Image Generation

ECMamba: A novel dual-branch framework efficiently corrects multiple exposure images by integrating Retinex theory and an innovative 2D selective state-space layer, achieving state-of-the-art performa…

DreamSteerer enhances source image-conditioned editability in personalized diffusion models via a novel Editability Driven Score Distillation objective and mode shifting regularization, achieving sign…

DRACO, a denoising-reconstruction autoencoder, revolutionizes cryo-EM by leveraging a large-scale dataset and hybrid training for superior image denoising and downstream task performance.

Doubly hierarchical geometric representations enable realistic human hairstyle generation by separating low and high-frequency details in hair strands, resulting in high-quality, detailed virtual hair…

DomainGallery: Few-shot domain-driven image generation via attribute-centric finetuning, solving key issues of previous works by introducing attribute erasure, disentanglement, regularization, and enh…

DiTFastAttn: A post-training compression method drastically speeds up diffusion transformer models by cleverly reducing redundancy in attention calculations, leading to up to a 1.8x speedup at high re…

This paper introduces a novel implicit Zero-Watermarking scheme using disentangled style domains to detect unauthorized dataset usage in text-to-image models, offering robust copyright protection via …

Direct Unlearning Optimization (DUO) robustly removes unsafe content from text-to-image models by using paired image data and output-preserving regularization, effectively defending against adversaria…

Boosting personalized image generation! Direct Consistency Optimization (DCO) fine-tunes text-to-image models, ensuring subject consistency and prompt fidelity, even when merging separately customized…

DiP-GO: A novel pruning method accelerates diffusion models via few-step gradient optimization, achieving a 4.4x speedup on Stable Diffusion 1.5 without accuracy loss.

DiMSUM: A novel diffusion model boosts image generation by unifying spatial and frequency information, achieving superior results and faster training.

Diffusion4D: Fast, consistent 4D content generation via a novel 4D-aware video diffusion model, surpassing existing methods in efficiency and 4D geometry consistency.

Adaptive likelihood estimation and MAP inference during reverse diffusion tackles real-world image noise.

AdvUnlearn enhances diffusion model robustness against adversarial attacks during concept erasure by integrating adversarial training, improving the trade-off between robustness and model utility.

AI-generated images contaminate online continual learning datasets, hindering performance. A new method, ESRM, leverages entropy and real/synthetic similarity maximization to select high-quality data…

Deep Degradation Response (DDR) uses image deep feature changes under degradation to create a flexible image descriptor, excelling in blind image quality assessment and unsupervised image restoration.

Unlearning synthesized images efficiently reveals influential training data for text-to-image models, improving data attribution accuracy and facilitating better model understanding.

Ctrl-X: Zero-shot text-to-image generation with training-free structure & appearance control!

CryoGEM: Physics-informed generative model creates realistic synthetic cryo-EM datasets, boosting particle picking and pose estimation accuracy for higher-resolution protein structure determination.

Self-supervised blind image deblurring (BID) breakthrough! A novel cross-scale consistency loss and progressive training scheme using implicit neural representations achieves superior performance wit…