Image Generation

Boosting diffusion-based human image animation, Test-time Procrustes Calibration (TPC) ensures high-quality outputs by aligning reference and target images, overcoming common compositional misalignmen…

TinyLUT achieves 10x lower memory consumption and superior accuracy in image restoration on edge devices using innovative separable mapping and dynamic discretization of LUTs.

Terra, a novel time-varying low-rank adapter, enables effective cross-domain fine-tuning of diffusion models by creating a continuous parameter manifold, facilitating efficient knowledge sharing and g…

R3GAN, a minimalist GAN baseline, surpasses state-of-the-art models by using a novel regularized relativistic GAN loss and modern architectures, proving GANs can be trained efficiently without relying…

TextCtrl: a novel diffusion-based scene text editing method using prior guidance control, achieving superior style fidelity and accuracy with a new real-world benchmark dataset, ScenePair.

BIR-D tames generative diffusion models for universal blind image restoration, dynamically updating parameters to handle various complex degradations without assuming degradation model types.

DoSSR: A novel SR model boosts efficiency by 5-7x, achieving state-of-the-art performance with only 5 sampling steps by cleverly integrating a domain shift equation into pretrained diffusion models.

SyncTweedies: a zero-shot diffusion synchronization framework generates diverse visual content (images, panoramas, 3D textures) by synchronizing multiple diffusion processes without fine-tuning, demon…

Boosting diffusion model features: This paper introduces GATE, a novel method to suppress ‘content shift’ in diffusion features, improving their quality via off-the-shelf generation techniques.

Stylus: an automatic adapter selection system for diffusion models, boosts image quality and diversity by intelligently composing task-specific adapters based on prompt keywords.

StoryDiffusion enhances long-range image & video generation by introducing a simple yet effective self-attention mechanism and a semantic motion predictor, achieving high content consistency without t…

StepbaQ enhances quantized diffusion models by correcting accumulated quantization errors via a novel sampling step correction mechanism, significantly improving model accuracy without modifying exist…

Stable-Pose: Precise human pose guidance for text-to-image synthesis.

DiGIT stabilizes image autoregressive models’ latent space using a novel discrete tokenizer from self-supervised learning, achieving state-of-the-art image generation.

D³GM, a novel score-based diffusion model, enhances stability & generalizability in solving inverse problems by leveraging measure-preserving dynamics, enabling robust image reconstruction across dive…

SSDiff: A novel spatial-spectral integrated diffusion model for superior remote sensing pansharpening.

SpikeReveal: Self-supervised learning unlocks sharp video sequences from blurry, real-world spike camera data, overcoming limitations of prior supervised approaches.

STIR: A novel spatio-temporal network reconstructs high-quality images from spiking camera data by jointly refining motion and intensity information for efficient and accurate high-speed imaging.

Smoothed Energy Guidance (SEG) improves unconditional image generation by reducing self-attention’s energy curvature, leading to higher-quality outputs with fewer artifacts.

Slightly corrupting pre-training data significantly improves diffusion models’ image generation quality, diversity, and fidelity.