Self-supervised ControlNet with Spatio-Temporal Mamba for Real-world Video Super-resolution

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Shi, Shijun, Xu, Jing, Lu, Lijing, Li, Zhihang, Hu, Kai
Formato: Preprint
Publicado: 2025
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866915317623554048
author Shi, Shijun
Xu, Jing
Lu, Lijing
Li, Zhihang
Hu, Kai
author_facet Shi, Shijun
Xu, Jing
Lu, Lijing
Li, Zhihang
Hu, Kai
contents Existing diffusion-based video super-resolution (VSR) methods are susceptible to introducing complex degradations and noticeable artifacts into high-resolution videos due to their inherent randomness. In this paper, we propose a noise-robust real-world VSR framework by incorporating self-supervised learning and Mamba into pre-trained latent diffusion models. To ensure content consistency across adjacent frames, we enhance the diffusion model with a global spatio-temporal attention mechanism using the Video State-Space block with a 3D Selective Scan module, which reinforces coherence at an affordable computational cost. To further reduce artifacts in generated details, we introduce a self-supervised ControlNet that leverages HR features as guidance and employs contrastive learning to extract degradation-insensitive features from LR videos. Finally, a three-stage training strategy based on a mixture of HR-LR videos is proposed to stabilize VSR training. The proposed Self-supervised ControlNet with Spatio-Temporal Continuous Mamba based VSR algorithm achieves superior perceptual quality than state-of-the-arts on real-world VSR benchmark datasets, validating the effectiveness of the proposed model design and training strategies.
format Preprint
id arxiv_https___arxiv_org_abs_2506_01037
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Self-supervised ControlNet with Spatio-Temporal Mamba for Real-world Video Super-resolution
Shi, Shijun
Xu, Jing
Lu, Lijing
Li, Zhihang
Hu, Kai
Computer Vision and Pattern Recognition
I.4.4; I.2.6
Existing diffusion-based video super-resolution (VSR) methods are susceptible to introducing complex degradations and noticeable artifacts into high-resolution videos due to their inherent randomness. In this paper, we propose a noise-robust real-world VSR framework by incorporating self-supervised learning and Mamba into pre-trained latent diffusion models. To ensure content consistency across adjacent frames, we enhance the diffusion model with a global spatio-temporal attention mechanism using the Video State-Space block with a 3D Selective Scan module, which reinforces coherence at an affordable computational cost. To further reduce artifacts in generated details, we introduce a self-supervised ControlNet that leverages HR features as guidance and employs contrastive learning to extract degradation-insensitive features from LR videos. Finally, a three-stage training strategy based on a mixture of HR-LR videos is proposed to stabilize VSR training. The proposed Self-supervised ControlNet with Spatio-Temporal Continuous Mamba based VSR algorithm achieves superior perceptual quality than state-of-the-arts on real-world VSR benchmark datasets, validating the effectiveness of the proposed model design and training strategies.
title Self-supervised ControlNet with Spatio-Temporal Mamba for Real-world Video Super-resolution
topic Computer Vision and Pattern Recognition
I.4.4; I.2.6
url https://arxiv.org/abs/2506.01037