A Chinese Traditional Opera Video Super-Resolution Dataset Based on the "Real-world+" Degradation Fusion

This dataset, named CTOVSR (A Chinese Traditional Opera Video Super-Resolution Dataset), is a large-scale, high-quality video collection created specifically for the research of video super-resolution (SR) on Chinese traditional opera and similar cultural heritage content. The dataset aims to provide computer vision researchers with a benchmark for training and evaluating advanced video restoration and enhancement algorithms, particularly for aged videos with complex, real-world degradations.The dataset was constructed through a rigorous multi-stage pipeline that integrates real-world degradation simulation with classic synthetic degradations. The “Real-world+” subset, derived from four opera videos, was processed through a three-stage alignment procedure—Temporal Alignment, Spatial Alignment, and Algorithm Filtering—to correct severe spatio-temporal misalignments between HR and LR sources. Additionally, to enhance dataset diversity and generalization, we generated synthetic degradation data from 41 other distinct HR opera videos using a second-order spatial degradation model and H.264-based temporal degradation.The detailed process can be found in our paper "A Chinese Traditional Opera Video Super-Resolution Dataset Based on the "Real-world+" Degradation Fusion".The dataset comprises a total of 900 strictly aligned LR-HR video sequence pairs. Each sequence contains 100 consecutive frames, totaling approximately 180,000 frames. All video frames are stored in the lossless 8-bit PNG format with the sRGB color space. The resolution of HR frames is 1920x1080, and the LR frames is 480x270, corresponding to a standard 4x super-resolution task.The dataset is partitioned into train (800 sequences) and test (100 sequences) folders. The root directory also contains a metadata.csv file, which documents detailed metadata for each sequence, including its source film, degradation type, and train/test assignment. It is important to note that any source video URLs provided in the metadata are for reference purposes only; due to the ephemeral nature of web content, their long-term availability cannot be guaranteed. The dataset is split into multi-part 4GB ZIP archives. To extract, please download all parts into the same directory and unzip the first .zip file.All processing code, alignment tools, and usage instructions for this dataset are open-sourced on GitHub: https://github.com/xiwang114/CTOVSR.git.