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¤
MultiBetaVAEBenchmarkSuite¤
Functions¤
init¤
init¤
run¤
run_all¤
Module Statistics¤
- Classes: 2
- Functions: 4
- Imports: 10