#Generative Modeling

Discover an innovative end-to-end TTS method that improves upon traditional two-stage systems using variational inference and adversarial learning. This approach enhances generative capabilities, resulting in natural-sounding speech. A stochastic duration predictor supports varied speech rhythms and tones from text. Human evaluations on the LJ Speech dataset demonstrate its superior performance, achieving MOS scores close to real human speech. Access the interactive demo for audio examples or explore available pretrained models.

GFlowNet is an innovative modeling framework optimized for discrete, combinatorial objects like graphs. The library centers on node-by-node graph construction via a graph neural network, which provides per-node and per-graph outputs. It supports multiple GFN algorithms and facilitates training with both existing and generated data. This resource offers detailed examples for practical application, such as molecule generation, and is compatible with Python 3.9+. It is available for installation through PIP, enabling flexible management of dependencies.

Discover Rectified Flow: a novel approach for transport map learning and efficient data generation. It uniquely enhances Stable Diffusion with improved diversity and FID scores. Applicable in generative modeling and domain transfer, explore its capabilities with practical Colab tutorials across datasets like CIFAR10 and CelebA-HQ.

This survey systematically categorizes key diffusion model research papers, as detailed in the paper accepted by ACM Computing Surveys. The repository highlights innovations in sampling acceleration, noise schedule optimization, and handling data with intricate structures. It covers extensive applications in fields like computer vision and natural language processing, focusing on advancements such as text-to-image generation and anomaly detection. Additionally, the survey explores the integration of diffusion models with other generative models and their impact on multi-modal learning.