Efficient Lossy Image Compression with Foundation Diffusion Models
Our novel image compression codec leverages foundation latent diffusion models to synthesize lost details and produce highly realistic reconstructions at low bitrates.
Diffusion Models Exhibit Remarkable Consistency Across Initializations and Architectures
Diffusion models with different initializations or architectures can produce remarkably similar outputs when given the same noise inputs, a rare property in other generative models.
Dynamic Backtracking in GFlowNet: Enhancing Decision-Making with Reward-Guided Exploration
Dynamic Backtracking GFN (DB-GFN) enhances the adaptability of GFlowNet decision-making through a reward-based dynamic backtracking mechanism, enabling more efficient exploration of the sampling space and generating higher-quality samples.
Exponentially Fast Distillation of Pretrained Diffusion Models for One-Step Generation
The proposed Score identity Distillation (SiD) method can distill the generative capabilities of pretrained diffusion models into a single-step generator, achieving exponentially fast reduction in Fréchet inception distance (FID) during distillation and surpassing the FID performance of the original teacher diffusion models.
Enhancing Consistency and Diffusion Models through Optimal Linear Combination of Saved Checkpoints
Linearly combining saved checkpoints from the training process of consistency and diffusion models can significantly enhance their performance in terms of generation quality and inference speed, outperforming the final converged models.
Efficient Sampling from Variational Flow Models via Systematic Transformations
This work proposes a systematic training-free method to transform the posterior flow of any "linear" stochastic process into a straight constant-speed flow, enabling efficient sampling without the need for extensive training.
ManiFPT: Defining and Analyzing Fingerprints of Generative Models
Generative models leave unique fingerprints on generated samples, aiding in model attribution and distinguishing between different generative processes.
SwiftBrush: One-Step Text-to-Image Diffusion Model with Variational Score Distillation
SwiftBrush introduces an image-free distillation scheme for one-step text-to-image generation, achieving high-quality results without reliance on training image data.
YOSO: One-Step Text-To-Image Synthesis with Self-Cooperative Diffusion GANs
YOSO introduces a novel generative model for high-quality one-step image synthesis by combining the diffusion process with GANs.
UniHDA: A Unified and Versatile Framework for Multi-Modal Hybrid Domain Adaptation
UniHDAは、複数のモーダルを持つハイブリッドドメイン適応のための統一された多目的フレームワークです。
© 2024 by Linnk AI