-
Download the images from CUB_200_2011
-
Download the preprocessed char-CNN-RNN text embeddings of BirdDataset
- [Optional] Follow the instructions reedscot/icml2016 to download the pretrained char-CNN-RNN text encoders and extract text embeddings.
Extract the Images and The Embeddings in the same directory. ('birds' and 'CUB_200_2011' directories would be made on Extraction.)
$ python3 main.py usage: main.py [--eval_] [--channels] [--epochs]
[--batchSize] [--saveInterval] [--embeddingsDim]
[--StageNum]
[--generatorLR] [--discriminatorLR][--beta1] [--beta2][--zDim ]
optional arguments:
--eval_ Boolean, whether to evaluate and save images or not
--channels Number of input channel of images
--epochs EPOCHS
--batch_size Batch Size
--saveInterval After How many epochs to save the Image
--embeddingsDim Embedding compressed dim
--StageNum Number Of Stages In StackGAN
--generatorLR Generator Learning Rate
--discriminatorLR Discriminator Learning Rate
--beta1 BETA1
--beta2 BETA2
--zDim Noise dimension
- Title: StackGAN++: Realistic Image Synthesis with Stacked Generative Adversarial Networks
- Authors: Han Zhang, Tao Xu, Hongsheng Li, Shaoting Zhang, Senior Member, IEEE, Xiaogang Wang, Member, IEEE, Xiaolei Huang, Member, IEEE, Dimitris N. Metaxas�, Fellow, IEEE
- Link: https://arxiv.org/pdf/1710.10916.pdf
- Year: 2017
Generative Adversarial Networks (GANs) were proposed by Goodfellow et al. In the original setting, GANs are composed of a generator and a discriminator that are trained with competing goals.The training process is usually unstable and sensitive to the choices of hyper-parameters.
Several papers argued that the instability is partially due to the disjoint supports of the data distribution and the implied model distribution. This problem is more severe when training GANs to generate high-resolution (e.g., 256 x 256) images because the chance is very low for the image distribution and the model distribution to share supports in a high-dimensional space. Moreover, a common failure phenomenon for GANs training is mode collapse, where many of the generated samples contain the same color or texture pattern. StackGAN++ expands upon the idea of StackGAN to include more Discriminator and by producing images at different levels of Generators which go through different Discriminators at once generating conditional and unconditional losses. The major idea behind the presence of multiple Generators is to enhance the data collection from the Embeddings by introducing Embeddings at each stage.
StackGAN-v2 consists of multiple generators (Gs) and discriminators (Ds) in a tree-like structure. Images from low-resolution to high-resolution are generated from different branches of the tree. At each branch, the generator captures the image distribution at that scale and the discriminator estimates the probability that a sample came from training images of that scale rather than the generator. The generators are jointly trained to approximate the multiple distributions, and the generators and discriminators are trained in an alternating fashion. In this section, we explore two types of multi-distributions: (1) multi-scale image distributions; and (2) joint conditional and unconditional image distributions.
