Skip to content

Multi Beta Vae Suite¤

Module: benchmarks.suites.multi_beta_vae_suite

Source: benchmarks/suites/multi_beta_vae_suite.py

Overview¤

Multi-β VAE controllable generation benchmark suite.

This module provides a comprehensive benchmark suite for evaluating multi-β VAE controllable generation models, targeting the Week 9-12 objectives:

  • MIG Score >0.3 (Mutual Information Gap for disentanglement)
  • FID Score <50 on CelebA (Fréchet Inception Distance)
  • Reconstruction Quality: LPIPS <0.2, SSIM >0.8
  • Training Efficiency: <8h per epoch on CelebA subset

Classes¤

MultiBetaVAEBenchmark¤

class MultiBetaVAEBenchmark

MultiBetaVAEBenchmarkSuite¤

class MultiBetaVAEBenchmarkSuite

Functions¤

init¤

def __init__()

init¤

def __init__()

run¤

def run()

run_all¤

def run_all()

Module Statistics¤

  • Classes: 2
  • Functions: 4
  • Imports: 10