How debopriya-ai/celebgan works
Project type and purpose
• This is a deep-learning prototype—specifically a custom Generative Adversarial Network (GAN) built with TensorFlow/Keras to generate photorealistic celebrity faces from the CelebA dataset.
• It’s research/experimental code rather than a full-stack or production microservice system.
File structure
• Top-level:
• LICENSE
• Lab10_DL__Debo (1).ipynb – the Jupyter notebook orchestrating experiments and visualizations
• README.md – project description
• According to the README, three key directories exist (even if not in the file tree snippet):
• dataset/ – raw & pre-processed CelebA images
• gan/ – model definitions (generator + discriminator), custom training loops, callbacks
• samples/ – snapshots of generated faces by epoch
README for additional insights
• Core features: DCGAN-style generator/discriminator, label smoothing, dropout, BatchNorm, custom GANMonitor callbacks.
• Tech stack: TensorFlow, Keras, Matplotlib.
• Workflow: ingest dataset → train GAN with stabilizing techniques → monitor & save samples.
Detail without over-splitting
• Essential pieces: data ingestion, two sub-models, training orchestration, monitoring outputs