The image generation component provides the mainstream generative adversarial network (GAN) model training feature used for image generation. You can use this component to train raw images to generate high-quality and diverse image generation models by using image generation networks, such as Deep Convolutional GAN (DCGAN), Wasserstein GAN with Gradient Penalty (WGAN-GP), Least Squares GAN (LSGAN), Graph GAN (GGAN), Progressive Growing of GAN (PGGAN), and Style Generative Adversarial Network (StyleGAN).
Supported computing resources
Inputs and outputs
Input ports
You can use the Read File Data component to read the Object Storage Service (OSS) path where training data is stored.
You can configure the Oss path to train data parameter of the image generation component to select the OSS path where image data is stored.
Output port
You can save trained models to the path specified by the oss path to output distributed parameter of the image generation component.
The following table describes the parameters of the component.
Configure the component
On the details page of a pipeline in Machine Learning Designer, add the image generation component to the pipeline and configure the parameters described in the following table.
Tab | Parameter | Required | Description | Default value | |
Fields Setting | oss path to train data | No | If no upstream OSS data is shipped, you must select the OSS path where the training data is stored. | None | |
oss path to pretrained model | No | The path of the pre-trained model. If you leave this parameter empty, no pre-trained model is loaded. The model must be in the same OSS bucket as the images that you train. | None | ||
oss path to output distributed | Yes | The OSS path that you can use to save a trained model. The model must be in the same OSS bucket as the images that you train. | None | ||
Parameters Setting | image generation network | Yes | The image generation network that you want to use. Valid values: DCGAN, WGAN-GP, LSGAN, GGAN, PGGAN, and StyleGAN. | dcgan | |
learning rate | Yes | The learning rate. | 0.001 | ||
number of train iterations | Yes | The total number of training iterations. One iteration indicates that the data specified by the batch size parameter is trained once. | 10000 | ||
visualization iterations | Yes | The output of the current model is saved to the training_samples folder in the path specified by the oss path to output distributed parameter after every specified number of iterations. The output is the image generated from random noise. | 1000 | ||
batch size | Yes | The number of training samples used in each iteration. | 32 | ||
model save interval | No | The model is saved after every specified number of training epochs. | 1 | ||
Tuning | Select Resource Group | Public Resource Group | No | The instance type and virtual private cloud (VPC) that you want to use. You must select a GPU instance type for the algorithm. | None |
Dedicated resource group | No | The number of CPU cores, memory, shared memory, and number of GPUs that you want to use. | None | ||
Maximum Running Duration (seconds) | No | The maximum period of time for which the component can run. If the specified period of time is exceeded, the job is terminated. | None | ||