Skip to content

kr1347/GAN-Model

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

3 Commits
 
 
 
 

Repository files navigation

DCGAN — Face Image Synthesis

Open In Colab

Deep Convolutional Generative Adversarial Network (DCGAN) trained on VGGFace2 to synthesize photorealistic human face images from random latent vectors.

Architecture

Component Design
Generator 5-layer transposed convolution, BatchNorm, ReLU, Tanh output
Discriminator 5-layer convolution, spectral normalization, LeakyReLU
Latent dim z = 100
Image size 64×64 RGB

Spectral normalization on the discriminator stabilizes training and prevents mode collapse (Miyato et al., 2018).

Dataset

VGGFace2 — 25,000 face images sampled from the full dataset, resized and center-cropped to 64×64.

Training

  • Optimizer: Adam (lr=1e-4, β₁=0.5, β₂=0.999) for both generator and discriminator
  • Loss: Binary Cross-Entropy
  • Epochs: 100 with checkpoint resumption
  • Hardware: Google Colab GPU

Results

Generated samples after 100 epochs show coherent facial structure with varied identity, lighting, and pose. Checkpoint resumption allows multi-session training without losing progress.

Usage

Open the notebook in Colab and run all cells. Checkpoints are saved to Google Drive every epoch.

References

  • Radford, A. et al. (2015). Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks. arXiv:1511.06434
  • Miyato, T. et al. (2018). Spectral Normalization for Generative Adversarial Networks. ICLR 2018
  • Cao, Q. et al. (2018). VGGFace2: A dataset for recognising faces across pose and age. FG 2018

About

DCGAN trained on VGGFace2 (25k images) for photorealistic face synthesis — PyTorch, spectral normalization, 100-epoch training with checkpoint resumption

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors