Questions tagged [gan]

Generative Adversarial Networks (GANs) are neural networks that are trained in an adversarial manner to generate data mimicking some distribution.

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

Constraining critic gradient norm in WGAN

From the WGAN-GP paper (emphasis mine): A differentiable function is 1-Lipschtiz if and only if it has gradients with norm at most 1 everywhere, so we consider directly constraining the gradient norm ...
user avatar
0 votes
0 answers
12 views

Can I adjust the Wasserstein GAN loss function for my particular data?

I am working on building Generative Adversarial Networks for the purpose of generating synthetic flight data. The GAN will be trained on actual time-series flight data in the form of a (n,m,9) array ...
user avatar
0 votes
0 answers
8 views

How to set a manual loss for the generator in a GAN?

I have a trained GAN model that generates images. I want to replace the discriminator with a human. Ideally, this person enters a score of how much he or she likes the images and that is the loss for ...
user avatar
  • 101
0 votes
0 answers
21 views

Techniques Used for DeepFake and Its Corresponding Research Field

I'm a beginner in image generative models, I'm trying to do some work similar to DeepFake, therefore I would like to find out first what techniques DeepFake use to generate the fake videos. Do they ...
user avatar
  • 1,653
2 votes
1 answer
85 views

Should I be using batchnorm and/or dropout in a VAE or GAN?

I am trying to design some generative NN models on datasets of RGB images and was debating on whether I should be using dropout and/or batch norm. Here are my thoughts (I may be completely wrong): ...
user avatar
1 vote
1 answer
50 views

Initialization of GAN discriminator

The question is pretty straightforward: how are GAN and WGAN discriminators typically initialized? I couldn't find much info on this. E.x. for GANs, I imagine you would theoretically want the ...
user avatar
  • 133
0 votes
0 answers
23 views

FID as a metric to evaluate the quality of synthetic datasets (Non GAN generated) for training models for a given classification task

I am working on a problem of generating synthetic data (algorithmically by blender, not using GANs) to aid the training of some CNN for a classification ask. Ideally, I want to generate an algorithm ...
user avatar
  • 135
2 votes
0 answers
76 views

Hyperparameters tuning on GANs

I have seen this post talking about how to tune hyperparameters on GANs. I'm actually wondering, more generally, how does one go about tuning hyperparameters on GANs. Obviously you cannot (I mean you ...
user avatar
1 vote
0 answers
42 views

The discriminator is classifying everything as fake. What does it mean?

I am using a conditional GAN with a relativistic loss function for both generator and discriminator (https://arxiv.org/abs/1807.00734). Before I added the relativistic part, the discriminator ...
user avatar
  • 11
0 votes
1 answer
31 views

Is there such a thing as intra-sample modal collapse in GANs?

Mode collapse is a known issue in generative adversarial networks (GANs) whereby the generator only learns a subset of the real data distribution. In those cases, it only outputs variations of a small ...
user avatar
1 vote
0 answers
41 views

Why does a GAN generate samples from a random prior?

I've been reading Goodfellow et. al.'s paper on GANs and also the conditional GAN one by Mirza et. al. While relatively straight forward, I'm not sure I understand why the prior for the generator is ...
user avatar
  • 649
0 votes
0 answers
17 views

Classifying images with categorical and numerical data in a GAN

I want to create a GAN model that accepts tabular data as well as a corresponding image. The data should be trained all together. For the final processing I want to be able to pass the tabular data ...
user avatar
0 votes
0 answers
29 views

How to evaluate quality of VAEs generated samples

I have a set of generated samples from a latent distribution (say 100 images) from a learned VAE. For GANs, the Inception score metric (which helps assess image quality and image diversity). Any idea ...
user avatar
  • 1
1 vote
0 answers
39 views

Synthetic data generation - GANs vs Simulator?

For synthetic data generation, does the GAN perform better than a simulator? If so, what are the limitations of the simulator? If we consider Conditional GANs, we could generate data based on the ...
user avatar
  • 11
0 votes
0 answers
11 views

Getting rid of the additive degree of freedom for discriminators of WGAN-GP's

Setting: Discriminators in WGAN-GP's are trained to minimise the following loss functional over functions D: Here I have been playing around with training a critic (simple convolutional network ...
user avatar
0 votes
0 answers
52 views

Why do GANs use binary cross entropy in practice instead of the Jensen-Shannon divergence?

I believe I understand the JSD formula which is derived from the loss of an optimal discriminator. However, I do not understand why the discriminator is usually implemented with a binary cross entropy ...
user avatar
1 vote
1 answer
31 views

Question: Optimal D notation in Generative Adversarial Network (GANs)

I am completely new to Computer Vision and how Deep Neural Networks work on images in general. In particular, I have questions on the Discriminator component of Adversarial Generative Network (GANs). ...
user avatar
1 vote
1 answer
91 views

Question about latent space for GANs

I am currently reading about GANs and I had a question about latent space. A site mentions: Latent space refers to an abstract multi-dimensional space containing feature values that we cannot ...
user avatar
  • 113
1 vote
0 answers
48 views

GANs training initially degrades pre-trained generator

I have an issue with the training of a GAN, which consists of a generator and two discriminators. The generator is used to generate waveforms. 1-The generator is independently pre-trained by ...
user avatar
  • 121
0 votes
0 answers
17 views

Symmetry when using the Kantorovich-Rubinstein duality

In the WGAN-paper, an equivalent formulation of the Wasserstein distance is used. From my understanding, the Wasserstein distance is symmetric, but the version in paragraph 3 doesn't seem like it is. ...
user avatar
2 votes
0 answers
24 views

Which part of an encoder/decoder generative network is improved by adding a discriminator loss term?

Lets say you're doing a superresolution image task with "deep learning" constructs. You encode to a latent representation using some parameterized model (like a neural network), then decode ...
user avatar
  • 141
0 votes
0 answers
33 views

What is a good approach to increase the depth of Nifti file format or Dicom file series?

I have a CT scan dataset of skull fracture consisting of multiple fractures and normal cases, the CT scans are in Dicom format. I want to do multi-class classification. But not every Dicom image ...
user avatar
0 votes
0 answers
64 views

I need help understanding the meaning of the loss values of a WGAN with Gradient Penalty

I am currently working on training a Auxiliary Classifier Wasserstein GAN with Gradient Penalty. I based my implementation off of https://keras.io/examples/generative/wgan_gp/ (to which I added the ...
user avatar
0 votes
0 answers
19 views

Result differences in Generative Adversarial Networks (GAN) across epochs

I'm using MNIST data. I'm confused on how to evaluate results across epochs. For instance, below are 100th and 500th epoch outputs. Should I be worried that what looks like 7 (row 1 column 2) at epoch ...
user avatar
  • 1
0 votes
1 answer
53 views

Generating the next heatmap in a sequence of labelled heatmaps without assuming continuity

I have a sequence of labelled heatmaps and I want to generate a new heatmap that is the 'best guess' at what the next heatmap in the sequence will look like, without assuming that the next heatmap ...
user avatar
1 vote
1 answer
30 views

The motivation of the value function of the generative adversarial network

The seminal paper on the generative adversarial network, proposes to $$\min_G\max_D V(D,G)$$ where the value function $$V(D,G):=\int p_{\text{data}}(x)\ln p(D=\text{data}|x)\,dx+\int p_{\text{gen}}(x)\...
user avatar
  • 895
0 votes
0 answers
21 views

From nn.MaxPool3d to "strided convolution layer"

I am working on GAN for medical images and the generator mode is Unet, but some bugs crash me. like Unet's task can be adversarial learning and semantic segmentation. but it seems some layers are ...
user avatar
0 votes
0 answers
10 views

Do Self-Attention GAN models belong to Autoregressive models group?

Do Self-Attention GAN (SAGAN) models belong to Autoregressive models group?
user avatar
  • 111
1 vote
0 answers
63 views

GAN discriminator loss randomly jumps up to over 1

I'm training my GAN, but it seems the losses (especially the discriminator loss) are quite erratic. It quickly converges to 0, but will randomly jump up to 1 or even 2-3 sometimes. I'm wondering what ...
user avatar
  • 11
1 vote
1 answer
26 views

How to approach making a GAN where row order does not matter

I am working on a project that has the aim of generating "recipes" which are the summation of "ingredients" (1-dimentional length N tensors, where each index is a value that ...
user avatar
-1 votes
1 answer
33 views

Can a single GAN learn multiple probability distributions

Assuming for example to generate a sentence, I have two distributions i.e one that represents a semantic and another a syntactic structure of a sentence. This distributions are different. Can I train ...
user avatar
0 votes
0 answers
53 views

how the lemma is applied on InfoGAN paper

I am reading the InfoGAN paper and I can understand how the lemma (attached below) is proved. The original paper uses the lemma in this way: first it finds out that (equation 4 in the paper) I can ...
user avatar
0 votes
0 answers
69 views

Using GAN for generating data to augment training data

I want to model an experiment data using a neural network but as the data set size is too small (25 samples), I decided to use a GAN to generate more data. The input is a 4×1 tensor representing 4 ...
user avatar
1 vote
1 answer
28 views

Affects of including generated data into "real" dataset

I was thinking about what the outcome of the following idea would be. Let's say that we have a Generative Adversarial Network (GAN) that has "successfully" (i.e., Discriminator is not able ...
user avatar
0 votes
0 answers
50 views

What is the mathematical notation in the attached figure means?

The full context can be found in this link: https://towardsdatascience.com/understanding-generative-adversarial-networks-gans-cd6e4651a29 Appreciate your kind help
user avatar
  • 1
0 votes
0 answers
122 views

Effect of batch normalization in GANs

I have built a deep convolutional GAN for generating artworks and explored the effect of including batch normalization (bn) in both the generator and discriminator. Including bn in generator helped to ...
user avatar
1 vote
0 answers
16 views

Systematic mode collapse and its avoidance in gans [closed]

I am looking for a gan architecture that does systematically mode collapse in one configuration (say A), and does not in a different configuration (say B). The difference between configurations A and ...
user avatar
  • 23
1 vote
0 answers
53 views

Training a Discriminator to guide Beam Search for a seq2seq model?

The idea is to train a discriminator during training of the seq2seq model to differentiate between 'fake' decoder outputs and 'real' decoder targets, while not propagating discriminator loss to the ...
user avatar
  • 934
1 vote
1 answer
87 views

Can GAN's be used to fill in gaps?

Is it possible to make a Generative Adversarial Network that, given a fragment of a valid piece of data, can identify the patterns involved and generate something close to the original piece of data? ...
user avatar
  • 13
1 vote
2 answers
205 views

How to use BinaryCrossEntropy (intutively) for Generator Network in DCGAN model?

TL;DR: Can someone tell me intuition working on BCE loss in the generator, specially for RGB as each pixel is having 3 values i.e. a list of values rather than just ...
user avatar
  • 234
1 vote
0 answers
48 views

How do GANs handle discrete outputs?

Let's consider some fictive task of generating binary images of size 200x200 (each pixel should be either 0 or 1). As far as I understand, the generator will output 200x200 values between 0 and 1 ...
user avatar
1 vote
0 answers
291 views

How is inpainting for self-supervised pre-training of convolutional neural networks usually done?

I read a nice blog post on self-supervised learning and computer vision, which suggests in-painting (amongst other ideas) as a possible self-supervised task for a neural network to "adapt" ...
user avatar
  • 23k
0 votes
1 answer
184 views

gradient penalty in GAN

In WGAN(https://arxiv.org/abs/1701.07875), we have both gradient clipping and gradient penalty. How both differ and why does standard GAN doesn't have any kind of gradient penalty or gradient clipping....
user avatar
2 votes
0 answers
53 views

Problem of jensen shannon

In GAN we want to minimize Jensen-Shannon distance and we use gradient descent. When can't we use this approach? What attribute might the training data and the distribution of the generating network ...
user avatar
1 vote
1 answer
50 views

Clarifications on Generative Adversarial Nets

I have just read the paper https://arxiv.org/pdf/1701.00160.pdf which is a tutorial on GAN. I have a few clarifications: Must the dimension of the output layer of Generator match the input layer of ...
user avatar
1 vote
0 answers
88 views

Why is reverse KL more suited for data generation

Here goes a first question! In a paper I'm reading in the context of GAN's (WGAN in particular) I came across the following quote when the authors discuss KL divergence: while maximum likelihood ...
user avatar
1 vote
1 answer
50 views

Understanding math notation in infoGAN paper

I'm reading this paper about mutual information in infoGAN infoGAN_paper_link and already have the code to run it. I pretty much found code for it which is fine and dandy except for the fact that I ...
user avatar
  • 247
1 vote
1 answer
341 views

Why GAN use adversarial MinMax formulation rather than Min formulation?

For generative adversarial neural network, originally Goodfellow used a MinMax formulation as $\text{Min}_D\text{Max}_G \mathbb{E}_{real}logD(x) dx+ \mathbb{E}_{fake}(1-D(G(z)))dz$. As long as the ...
user avatar
3 votes
1 answer
267 views

What is the intuition behind the expected value in orginal GAN papers objective function?

I know how Generative Adversarial Network(GAN) works but it became a daunting task to grasp the non mathematical intuition behind the expected value in the objective function $L_D = - \left[\; \mathbb{...
user avatar
  • 2,031
1 vote
0 answers
17 views

What does it mean to perform backprop "through the operations of an SDE solver"?

I am reading this cool paper about Neural SDEs as GANs. I've gotten through all of it and I understand fairly well. I've taken a couple classes on SDEs so I'm comfortable with the math. What I don't ...
user avatar

1
2 3 4 5