Questions tagged [generative-adversarial-networks]

For questions related to generative adversarial networks (GANs), introduced in the paper Generative Adversarial Nets (2014) by J. Goodfellow et al. A GAN is composed of a discriminative model (D) and a generative model (G). The discriminator D needs to distinguish between data generated by the generator G and data in the training set, while the generator G needs to generate data such that the discriminator D is not able to accomplish its task.

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

How accurate is this table from the original GAN paper summarizing difficulties and properties for deep generative models?

In the original GAN paper, they talk about how inference and training might be done in other deep generative models. In no particular order I was confused by: what is meant by "Learned ...
1
vote
0answers
12 views

What does it mean by strong or sufficient gradient for training in this context?

It has been mentioned in the research paper titled Generative Adversarial Nets that we generator need to maximize the function $\log D(G(z))$ instead of minimizing $\log(1 −D(G(z)))$ since the former ...
0
votes
1answer
26 views

Does average loss function in GAN training is just an approximation of value function and does not ensure convergence of generator and discriminator?

The value function on which convergence has been proved by the original paper of GAN is $$\min_G \max_DV(D, G) = \mathbb{E}_{x ∼ P_{data}}[\log D(x)] + \mathbb{E}_{z ∼ p_z}[log (1 - D(z))]$$ and the ...
0
votes
0answers
12 views

Why proof of convergence in GAN paper is not applicable practically?

This question is about generative adversarial networks and restricted to the research paper titled Generative Adversarial Nets. If I select a particular architecture of MLP as a generator and trained ...
0
votes
0answers
9 views

How to handle critical points during generator training?

Using an MLP as generator introduces multiple critical points in parameter space. You can read this excerpt from research paper titled Generative Adversarial Nets In practice, adversarial nets ...
1
vote
1answer
31 views

What are the iid random variables for a dataset in the GAN framework?

I am trying to understand why mean is used for expectation in training Generative Adversarial Networks. The answer tells that it is due to the law of large numbers which is based on the assumption ...
0
votes
0answers
27 views

Are there any Generative Adversarial Networks without Multi Layer Perceptrons?

Although the main stream research is on Generative Adversarial Networks(GANs) using Multi Layer Percepteons (MLPs). The original paper titled Generative Adversarial Nets clealry says, in abstract, ...
0
votes
0answers
19 views

Which probability distribution a Generative Adversarial Network (GAN) is capturing: dataset or ground truth?

Consider the following statement from the abstract of the paper titled Generative Adversarial Nets We propose a new framework for estimating generative models via an adversarial process, in which we ...
1
vote
2answers
38 views

What is Lipschitz constraint and why it is enforced on discriminator?

The following is the abstract for the research paper titled Improved Training of Wasserstein GANs Generative Adversarial Networks (GANs) are powerful generative models, but suffer from training ...
0
votes
0answers
20 views

Could an adaptive discriminator augmentation (ADA) be used for a discriminatory task?

Was wondering if I could use an adaptive discriminator augmentation (ADA) on a data set like MNIST (multi-class classification task). It seems that this is focused on generative modeling, so not sure ...
0
votes
0answers
30 views

GAN performs worse after 50 epochs than after 2

I am training GAN on SVHN dataset (house numbers in Google Street View images, dimensions: 3x32x32 - 3 color channels). The problem is that it performs worse after some training (e.g. after 50 epochs) ...
0
votes
0answers
14 views

What is the state of the art in melody generation?

Generative Adversarial Networks can generate realistic photos of people, such as thispersondoesnotexist.com. I wonder whether one can train an artificial intelligence on a batch of plain solo melodies ...
0
votes
0answers
45 views

Is it possible to mimic someone's handwriting given their sample notes?

I have a pet project, but I'm not very well versed with AI. I posted this question on datascience.stackexchange with no luck. I think this forum is more apt, so can anyone here help me start this in ...
0
votes
0answers
8 views

Using GAN to create multiple images as per multiple input layers

How to use GAN to create multiple images like https://opensea.io/collection/boredapeyachtclub here, each attribute is variable like fur, BG, Expression, Sunglasses, etc. Is there any GAN software to ...
0
votes
0answers
16 views

GAN - relation of input size and hidden layer size

I'm adapting a GAN described here used for generating binary output. It's trained on binarized MNIST data, with a size of 28x28 so 784 values. I want to adapt it to train on and generate 1D vectors ...
0
votes
0answers
21 views

Text to image GANs and failure

My knowledge of GANs is relatively basic at the moment but I seem to remember reading somewhere that GANs that generate images from a text prompt, when they fail to understand some of the text/words ...
0
votes
1answer
34 views

Is there a performace benefits using VAE-GAN instead of just GAN?

I have read that when using VAE-GANs, first what happens is the VAE's encoder encodes some image to another encoded image, which from GAN's point of view is considered a noise, and then the GAN part ...
0
votes
0answers
8 views

GAN unexpected behavior depeding on scaling and generator output activation function

I am training a GAN using data that underwent PCA. When I scale the data between -1 and 1, no matter when I use 'tanh' or 'sigmoid' at the last layer of the generator, the network is not stable. ...
1
vote
0answers
14 views

Hyperparameters for Reproducing the Results of IRGAN on MovieLens 1M

I am trying to reproduce results reported for IRGAN (information retrieval GAN) on the MovieLens 1M dataset. The results I want to reproduce and their sources are listed in the table below. Model ...
2
votes
1answer
65 views

In style transfer, why does the comparison between channels give a good sense of style?

I have been learning about Style Transfer recently. Style is defined as The correlation of activations between channels. I can't seem to understand why that would be true. Intuitively, style seems ...
4
votes
0answers
38 views

Are generative models actually used in practice for industrial drug design?

I just finished reading this paper MoFlow: An Invertible Flow Model for Generating Molecular Graphs. The paper, which is about generating molecular graphs with certain chemical properties improved the ...
0
votes
0answers
12 views

Things to consider while adding custom function to generator output in GAN

I am training a GAN model (DCGAN) to generate 128x128 images. Now, I wish to add a function which will take the generator output, perform some pre-defined operations on it, and return the modified ...
1
vote
0answers
34 views

Which approach best suits vector encodings?

I want to build a model that when given two vectors, outputs the probability of one vector being the encoded form of the other. I have 2 strategies for this: (Dataset available) I can directly feed ...
1
vote
1answer
32 views

Are there architectures to generate pictures from four labels? (VAEs, GANs)

I want to try something with image creation via NNs. I have come across Variational Autoencoders and Generative Adversarial Networks as possible solutions but have only found image creatinon with ...
0
votes
1answer
63 views

Decision boundary figure in Least square GAN paper

I currently reading Least Square GAN paper. But, I cannot interpret the one of the its figures. . Explanation of the figure goes like this: Figure 1: Illustration of different behaviors of two loss ...
2
votes
1answer
56 views

Is it possible to use an internal layer's outputs in a loss function?

For a network of the form: Input(10) Dense(200) Dense(100+10) Dense(20) Output() Those +10 outputs are what I want to add to ...
0
votes
0answers
7 views

Feasibility f using Word2Vec embeddings as input for GANs

If I have a dataset of artwork and a list of say, 10-12 words that describe each piece of artwork, how feasible would it be to convert those words into a word2vec embedding and use that (plus lets say ...
2
votes
0answers
38 views

How does the output distribution of a GAN change if the parameters are slightly purturbed?

Suppose $G_{\phi}:\mathcal{Z}\rightarrow \mathcal{X}$ is a generator (neural network, non-invertible) that can sample from some distribution $\pi$ on $\mathcal{X}$. That is, $G_{\phi}(z)\sim \pi$ when ...
0
votes
0answers
14 views

Do dataset sizes matter in a Style GAN?

When working with classifiers, a class imbalance is a huge issue for our models. If we have too many images of class 1 and too few images from ...
1
vote
0answers
38 views

SAGAN - is there a mistake in the original paper?

in the original paper the following scheme of the self-attention appears: https://arxiv.org/pdf/1805.08318.pdf In a later overview: https://arxiv.org/pdf/1906.01529.pdf this scheme appears: ...
1
vote
1answer
30 views

Why does this paper say that the Nash-equilibrium of GAN is given by a discriminator which is 0 everywhere on the data distribution?

I am facing difficulty in understanding the bolded portion of the following statement from this paper GANs are defined by a min-max two-player game between a discriminative network $D_\Psi(x)$ and ...
0
votes
1answer
41 views

What to do with a GAN that trained well but got worse over time?

I am training a WGAN-GP network based on the following paper, though I am using a different dataset. Now, for the first ~ 60-70 epochs, my network trained really well, which I could see in the loss ...
3
votes
1answer
58 views

Why do we sample vectors from a standard normal distribution for the generator?

I am new to GANs. I noticed that everybody generates a random vector (usually 100 dimensional) from a standard normal distribution $N(0, 1)$. My question is: why? Why don't they sample these vectors ...
4
votes
1answer
66 views

GAN Generator Output w/ Periodic Noise

I am training a Semi-Supervised GAN, using multivariate time-series with window of shape (180*80) with the generator and discriminator architecture below. My data is scaled using Robust Scaler, so I ...
1
vote
1answer
45 views

Classifying generated samples with Wasserstein-GAN as real or fake

I'm quite new to GANs and I am trying to use a Wasserstein GAN as an augmentation technique. I found this article https://www.sciencedirect.com/science/article/pii/S2095809918301127, and would like to ...
0
votes
0answers
14 views

Why do the inception score and the Fréchet inception distance use the inception network and not another network?

So I was researching about the evaluation of GANs and found these two metrics which seem to be the most popular. I understand that the main ideia is to apply the data to a pre-trained network in order ...
3
votes
0answers
39 views

Best Machine Learning Model for “Predicted” Image Generation

I am currently working on undergraduate research to determine hotspots for hand-surface contact. Ideally, I would like to give the model a depth image as input: Example of synthetic depth image and ...
1
vote
0answers
18 views

Why does the relativistic discriminator increase the probability that generated data are real and decrease the probability that real data are real?

I was reading the ESRGAN whitepaper, where I came across this line: Relativistic discriminator [2] is developed not only to increase the probability that generated data are real but also to ...
3
votes
0answers
46 views

Is the GAN architecture better suited for medical image denoising than the CNN?

I'm considering using GANs for medical image denoising, based on previous literature, like this and this. My input to the GAN would be a high-noise image and my ideal output would be a low-noise, high-...
0
votes
0answers
24 views

Where can I find a progressively trained GAN's pretrained low-resolution models?

StyleGAN is trained progressively, meaning that it starts as a small network trained to produce 4x4 images, then a layer is added which doubles the resolution to 8x8, then 16x16, etc. The final result ...
7
votes
0answers
77 views

Why is my GAN more unstable with bigger networks?

I am working with generative adversarial networks (GANs) and one of my aims at the moment is to reproduce samples in two dimensions that are distributed according to a circle (see animation). When ...
4
votes
1answer
194 views

Why don't those developing AI Deepfake detectors use two detectors at once so as to catch deepfakes in one or the other?

Why don't those developing AI Deepfake detectors use two differently trained detectors at once that way if the Deepfake was trained to fool one of the detectors the other would catch it and vice-versa?...
2
votes
1answer
84 views

Aren't scores in the Wasserstein GAN probabilities?

I am quite new to GAN and I am reading about WGAN vs DCGAN. Relating to the Wasserstein GAN (WGAN), I read here Instead of using a discriminator to classify or predict the probability of generated ...
1
vote
1answer
102 views

What dataset might Elon Musk's Dall-E have used?

Dall-E, it can generate many imaginative images from the description, even some peculiar images, how did they actually create this kind of dataset to train this AI , because there is not much of that ...
1
vote
1answer
84 views

In this implementation of pix2pix, why are the weights for the discriminator and generator losses set to 1 and 100 respectively?

I am working on a pix2pix GAN model that was inspired by the code in this Github repository. The original code is working and I have already customized most of the code for my needs. However, there is ...
7
votes
1answer
914 views

What are the fundamental differences between VAE and GAN for image generation?

Starting from my own understanding, and scoped to the purpose of image generation, I'm well aware of the major architectural differences: A GAN's generator samples from a relatively low dimensional ...
0
votes
0answers
10 views

Parametrizing non-analytical functions using generative models

My questions centers around what method is best to use parametrize a response function for an experiment. We are currently using ab initio simulation to model our experiment's response. Unfortunately, ...
0
votes
1answer
70 views

How is the latent vector transforming to a feature map in DCGAN (Generator structure)?

I'm working on the code trying to generate new images using DCGAN model. The structure of my code is from the PyTorch tutorial here. I'm a bit confused trying to find and understand how the latent ...
3
votes
2answers
147 views

Mathematical Analysis of the Loss function of GAN

I was pondering on loss function of GAN and following thing turned out \begin{aligned} L(D, G) & = \mathbb{E}_{x \sim p_{r}(x)} [\log D(x)] + \mathbb{E}_{x \sim p_g(x)} [\log(1 - D(x)] \\ & =...
3
votes
1answer
562 views

Can someone explain R1 regularization function in simple terms?

I'm trying to understand the R1 regularization function, both the abstract concept and every symbol in the formula. According to the article, the definition of R1 is: It penalizes the discriminator ...