SouthKen SV: A Real-World Stereo Video Dataset

Current stereo video datasets are limited in both diversity and realism. To address the lack of well-rectified data and offer a more reliable dataset for visualization and qualitative evaluation, we present the South Kensington SV dataset, a real-world stereo videos covering indoor and outdoor scenes recorded via a ZED 2 stereo camera. Videos were captured across locations in South Kensington, London, including the Imperial College campus, Hyde Park, and Chelsea. The dataset includes diverse urban scenarios with dynamic objects, encompassing various weather conditions (sunny, cloudy, rainy, night, etc) and camera movements, primarily from walking.

It consists of 266 stereo videos, each between 10 and 70 seconds in length, recorded at 1280x720 resolution and 30 fps. The raw videos were inspected to select sequences with consistent exposure without abrupt changes in recording parameters. The data collection process spanned over three months, amounting to approximately 300 man-hours. In addition to its value for stereo video tasks, this dataset is also suitable for lite stereo model distillation (e.g., Lite Any Stereo), monocular depth training with pseudo labels, and low-level vision tasks such as stereo video super-resolution, denoising, and compression.

The releasing of SouthKen SV dataset is under strict conditions. SouthKen SV cannot be used or distributed for any commercial purposes. It can only be used for academic and research purposes. SouthKen SV is licensed under Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0). Under this license, if you want to modify SouthKen SV or generate new data from SouthKen SV (e.g., super-resolution, denoising, defocus), the releasing of your new data should be licensed under the same CC BY-NC-SA 4.0. If you need more help or information, please contact: j.jing23@imperial.ac.uk.