analiza aiqbench
This research paper investigates visual co-denoising within pixel-space diffusion models, isolating specific architectural and loss-function components that enhance semantic supervision without relying on pretrained autoencoders. The study demonstrates that a dual-stream architecture combined with perceptual-drifting hybrid losses yields superior generation quality and training efficiency compared to standard baselines. These findings provide a clear technical roadmap for improving high-level visual structure capture in next-generation generative models.
V-Co: A Closer Look at Visual Representation Alignment via Co-Denoising
arXiv:2603.16792v2 Announce Type: replace-cross Abstract: Pixel-space diffusion has recently re-emerged as a strong alternative to latent diffusion, enabling high-quality generation without pretrained autoencoders. However, standard pixel-space diffusion models receive relatively weak semantic supervision and are not explicitly designed to capture high-level visual structure. Recent representation-alignment methods (e.g., REPA) suggest that pretrained visual features can substantially improve diffusion training, and visual co-denoising has emerged as a promising direction for incorporating such features into the generative process. However, existing co-denoising approaches often entangle multiple design choices, making it unclear which are truly essential. We therefore present V-Co, a systematic study of visual co-denoising in a unified JiT-based framework. This controlled setting allows us to isolate the ingredients that make visual co-denoising effective. Our study reveals two main ingredients. First, co-denoising benefits from preserving feature-specific computation while enabling flexible cross-stream interaction, which leads to a fully dual-stream architecture together with a structurally defined unconditional prediction for classifier-free guidance. Second, it requires both stronger semantic supervision and proper cross-stream calibration, which we realize through a perceptual-drifting hybrid loss and RMS-based feature rescaling. Together, these findings yield a simple recipe for visual co-denoising. Experiments on ImageNet-256 show that, at comparable model sizes, V-Co outperforms the underlying pixel-space diffusion baseline and strong prior pixel-diffusion methods while using fewer training epochs, offering practical guidance for future representation-aligned generative models.
Abstract:Pixel-space diffusion has recently re-emerged as a strong alternative to latent diffusion, enabling high-quality generation without pretrained autoencoders. However, standard pixel-space diffusion models receive relatively weak semantic supervision and are not explicitly designed to capture high-level visual structure. Recent representation-alignment methods (e.g., REPA) suggest that pretrained visual features can substantially improve diffusion training, and visual co-denoising has emerged as a promising direction for incorporating such features into the generative process. However, existing co-denoising approaches often entangle multiple design choices, making it unclear which are truly essential. We therefore present V-Co, a systematic study of visual co-denoising in a unified JiT-based framework. This controlled setting allows us to isolate the ingredients that make visual co-denoising effective. Our study reveals two main ingredients. First, co-denoising benefits from preserving feature-specific computation while enabling flexible cross-stream interaction, which leads to a fully dual-stream architecture together with a structurally defined unconditional prediction for classifier-free guidance. Second, it requires both stronger semantic supervision and proper cross-stream calibration, which we realize through a perceptual-drifting hybrid loss and RMS-based feature rescaling. Together, these findings yield a simple recipe for visual co-denoising. Experiments on ImageNet-256 show that, at comparable model sizes, V-Co outperforms the underlying pixel-space diffusion baseline and strong prior pixel-diffusion methods while using fewer training epochs, offering practical guidance for future representation-aligned generative models.
Accepted by ECCV 2026. code: this https URL
Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
arXiv:2603.16792 [cs.CV]
(or arXiv:2603.16792v2 [cs.CV] for this version)
https://doi.org/10.48550/arXiv.2603.16792 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 323 cuvinte
Subiecte conexe
RynnWorld-4D: 4D Embodied World Models for Robotic Manipulation
arXiv · 30 sept. 2026
Beyond Trial Averaging: Anchoring Neural and Visual Representations for Few-Repetition Brain-to-Image Retrieval
arXiv · 20 aug. 2026
Problems of Proliferation
LessWrong · 29 sept. 2026
Inception in DiffusionGemma - Jailbreaking a Diffusion Language Model by Pinning Tokens Anywhere on the Canvas
LessWrong · 23 iul. 2026