Scope-VSR: A Broad-Coverage Training Dataset for Video Super-Resolution

Gengyu Liao*, Yechun Yong*, Zheng Chen*, Jianze Li, Ziqing Zhang, Linghe Kong, Yulun Zhang†
1Shanghai Jiao Tong University, China
*Equal Contribution, †Corresponding Author
Under Review
Paper arXiv Code
The semantic atlas of SCOPE-VSR

The semantic atlas of SCOPE-VSR. Colors denote topics, with representative frames illustrating the diverse semantic content of our dataset.

Abstract

Diffusion models have shown superior performance in video super-resolution (VSR). However, restoring diverse real-world content remains challenging due to the lack of balanced semantic coverage in existing training data. In this paper, we introduce a structured curation pipeline for open-domain VSR, SCOPE. In this pipeline, we combine task-oriented video selection with hierarchical semantic rebalancing. Applying this pipeline to a large-scale collection of data sources, we construct the SCOPE-VSR dataset for training and SCOPE-test for testing. SCOPE-VSR has approximately 27k videos with a balanced and diverse coverage of various real-world scenarios and subjects. In addition, we propose GraCo-VSR, a novel framework for one-step video restoration. GraCo-VSR adopts a reliability-aware design with explicit supervision to reduce degradation-induced distortions in motion and texture guidance. This design enables the model to better leverage the diverse training content in SCOPE-VSR, thereby supporting generalization across varied scenarios. Experiments demonstrate that our model exhibits strong performance on both perceptual fidelity and visual quality.

Method

SCOPE: Dataset Construction

Overview of the SCOPE data curation pipeline
Overview of the SCOPE data construction pipeline

We propose SCOPE, a systematic pipeline that combines multi-stage curation with semantic rebalancing. We collect approximately 0.6 million video clips from multiple domains, including OpenVid-HD, UltraVideo, MiraData, HumanVid, and HOIGen. The original collection contains. By applying the pipeline, we extract 27,351 clips to form the final SCOPE-VSR dataset.

Representative Videos from SCOPE-VSR

Everyday People
Food Cooking
Performance Music
Sports Exercise
Transportation
Agriculture Rural
Animals
Everyday People
Open Natural Landscape
Presentation Communication
Presentation Communication
Product Showcase
Street Road
Transportation
Urban Building
Vegetation
Water
Work Making
Animation
Digital Screen Content

GraCo-VSR: Architecture

Overview of the GraCo-VSR framework
Illustration of the GraCo-VSR framework

We introduce GraCo-VSR, a one-step video restoration framework which provides reliable and granularity-aligned guidance for open-domain VSR. The Reliability-Aware Texture Encoder (RATE) is designed to extract confidence-weighted local texture cues, while the Clip-Level Dynamics Conditioning (CLDC) is designed to provide robust clip-level motion guidance.

Results

Quantitative Results (click to expand)
  • Quantitative results on synthetic and real-world datasets (Tab. 1 of the main paper)
Quantitative comparison with state-of-the-art video super-resolution methods
Qualitative Results (click to expand)
  • Visual comparisons on the real-world RealVSR dataset (Fig. 6 of the main paper)
Qualitative comparison on RealVSR
  • Visual comparisons on the synthetic YouHQ40 and SCOPE-test datasets (Fig. 7 of the main paper)
Qualitative comparison on YouHQ40 and SCOPE-test
More Results (click to expand)
  • Additional visual comparisons on UDM10 and SCOPE-test
Additional qualitative results on UDM10 and SCOPE-test
  • Additional visual comparisons on RealVSR
Additional qualitative results on RealVSR

BibTeX

@article{liao2027scopevsr,
  title={Scope-VSR: A Broad-Coverage Training Dataset for Video Super-Resolution},
  author={Liao, Gengyu and Yong, Yechun and Chen, Zheng and Li, Jianze and Zhang, Ziqing and Kong, Linghe and Zhang, Yulun},
  journal={arXiv preprint},
  year={2027}
}