How byronliang8/hubnessgansampling works
This repository is a Colab research artifact demonstrating a GAN latent-sampling method, rather than a general-purpose application. The documented workflow uses hubness priors to identify promising latents before image synthesis; the paper also frames truncation as an approximation to hub latents’ central-clustering effect. The tree contains only one notebook, so the diagram reflects documented behavior rather than inspected implementation details. The README identifies StyleGAN2 as the research basis but does not specify the notebook’s exact model-loading or synthesis calls.
Main entry points: Exploring_and_Exploiting_Hubness_Priors_for_High_Quality_GAN_Latent_Sampling.ipynb Colab Notebook
Components 8
Notebook Workflow
- Colab Notebook
Colab/Exploring_and_Exploiting_Hubness_Priors_for_High_Quality_GAN_Latent_Sampling.ipynb - Synthesized Images
Hubness Method
- Hubness Priors
Colab/Exploring_and_Exploiting_Hubness_Priors_for_High_Quality_GAN_Latent_Sampling.ipynb - Quality Latent Selection
Colab/Exploring_and_Exploiting_Hubness_Priors_for_High_Quality_GAN_Latent_Sampling.ipynb - Truncation Comparison
Colab/Exploring_and_Exploiting_Hubness_Priors_for_High_Quality_GAN_Latent_Sampling.ipynb
GAN Synthesis
- GAN Latent Distribution
- StyleGAN2 Synthesis
Other components
- Researcher
Connections 7
- Researcher → Colab Notebook: runs
- Colab Notebook → Hubness Priors: explores
- GAN Latent Distribution → Hubness Priors: provides latents
- Hubness Priors → Quality Latent Selection: guides selection
- Quality Latent Selection → StyleGAN2 Synthesis: selects latents
- StyleGAN2 Synthesis → Synthesized Images: synthesizes
- Colab Notebook → Truncation Comparison: compares