Questions tagged [gan]

GAN refers to Generative Adversarial Networks. Such networks is made of two networks that compete against each other. The first one generates new samples and the second one discriminates between generated samples and true samples.

0
votes
0answers
12 views

Using multi-channel input for a CNN, can one channel benefit from the other (in terms of resolution, texture etc.)?

I have one low resolution image (A) and a high resolution image (B) from another modality. A and B have statistical correlations at some arbitrary scale space. I would like to use the information from ...
1
vote
0answers
16 views

When is bias set to False?

I have been working with DC-GAN to generate pairs of images based on WGAN paper. For which I referred fastai notebook on WGAN where the bias in the network has been set to False. Jeremy Howard in his ...
0
votes
0answers
36 views

Map predictions to real text

I have read the paper "Learning to Read by Spelling" by Gutpa et al. They present a method for visual text recognition without using any paired supervisory data. In chapter 4 they describe how to ...
0
votes
1answer
18 views

Generative network understanding

I was going through GAN's notebook by fchallot on Generative Adversarial Networks where, in the Generator Network, he creates a Dense layer with $16*16 * 128$ (...
0
votes
0answers
12 views

Get label from CGAN or InfoGAN

So i'm trying to build a CGAN. I know you can have several labels and tell the GAN to generate an image from one. But what if i want to feed it an image and have it tell me if it's real or fake? Let'...
1
vote
0answers
15 views

CGAN - Shape must be rank 2 error

I'm trying to create a conditional GAN with the following code: learning_rate = 0.0002 batch_size = 128 epochs = 10000 ...
0
votes
1answer
27 views

Error on CGAN / InfoGan [closed]

So i'm trying to execute the code provided here: https://github.com/eriklindernoren/Keras-GAN/blob/master/infogan/infogan.py Both for InfoGAN and CGAN i get the same error: ...
2
votes
1answer
27 views

How to optimize the lambdas of a hybrid loss in a deep learning model

I am using a generative adversarial deep learning model (GAN) with a hybrid loss represented by a linear combination of four losses with three $\lambda$'s, something like: $total\_loss = loss_1 + \...
0
votes
0answers
30 views

Transposed convolution as umpsampling in DCGAN

I read several papers and articles where it is suggested that transposed convolution with 2 strides is better than upsampling then convolution. However implementing such model with the transposed ...
1
vote
2answers
92 views

GAN - am I seeing mode collapse? Common fixes not working

I have a 2 part question. Context I am learning about GANs and writing my own starting from the very simplest example of adversarial learning (1-parameter node), then implementing a very simple 1-...
0
votes
0answers
32 views

Label embedding in Auxiliary Classifier GANs

In Auxiliary Classifier GAN the generator takes two inputs, 1. one hot encoding of the labels, and 2. noise vector. But in the implementation of the GAN (e.g.:) some embedding is used, I think it is ...
1
vote
0answers
43 views

Where can I find out about the “Helvetica scenario”?

From the paper introducing GANs: It makes sense that collapsing too many $\vec{z}$-values to a single $\vec{x}$-value will cause problems. However, I was a bit confused as to how training $G$ for a ...
1
vote
0answers
31 views

Something is disastrously wrong with my neural network and what it's produced

I just got a neural network to run and although it doesn't raise any exceptions, I'm left with a horrible mess after 80 to 100 epochs: After 100 epochs with the adapted code: After 100 epochs with ...
0
votes
0answers
25 views

Understanding notation of Goodfellow's GAN objective function

What is the meaning of $V(D,G)$? How do we get these expectation parts? I was trying to understand it following this article: Understanding Generative Adversarial Networks (D.Seita), but, after many ...
1
vote
0answers
34 views

GAN optimizer settings in Keras

I am working on a Generative Adversarial Network, implementing in Keras. I have my generator model, G, and discriminator D, both are being created by two functions, and then the GAN model is created ...
0
votes
0answers
34 views

Dimension of feature vectors for classification task in the DCGAN paper

I am trying to implement one of the section in the DCGAN paper (https://arxiv.org/pdf/1511.06434.pdf) i.e. Using the Discriminator network trained on ImageNet-1k as a feature extractor to classify ...
2
votes
0answers
63 views

How to design the generator in generative adversarial network (GAN)?

I am new to GAN. And recently read a paper about how to implement GAN in recommender system here I have a question in the paper. The Eq (20) in the paper should ...
0
votes
1answer
23 views

In CycleGAN are there two different generators and two different discriminators?

I am trying to assimilate the contents of this paper. I have a confusion about how many different networks are there in the architecture of CycleGAN. To my understanding, the concept of cycle means ...
2
votes
0answers
48 views

Knowing when a GAN is overfitting (sequence classification study)

I have sequences of long, sparse 1_D vectors (3000 digits, made of of 0s and 1s) that I am trying to classify. I have previously implemented a simple CNN to classify them with relative success (with ...
0
votes
0answers
12 views

How to custom build a convolution generator network?

I learned GAN's by using mnist dataset(28x28) and codes available in the web. Now I am trying to build a GAN for dataset with images containing custom channel, rows, columns. eg:(3,300,200). I have ...
0
votes
0answers
40 views

Transpose-CNN with batch normalization

I am new to GAN (generative adversarial networks) and am trying to implement a Transpose-CNN (T-CNN) in Matlab (which is used as the generator in the GAN). I was able to build a Transpose-CNN with ...
0
votes
1answer
84 views

Which is better: GPT or RelGAN for text generation?

Based on my understanding, gpt or gpt-2 are using language model loss to train and generate text, which do not contains GAN. So which is better: GPT vs RelGAN/LeakGAN/SeqGAN/TextGAN I am so ...
1
vote
1answer
213 views

Why is my generator loss function increasing with iterations?

I'm trying to train a DC-GAN on CIFAR-10 Dataset. I'm using Binary Cross Entropy as my loss function for both discriminator and generator (appended with non-trainable discriminator). If I train using ...
3
votes
1answer
41 views

What is the difference between ImageNet and ImageNet1k? How to download it?

Some papers mention just ImageNet and some papers mention ImageNet 1k database? What is the difference between these 2? Are they same or is the latter one subset of the former one? I'm working on ...
6
votes
1answer
220 views

Why Gaussian latent variable (noise) for GAN?

When I was reading about GAN, the thing I don't understand is why people often choose the input to a GAN (z) to be samples from a Gaussian? - and then are there also potential problems associated with ...
1
vote
0answers
71 views

Keras regularizers (kernel, bias and activity) vs tf.contrib.layers.apply_regularization

I have a DCGAN set up in tensorflow that is working well on the faces in the wild dataset. As an experiment, I tried using the same architecture in keras to better understand the difference in ...
1
vote
2answers
159 views

Why i get OOM error although my model is not that large?

I am a newbie in GPU based training and Deep learning models. I am running cDCGAN (Conditonal DCGAN) in tensorflow on my 2 Nvidia GTX 1080 GPU's. My data-set consists of around 32,0000 images with ...
0
votes
0answers
14 views

How can I enrich train data in case of cnn using target and time features

I have a sequence of images, let's say we ignore time specificity for now. In the other hand, target is a multivariate continuous time series. Let's consider it just a univariate one. Training a cnn ...
0
votes
0answers
93 views

What is the difference between TextGAN and LM for text generation?

I'm new to LeakGAN or SeqGAN or TextGAN. I know GAN is to generate text and let discriminator un-judge-able to real text and gen-text. LM(language model) is the task of predicting the next word and ...
1
vote
1answer
42 views

What is meant by Average Content Distance in Videos generated by GANs?

I'm reading a research paper on generating/synthesizing videos: MoCoGAN: Decomposing Motion and Content for Video Generation To evaluate the generated videos, they have used a metric called 'Average ...
1
vote
0answers
26 views

Deeplearning without an objective function?

In this article, the author talks about how deeplearning models no longer are trained for an objective function that humans specify, but find their own objective function. Specifically, he is talking ...
1
vote
0answers
216 views

Tensorflow Deep learning network not utilizing GPU?

I have a Nvidia GeForce GT 755M (PC), which I heard should be at least functional for running deep learning models. But when I train my model (DCGAN) and check the task manager process info (Win 10) I ...
2
votes
1answer
115 views

Why do most GAN (Generative Adversarial Network) implementations have symmetric discriminator and generator architectures?

For example, if the discriminator is a vanilla network of n layers, each with n(i) units, then, typically, the generator will also be a vanilla network of n layers, each with n(n-i) units (except the ...
7
votes
1answer
139 views

Are there any actual, commercial uses of GANs already?

Doing research on the internet, I found many scientific papers, ideas, and experiments concerning GANs. But I was unable to find a single example of it being already used commercially. Q1 can you ...
0
votes
0answers
165 views

In generative adversarial models (GANs), why should we solve min-max problem and not max-min?

I know that in GANs model, there is min-max game between generator and discriminator which discriminator tries to maximize the loss function and the goal of generator is to minimize it. But why we ...
0
votes
1answer
157 views

Can a Conditional GAN be used for a regression task, where the condition can take any continuous value?

I came across conditional GAN, where the classes can be provided as input to the GAN. I would like to know, if the same network can be used for classes which can take any continuous value.ie, the ...
0
votes
0answers
37 views

Dataset for GANs (logo generation)

I'm begining work on a new project that involves GANs. So far what I've learnt from some publications (e.g. this) is that these models require literally tonnes of images, e.g. 80K. The problem I'm ...
2
votes
0answers
268 views

How to train the generator in a recurrent GAN (Keras)

I am trying to train a Recurrent GAN that is meant to generate geospatial movement data (sequences of 3-tuples of latitude, longitude and time). You may simply consider it a sequences of vectors with ...
1
vote
0answers
62 views

how to implement infoGAN's loss function in Keras' functional API

I have been trying to make an infoGAN based on this, but it is in pure tensorflow and I can't really figure out how I would implement the Q_loss in Keras (preferably the functional API). here is ...
0
votes
1answer
33 views

Evaluating performance of Generative Adverserial Network?

What is the best way to evaluate performance of Generative Adverserial Network (GAN)? Perhaps measuring the distance between two distributions or maybe something else?
1
vote
0answers
22 views

Confused about transpose convolution and tensor shapes in tensorflow GAN tuturial

https://github.com/tensorflow/tensorflow/blob/r1.11/tensorflow/contrib/eager/python/examples/generative_examples/dcgan.ipynb class Generator(tf.keras.Model): def init(self): super(Generator, ...
0
votes
2answers
151 views

Value Function of Generative Adversarial Network

My question concerns the notation used in the Value function of a GAN Does "x ~ p_data(x)" mean $$ E_{x\sim p_{data}}[log(D(x)]= \sum_{x} logD(x)(x)p_{data}(x)$$ ?? Many thanks,
2
votes
1answer
35 views

How to calculate $\phi_{i,j}$ in VGG19 network?

In the paper Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network by Christian Ledig et al., the distance between images (used in the loss function) is calculated from ...
2
votes
1answer
45 views

Can I use a GAN to increase my Dataset used for Image detection?

I am currently working on a machine learning project where I use the YOLO Algorithm to detect an object inside of a picture or video. The problem I face is that the specific image set (side-scan sonar)...
1
vote
1answer
35 views

EGAN Paper With Confusing Notation

I am reading a research paper that tries to fix some issues with GAN, but for one of the equations, the paper does not fully explain where it comes from and why it works. Although the overall ...
1
vote
0answers
33 views

Architecture Advice for training a GAN

I'm trying to create a model that generates worlds for a game. The game is 2-dimensional and small worlds sizes. I've sampled about 15,000 worlds that are about 200x200 blocks. Each block has its own ...
3
votes
0answers
23 views

Fully connected layer output explodes, but weights, gradients, and inputs all have sane values

I'm trying to train a GAN, and the architecture includes a fully connected layer before the output activation function. In my case, by the second training iteration this layer's output always explodes....
1
vote
0answers
756 views

Is it possible for a DCGAN to do regression? What are some examples of this?

I'm currently a student doing some machine learning projects, and I want to use generative adversarial networks to train some data to discern for example, how old someone is. The intended output is a ...
1
vote
0answers
273 views

Why use GAN in NLG?

I am interested in the GAN recently. There are many papers that recently applied GAN to NLG. I do not know much about NLP or NLG, but I wonder why I use GAN for NLG. For better quality? Or for many ...
2
votes
1answer
462 views

Gumbel Softmax vs Vanilla Softmax for GAN training

When training a GAN for text generation, i have seen many people feeding the gumbel-softmax from the generator output and feed into the discriminator. This is to bypass the problem of having to sample ...