Back to datasets
Dataset assetOpen Source CommunityVideo ProcessingVideo Super-Resolution

RealVSR

The RealVSR dataset is a benchmark dataset for real-world video super-resolution, containing low-quality and high-resolution video sequences in various formats, used for evaluating and training video super-resolution algorithms.

Source
github
Created
Aug 4, 2021
Updated
Nov 25, 2021
Signals
140 views
Availability
Linked source ready
Overview

Dataset description and usage context

Dataset Overview

Dataset Name

RealVSR

Dataset Description

RealVSR is a benchmark dataset for video super-resolution research. The dataset contains high-quality (HR) and low-quality (LR) video sequences, suitable for training and testing video super-resolution models.

Dataset Structure

The dataset includes the following files:

FileDescription
GT.zipAll high-quality RGB format sequences
LQ.zipAll low-quality RGB format sequences
GT_YCbCr.zipAll high-quality YCbCr format sequences
LQ_YCbCr.zipAll low-quality YCbCr format sequences
GT_test.zipHigh-quality RGB format test sequences
LQ_test.zipLow-quality RGB format test sequences
GT_YCbCr_test.zipHigh-quality YCbCr format test sequences
LQ_YCbCr_test.zipLow-quality YCbCr format test sequences
videos.zipOriginal videos, containing over 500 LR‑HR video pairs

Dataset Storage Location

The dataset is stored on Google Drive and Baidu Drive.

Dataset Samples

Sample images of the dataset illustrate several scenes and can be used to preview the dataset contents.

Dataset License

The dataset follows the Apache 2.0 license.

Need downstream help?

Pair the dataset with AI analysis and content workflows.

Once the source passes your review, move straight into summarization, transformation, report drafting, or presentation generation with the JuheAI toolchain.

Explore AI studio