with mnist data. To sample, one draws from normal distribution. My question is: suppose I am interested in generating only 1s and no other digits. ...
0
votes
1
answer
189
views
Help needed in interpreting the loss, val_loss vs epoch plots for an autoencoder training?
I am training a variational autoencoder and I am getting a loss-plot as follows:
Rigt after epoch 224, val-loss overtakes train-loss and sort of getting bigger but at an extremely slow pace as you ...
1
vote
0
answers
10
views
Training with different datasets for the same better VAE model yields poor results
The VAE model I used here https://github.com/keras-team/keras-io/blob/master/examples/generative/vae.py.
It can produce very well results for the minist and fashion minist dataset.
But when I use my ...
4
votes
1
answer
3k
views
1D CNN Variational Autoencoder Conv1D Size
I am trying to create a 1D variational autoencoder to take in a 931x1 vector as input, but I have been having trouble with two things:
Getting the output size of 931, since maxpooling and upsampling ...
2
votes
1
answer
169
views
What are the Most Dissimilar MNIST Digits?
Using whatever definition of dissimilarity over sets that you'd like, what are the most dissimilar two digits in MNIST? I was thinking that a reasonable approach to answering the question would be to ...
1
vote
0
answers
182
views
pytorch code for VAE for MINST forces mu and logvar to zero
I am new to pytorch and trying to implement a VAE for MNIST data. When I try to train my model, it appears that the model forces mu and logvar to zero (or something very close to zero) independent of ...
1
vote
0
answers
139
views
Is vanishing KL a problem in vision based VAEs as well?
I came across some work on the problem of a vanishing KL contrbution in Variational Auto Encoders Cyclical Annealing Schedule: A Simple Approach to Mitigating KL Vanishing. This work particularly is ...
4
votes
1
answer
551
views
Why KL Divergence instead of Cross-entropy in VAE
I understand how KL divergence provides us with a measure of how one probability distribution is different from a second, reference probability distribution. But why are they particularly used (...
0
votes
1
answer
937
views
What makes the posterior intractable?
In the setting of Variational AutoEncoders, i.e. when we want to find the posterior distribution
over the data generating, latent variable z, given some ...
3
votes
1
answer
1k
views
VAE generates bad images. due to unbalanced loss functions?
I'm training a variational autoencoder on CelebA dataset using TensorFlow.keras
The problem I'm facing is that the generated images are not diverse enough and look kinda bad.
(new) Example:
What I ...
1
vote
1
answer
496
views
How to make custom callback in keras to generate sample image in VAE training?
I'm training a simple VAE model on 64*64 images and I would like to see the images generated after every epoch or every couple batches to see the progress.
when I train the model I wait until the ...
9
votes
1
answer
504
views
Train a GAN on "before and after" images of dental surgeries [closed]
I want a GAN to train on "before and after" images of dental surgeries; so that it can generate "after" pictures for fresh patients.
Input images are like these: https://img.webmd.com/dtmcms/live/...
2
votes
2
answers
66
views
Can VAEs be used to generate multivariate data?
Most of the tutorials online seem to use VAEs to generate images and use CNNs to generate data. I am working on a game with multivariate data consisting of character position and the character ...
1
vote
0
answers
241
views
Variational Autoencoder: Negative log likelihood not optimized
I am using the auto encoding variational Bayes algorithm for one unsupervised object detection task. In the loss function, the reconstruction loss is calculated as the log likelihood of the original ...
1
vote
0
answers
210
views
How to estimate total correlation KL[q(z)||Πjq(zj)] of VAE after training (useful for latents disentanglement evaluation)
FactorVAE and β-TCVAE both use total correlation (TC) batch estimation for their objectives. Where TC is:
$$
KL\bigl( q(z)||\prod\nolimits_{j} q(z_{j})\bigr)
$$
both estimates are applied to $q(z|x)$...
0
votes
2
answers
244
views
What is the meaning of "probability distribution of p(x)" of something uncountable?
I'm studying VAE and new to both of the neural network and the statistic.
After some researches, I could understand the rough concept of VAE.
But what makes me confused is, the meaning of probability ...
3
votes
1
answer
1k
views
Intractability in Variational Autoencoders
I'm having difficulty understanding when integrals are intractable in variational inference problems.
In a variational autoencoder with observation $x$ and latent variable $z$ we want to maximize data ...
6
votes
1
answer
5k
views
Why maximize ELBO in the variational autoencoder?
For a variational autoencoder, we have that:
$$\mathcal{L}(x,\theta,\phi) := \mathbb{E}_{z \sim q_\phi(z|x)}[\log p_{\theta}(x|z)] -KL[q_{\phi}(z|x) ||p(z)] $$
This is called the variational lower ...
5
votes
1
answer
7k
views
ValueError: Cannot convert a partially known TensorShape to a Tensor: (?, 256)
I'm working on a sequence to sequence approach using LSTM and a VAE with an attention mechanism.
...
3
votes
1
answer
6k
views
What do we visualize in showing a VAE latent space?
I am trying to wrap my head around VAE's and have trouble understanding what is being visualized when people make scatter plots of the latent space. I think I understand the bottleneck concept; we go ...
19
votes
1
answer
21k
views
What is "posterior collapse" phenomenon?
I was going through this paper on Towards Text Generation with Adversarially Learned
Neural Outlines and it states why the VAEs are hard to train for text generation due to this problem. The paper ...