Efficient Differentially Private Synthetic Image Generation using Semantically-Aligned Pre-training
PRIVIMAGE, a novel method for efficiently generating differentially private synthetic images, leverages semantic-aware pre-training on a carefully selected subset of public data to achieve superior fidelity and utility compared to state-of-the-art approaches.
Boosting High-Resolution Image Synthesis with Coupling Flow Matching
Combining the diversity of diffusion models, the efficiency of flow matching, and the effectiveness of convolutional decoders enables state-of-the-art high-resolution image synthesis at minimal computational cost.
Quantization Noise Correction Scheme for Efficient Diffusion Model Inference
A novel post-training quantization scheme, QNCD, that effectively mitigates both intra and inter quantization noise in diffusion models, enabling efficient low-bit inference while preserving high-quality image synthesis.
FouriScale: A Frequency Perspective on Training-Free High-Resolution Image Synthesis
Innovative training-free approach FouriScale enhances high-resolution image generation by addressing structural and scale consistency through frequency domain analysis.
Pyramid Diffusion Model for Ultra-High-Resolution Image Synthesis
Pyramid Diffusion Model enables ultra-high-resolution image synthesis through innovative architecture and latent representation.
Analyzing and Improving Training Dynamics of Diffusion Models by NVIDIA Researchers
Improving training dynamics in diffusion models leads to better image synthesis results.
Ultra-High-Resolution Image Synthesis with Pyramid Diffusion Model
Pyramid Diffusion Model enables ultra-high-resolution image synthesis through a novel architecture and latent representation.
Pose-Guided Image Synthesis with Progressive Conditional Diffusion Models at ICLR 2024
PCDMs incrementally bridge the gap between source and target poses through three stages, generating high-quality synthesized images.
Fast High-Resolution Image Synthesis with Latent Adversarial Diffusion Distillation: A Novel Approach for Efficient Training and High-Quality Image Generation
Efficiently distilling large latent diffusion models for high-resolution image synthesis.
SynArtifact: Classifying and Alleviating Artifacts in Synthetic Images via Vision-Language Model
合成画像のアーティファクトを分類し、軽減するためにVision-Languageモデルを使用する方法を提案します。
