carpedm20/DCGAN-tensorflow
DCGAN Tensorflow
This is a TensorFlow implementation of the Deep Convolutional Generative Adversarial Network (DCGAN) architecture, providing a framework for training generative models that produce synthetic images from random noise vectors. The project implements the core DCGAN design, using transposed convolutions for upsampling, batch normalization for training stability, and leaky ReLU activations in the discriminator, all executed as static TensorFlow computation graphs.
The implementation supports training on custom image datasets by accepting user-supplied image folders without requiring a predefined format, and includes tools for monitoring the adversarial training process. During training, the system logs generator and discriminator loss curves and records histograms of discriminator output distributions, enabling users to track model convergence and training dynamics. The trained generator can produce realistic face images from 100-dimensional random noise vectors, and supports feature interpolation by varying individual input dimensions to smoothly transition between facial attributes.
Features
- Generative Adversarial Architectures - Implements the DCGAN architecture using deep convolutional layers in both generator and discriminator networks.
- Batch Normalization - Applies batch normalization layers in both generator and discriminator to stabilize deep GAN training.
- Leaky - Applies leaky ReLU activations in the discriminator to allow small negative gradients and improve training stability.
- GAN Implementations - Provides a TensorFlow implementation of the DCGAN architecture for generating synthetic images from noise vectors.
- GAN - Generates realistic synthetic images by feeding random noise vectors through a trained DCGAN generator.
- GAN Noise-to-Image Generators - Transforms random noise vectors through a trained generator to produce synthetic images.
- GAN Noise-to-Image Sampling - Generates new images by feeding random noise vectors through a trained generator network.
- Generator and Discriminator Training - Implements the core adversarial training loop between generator and discriminator networks for DCGAN.
- Adversarial Training Procedures - Implements the minimax adversarial training loop that simultaneously optimizes generator and discriminator networks.
- Synthetic Face Generators - Creates realistic face images from a 100-dimensional random noise vector using a trained neural network.
- TensorFlow Framework Implementations - Provides a TensorFlow framework for training GANs on custom image datasets with training monitoring.
- TensorFlow Graph Execution - Defines neural network operations as static TensorFlow computation graphs for efficient GPU-accelerated training.
- Transposed Convolutions - Uses transposed convolution filters to upsample random noise into spatial feature maps for image generation.
- Real-Time Face Synthesis - Creates realistic face images from a 100-dimensional random noise vector using a trained neural network.
- Loss and Distribution Monitoring - Logs generator and discriminator loss curves and discriminator output distributions for convergence tracking.
- Model Checkpointing - Saves and restores trained model parameters using TensorFlow checkpoint files for inference or continued training.
- Custom Image Folder Training - Accepts user-supplied image folders as training data without requiring a predefined dataset format.
- Raw Image Folder Training - Accepts user-supplied image folders as training data without requiring a predefined dataset format.
- Training Progress Monitors - Logs discriminator and generator loss curves and output distributions during training to track model convergence.
- Latent Space Interpolations - Varies individual input vector dimensions to smoothly transition facial attributes like expression or gender.
- Computer Vision Libraries - TensorFlow implementation of deep convolutional GANs.
- Generative Models - Deep convolutional GANs for unsupervised representation learning.
Star history
How this analysis was created: This summary and feature list were written by an AI model that read the project's README and public documentation pages. Each feature links to the documentation it came from; stars, license and language come straight from the GitHub API. The model does not read the source code, and the analysis is refreshed when the project is re-analysed. Learn more on our About page.
