평범한 필기장

Paper : https://arxiv.org/abs/2412.09611 FluxSpace: Disentangled Semantic Editing in Rectified Flow TransformersRectified flow models have emerged as a dominant approach in image generation, showcasing impressive capabilities in high-quality image synthesis. However, despite their effectiveness in visual generation, rectified flow models often struggle with disentanarxiv.orgAbstract기존 문제 Rectifi..

Paper : https://arxiv.org/abs/2412.15205 FlowAR: Scale-wise Autoregressive Image Generation Meets Flow MatchingAutoregressive (AR) modeling has achieved remarkable success in natural language processing by enabling models to generate text with coherence and contextual understanding through next token prediction. Recently, in image generation, VAR proposes scale-wisarxiv.orgAbstract기존 VAR은 다음 두가지..

Paper : https://arxiv.org/abs/2502.09616 Variational Rectified Flow MatchingWe study Variational Rectified Flow Matching, a framework that enhances classic rectified flow matching by modeling multi-modal velocity vector-fields. At inference time, classic rectified flow matching 'moves' samples from a source distribution to the tararxiv.orgAbstract 본 논문은 multi-modal velocity vector-fields를 모델링함으로..

https://yuxiangbao.github.io/FreeInv/ FreeInv: Free Lunch for Improving DDIM InversionNaive DDIM inversion process usually suffers from a trajectory deviation issue, i.e., the latent trajectory during reconstruction deviates from the one during inversion. To alleviate this issue, previous methods either learn to mitigate the deviation or deyuxiangbao.github.ioAbstract Naive DDIM은 reconstruction과..

paper : https://arxiv.org/abs/2410.12557 One Step Diffusion via Shortcut ModelsDiffusion models and flow-matching models have enabled generating diverse and realistic images by learning to transfer noise to data. However, sampling from these models involves iterative denoising over many neural network passes, making generation slow aarxiv.orgAbstract본 논문은 shortcut model을 제안한다. 이는 single network를..

Paper : https://arxiv.org/abs/2504.13109 UniEdit-Flow: Unleashing Inversion and Editing in the Era of Flow ModelsFlow matching models have emerged as a strong alternative to diffusion models, but existing inversion and editing methods designed for diffusion are often ineffective or inapplicable to them. The straight-line, non-crossing trajectories of flow models posearxiv.orgAbstractDiffusion mo..

Paper : https://arxiv.org/abs/2411.15843 Unveil Inversion and Invariance in Flow Transformer for Versatile Image EditingLeveraging the large generative prior of the flow transformer for tuning-free image editing requires authentic inversion to project the image into the model's domain and a flexible invariance control mechanism to preserve non-target contents. However, thearxiv.org1. Introductio..

Paper : https://arxiv.org/abs/2411.00113 A Geometric Framework for Understanding Memorization in Generative ModelsAs deep generative models have progressed, recent work has shown them to be capable of memorizing and reproducing training datapoints when deployed. These findings call into question the usability of generative models, especially in light of the legal andarxiv.orgAbstract본 논문은 memori..

Paper : https://arxiv.org/abs/2406.03537 A Geometric View of Data Complexity: Efficient Local Intrinsic Dimension Estimation with Diffusion ModelsHigh-dimensional data commonly lies on low-dimensional submanifolds, and estimating the local intrinsic dimension (LID) of a datum -- i.e. the dimension of the submanifold it belongs to -- is a longstanding problem. LID can be understood as the number ..

Paper : https://arxiv.org/abs/2412.07517 FireFlow: Fast Inversion of Rectified Flow for Image Semantic EditingThough Rectified Flows (ReFlows) with distillation offers a promising way for fast sampling, its fast inversion transforms images back to structured noise for recovery and following editing remains unsolved. This paper introduces FireFlow, a simple yet effarxiv.orgAbstract Rectified Flow..