This repository provides the code for "Glow: Generative Flow with Invertible 1x1 Convolutions," a generative model for image synthesis. It is intended for researchers and practitioners in deep learning and generative modeling who want to reproduce the paper's results or experiment with normalizing flows. The primary benefit is the implementation of invertible 1x1 convolutions for efficient and high-quality generative modeling.

How It Works

Glow utilizes normalizing flows, a class of generative models that learn an invertible transformation from a simple base distribution (e.g., a Gaussian) to a complex data distribution. The key innovation is the use of invertible 1x1 convolutions, which allow for efficient permutation of features within a layer without increasing computational complexity. This design enables the model to capture long-range dependencies and achieve state-of-the-art results on various image generation tasks.

Quick Start & Requirements

• Install: pip install -r requirements.txt
• Prerequisites: Tensorflow (v1.8.0 tested), Horovod (v0.13.8 tested), (Open)MPI.
• Datasets: MNIST/CIFAR-10 (downloaded by train.py), or large datasets from provided Azure URLs (e.g., ImageNet, LSUN, CelebA-HQ). Data preprocessing instructions are detailed in the README.
• Setup: Requires MPI setup for multi-GPU training.
• Links: Demo Folder

Highlighted Details

• Reproduces quantitative results on ImageNet (32x32, 64x64) and LSUN (96x96).
• Enables qualitative results on CelebA-HQ (256x256, 1024x1024) and LSUN.
• Supports conditional generation for CIFAR-10 and ImageNet.
• Provides commands for single-GPU and multi-GPU (MPI/Horovod) training.

Maintenance & Community

• Status: Archived. Code is provided as-is with no expected updates.
• Contributors: Primarily associated with OpenAI.

Licensing & Compatibility

• License: Not explicitly stated in the provided README snippet.
• Compatibility: Requires specific older versions of Tensorflow and Horovod, which may pose compatibility challenges with modern environments.

Limitations & Caveats

The project is archived, meaning no further development or support is expected. The reliance on older versions of Tensorflow (v1.8.0) and Horovod (v0.13.8) may present significant challenges for setup and integration into current deep learning workflows.